Demand generation never existed.

Not the way we sold it. Not the way we put it on org charts and comp plans and the second slide of every board deck between 2010 and 2020. What we called demand generation was demand capture wearing a better jacket. Venture capital paid for the jacket.

Brand marketing is demand generation. It always was. It is the only discipline that ever created a want where there wasn’t one. Everything the demand gen team did was walk into a market and find the want that was already there.

I ran demand gen through the whole era. I believed the story. I sold it upstairs. It was all bullshit - when my team snagged a deal it was high velocity, fully BANT qualified. When you generate a deal, it’s not BANT qualified. Full stop. You might nail getting the buyer on the phone (authority), you might luck into the need, but if you’re finding someone with budget and a proscribed timeline, the deal existed before we bumped into it.

We’ve all learned this lesson at some point. So, here’s the legend and what actually happened.

The Robin Hood years

Gong was founded in 2015. It launched in 2016 into a category called Conversation Intelligence. Record the calls, coach the reps. Good product. No one in the room. Just Gong. They did the yeoman’s work of creating the category.

The room filled in about three years. Everybody recorded calls. Senior revenue leaders kicked the meeting down to enablement because enablement is where call recording lives. Gong had a bigger vision and a category that couldn’t hold it.

So in October 2019 they renamed the category. Revenue Intelligence.

In October 2019 there were zero Google searches for the term. Zero. Eighteen months later every competitor was using it. Forrester published a Wave. Gartner published a market report. Gong raised $584 million all in, grew ARR 2.3x in a year, crossed 2,000 customers, and took a $7.25 billion valuation in 2021.

That is the best category story of the decade. Read it again and count the demand generation.

There isn’t any. Nobody was searching. Nobody had a budget line. Nobody woke up wanting Revenue Intelligence. What Gong did was pick a pain (revenue leaders can’t see what’s actually happening in their deals), draw a boundary around the pain, put a name on the boundary, and send a sales team through every account that matched the ICP until they bumped into it.

Inefficient. Didn’t matter. The burn was underwritten.

Now, we have to make a nod to the brilliant brand marketing in sheep’s clothing, the content marketing play that had Gong sharing unique sales insights.

That’s category definition. We called it demand gen because the leads landed in the CRM and somebody needed credit for them.

Why the 5% didn’t exist

Professor John Dawes at the Ehrenberg-Bass Institute gave us the 95:5 rule. At any moment, 5% of a B2B market is in-market. The other 95% is not.

In a new category the 5% is zero. You cannot capture demand for a thing that has no name. There is nothing to search, nothing to compare, nothing to put in an RFP.

So the category creators hunted the 20%. The 5% who would buy if the thing existed, plus the 15% who have the pain right now and aren’t shopping. (The 15% is my number from twenty years of doing this, not a published stat. Fight me.) They hunted it by hand. Outbound only. Unit economics be damned.

And then the category crossed into the late majority.

Rogers gave us the curve: 2.5% innovators, 13.5% early adopters, 34% early majority, 34% late majority, 16% laggards. Laggards are hard to reach. Laggards often don’t have the pain. Laggards cost a fortune to educate. Once you’re selling to them, the cost of the education breaks the unit economics that were never good to begin with.

The math died. The legend didn’t. Demand gen budgets survived on the legend.

So now everyone in the category is fighting over the same 5%, and we told ourselves we were creating demand. We were chasing the same five accounts in a hundred as every competitor with the same term in their tagline.

The subsidy left. Then it came back and walked past you.

Venture funding peaked in 2021 and fell off a cliff through 2023. Everyone knows this part.

Here’s the part that matters. 2025 was the third-highest venture funding year on record, behind only 2021 and 2022. Roughly half of it went to AI.

The money came back. It did not come back to your stack. It walked past RevTech and MarTech and went to the AI natives, who are drawing new boundaries across the old categories. That is the last frontier of category definition, and it’s real. If you’re standing on it, there’s a section for you below.

If you’re not, keep reading.

Brinker’s graph stopped growing

Scott Brinker’s martech landscape grew 100x from 2011 to 2025. In 2026 it grew 0.79%. Fifteen thousand five hundred and five logos. Flat.

Under the flat number: 1,488 products added, 1,367 removed. Churn, not growth. Brinker calls it peak martech, with a question mark. He expects a split: big platforms and data infrastructure on one side, a long tail of small contextual tools and agents on the other. Nothing in between that looks like a category.

Translation. Wherever the RevTech or MarTech stack is stable, the boundaries are drawn. There is nothing left to define. And demand gen, which was only ever category definition with a checkbook, has nothing left to do.

Stop funding it. There is no demand to generate. There is demand to capture, and there is a 15% nobody is talking to.

Where are you?

Three positions. The Cannonball methodology and 5/15/80 apply to all three. What changes is which number you’re hunting and what you’re allowed to ignore.

You’re creating a category. There is no 5%. Hunt the 15%. Outbound only. Pre-scale. Follow the pain with EDPs, pain-based segments, and PVPs, the same way you would in a mature market, except you’re not allowed to care about CAC yet. You’re not generating demand. You’re defining a boundary, and the boundary is the pain. The AI natives get this. The ones that don’t will run a demand gen playbook into a market with zero searches and blame the SDRs.

You’re in the early majority. The 5% exists now, and it rewards efficiency. This is a citation game. Buyers ask the model, the model cites what it trusts, and what it trusts is earned media, reviews, and third-party coverage. Build the best product, get the reviews, earn the citations, own the answer. This is where flywheels and bow ties finally make sense, because the flywheel is the citation engine. It’s also where the brand marketer gets revenge. I wrote about that here: The Brand Marketer’s Revenge. Brand is the demand generation. It always was. Meanwhile your outbound goes to the 15%, because the 5% is being handled by your reputation.

You’re in the late majority. Everyone is fighting for the 5%. Saturate it. You should. But the 15% is the only open ground left, and it’s the ground demand gen budgets ignore because it doesn’t look like a lead.

Here’s your tell. Matt Dixon and Ted McKenna analyzed 2.5 million sales conversations for The JOLT Effect. Somewhere between 40% and 60% of qualified deals are lost to no decision. Not to a competitor. To nothing. Of those, 44% never accepted the pain existed. The other 56% accepted it and froze.

That 44% is the laggard education bill showing up in your pipeline. Every no-decision loss is a demand gen dollar that paid to find someone who wasn’t in the 20%. A rising no-decision rate is a category telling you it’s late.

How to tell which one you are

Six signals in public data. Three in your own pipeline. None of them need a consultant.

Public:

• Search volume for the category term. Zero means pre-category. Rising means early majority forming. Flat or falling means late majority. The shape matters more than the number.

• Analyst coverage. The first Forrester Wave or Gartner market report is the crossing. Gong got both within eighteen months of the rename.

• Competitor adoption of the term. When every competitor uses your category name, you’re mainstream.

• Brinker churn inside your category. Adds versus removals. Removals winning means consolidation, which means late.

• Funding stage mix. Seed and Series A dominance is early. Late-stage rounds, M&A, and valuation compression are late. Gong’s $7.25 billion is reported closer to $4.5 billion now.

• The objection you hear most. “Does it work” is early. “How does it integrate and what’s the ROI” is early majority. “Why you and not them, and why this price” is late.

Pipeline:

• Share of new opportunities from in-segment references. Under 40% and you haven’t crossed. This one’s a consultant’s benchmark, not a law, but it’s directionally right.

• CAC payback. Early categories run 24 months and up. Mainstream runs 12 to 18. If yours is climbing, the category is aging under you.

• No-decision rate. Track it by segment. Watch the 44% who never bought the pain. That’s your laggard exposure, measured.

The unicorn

Brand marketing is demand generation. Outbound is demand capture. Say that at dinner and watch a CMO’s face.

Demand gen is a unicorn ridden by a fairy in a playland of self-delusion, built on the legend of a category boom that ended when the checks did.

The checks came back. They’re funding the last frontier. Everywhere else, the boundaries are drawn, the 5% is spoken for, and the 15% is sitting there with the pain and no one in their inbox.

Stop generating. Start capturing.

New here? This post is one argument inside a larger methodology. Start with the post it picks a fight with, then the playbook that shows the 15% being found in the wild.

• The Brand Marketer’s Revenge. Why brand is the demand generation, and why the citation game belongs to it.

• Finding Hidden Customers: 0.05% to 1.2%. A construction vertical SaaS company, public permit data, and a meeting booked rate that went up 24x.

• The Category Position Scorecard. One page. Nine signals. Tells you which number to hunt.

The signals market finally agreed on its words. First-party, third-party, intent, surge. And then, the ink barely dry, it started fighting about the number: is 5% of your market active at any moment, or 40%?

When a market stabilizes its vocabulary before it stabilizes its facts, buyers attach to the wrong things. The most expensive wrong thing right now is the assumption that a signal, on its own, is useful. A lot of money has been spent finding out it isn’t.

This post is here to save you that line item.

The words everyone agreed on

Quick tour, so we’re standing on the same floor.

A prospect visits your pricing page three times in a week. That’s a first-party signal: behavior on property you own, captured by systems you run. A buyer sits on G2 comparing you to the two vendors you lose to most. That’s second-party: someone else’s property, but the behavior is explicitly about purchase. An account’s research activity surges on topics in your category, aggregated across a publisher network and sold to you by Bombora or ZoomInfo. Third-party.

That’s the taxonomy. Every vendor deck you’ll see this year is built on it.

We covered where signals lead in Signals: The Gateway Drug to Public Data. That post was the primer. This one is about what the taxonomy can’t see, and why the accounts you most need to reach live in its blind spot.

Signals earn their keep. Let’s say that plainly.

Signals see the 5%.

Professor John Dawes at the Ehrenberg-Bass Institute put the number on it: at any given moment, roughly 5% of your market is actively buying. (My working extension, from campaign data rather than a study: another 15% has the pain but isn’t shopping yet. Hold that thought.)

Seeing the 5% clearly is worth real money. Signals put you at the top of the in-market funnel. Better than that, they get you into accounts that already chose to start a conversation with another brand, and nothing else opens that door. If a buyer is comparing your competitor’s pricing page to a G2 grid right now, a signal is the only thing that tells you.

I’m not writing from a distance here. I’ve bought the platforms. I’ve bought the data raw and integrated it into my workflows. Guess what? My competitors had the same fucking data.

Now, about the fight over the number. 6sense published research last year arguing the 5% figure came from consumer contexts and that around 40% of B2B accounts show meaningful signs of buying activity at any moment. Interesting work, worth reading. Also worth noticing: the companies telling you the in-market number is eight times bigger than Dawes measured are the companies whose revenue depends on the in-market number being big. Draw your own line there. I’ve drawn mine.

And then the part nobody says out loud

Every signal you can buy, your competitor can buy.

Bombora sells the same surge to you and to the company you lose deals to. Same account, same topic, same week. Your first-party data is exclusively yours, which is genuinely valuable, but it has the opposite problem: it only captures buyers who already found you. The 5%, or the fraction of the 5% that walked through your door.

Both buckets, first-party and third-party, share one trait the taxonomy never names. They are sourced. Somebody captured the behavior. You collected it or you paid for it. And a sourced signal can only ever answer one question: who is moving right now.

There’s a second question. Who is suffering right now, whether they’re moving or not.

No vendor sells the answer, because the answer can’t be sourced. It has to be built.

Three records that mean nothing. Then everything.

A worked example. Commercial construction, a mid-size contractor, and three public records.

Record one: OSHA’s inspection database shows the contractor took two serious fall-protection citations in the last fourteen months. Public, free, sitting in IMIS for anyone who looks.

Record two: permit data (we pull it from Shovels) shows their permit volume up roughly 40% year over year. They’re scaling, fast.

Record three: USAspending shows they just won their first federal contract. Federal work means federal compliance standards, and a paper trail that follows you.

Each record alone is trivia. A citation happens. Growth happens. Contractors win contracts. Nobody at Bombora scores any of it.

And then you put them together. A contractor with a live citation history is scaling headcount onto more job sites while stepping into a compliance regime with no tolerance for their existing record. That’s not trivia. That’s a company whose next serious incident doesn’t cost a fine. It costs the federal relationship, the surety rates, possibly the license. If you sell safety compliance software, this account has an existential problem today, and there is a better than even chance nobody inside the building has assembled these three records into one sentence.

That sentence is an Existential Data Point. Not captured. Not bought. Built.

And because it’s built, it’s the one signal with no second customer. Bombora can’t sell it to your competitor, because Bombora doesn’t have it. Nobody has it until someone does the triangulation.

Here’s the full map, the one the taxonomy hides:

The market sells three quadrants and calls it the whole map.

Shopping and suffering

So the clean version:

Signals tell you who’s shopping. The EDP tells you who’s suffering.

Shopping is the 5%. Suffering is the 15%.

You don’t pick one. You use the second to define the universe and the first to time your entry into it.

The layering inversion

Which brings us to why this post lives in the channel-stacking playbook.

The market’s layering advice, and there’s a lot of it right now, stacks signals on signals: surge plus pricing-page visit plus new CRO hire equals a higher account score. Fine, as far as it goes. But look at what’s underneath the stack. A list somebody else built. Layering sourced signals re-ranks the 5%. It never expands it.

The inversion: the EDP builds the list. Pain segments divide it. And then, only then, first-party and third-party signals get layered on top, doing the one job they’re actually great at: telling you when, and who first.

That’s a demotion from strategy to sequencing. It’s also a promotion, because sequencing is where signals stop wasting your money and start earning it.

What it looks like on a Tuesday: the segment already exists, built from public data months before anyone surged. An account in the segment fires a third-party signal. It doesn’t trigger a “noticed you were researching solutions like ours” email. It moves that account to the front of a line it was already standing in, and the first touch is a PVP built on the EDP: the three-record sentence nobody inside their building has written yet.

The full sequencing method, channel by channel, is in the Channel Layering Playbook for paid subscribers.

The close

Buy signals to know who’s moving. Build the one that knows who’s hurting.

The market sells the first kind because the first kind can be sold.

New here? Start with Start Here: Your Complete Guide to Cannonball GTM. It maps the whole methodology in one read. For the primer on signals and public data, read Signals: The Gateway Drug to Public Data. And if you want to know what a good outbound number actually looks like, the Meeting Booked Rate Calculator is free and takes five minutes.

Want to start a fistfight? When there’s a lull in a sales call, bark out:

Never call into a non-responsive account!

Not “call less.” Not “deprioritize.” We’re talking never. Never-ever, even.

Brace for fisticuffs.

The tried-and-true (and tired) five-to-seven-email sequence with a follow-up call baked into every step no matter what the prospect does… that’s TAM spam (and it screws up your email deliverability). Sure, it’s a playbook that works if by ‘works’ you mean throwing enough darts eventually hits a bullseye. But, it’s not what you should do if you have a real number attached to your pipeline.

Instead, treat the phone call as a signal-triggered action. Not just a step in the cadence.

The call only happens because something told me to make it. And figuring out what “something” means is the whole game.

From One Baseline to Two

You already know your meeting booked rate. If you don’t, go run the stripped-down campaign we walked through in Punk Rock Campaigns. A pain-only segment with a mediocre message typically lands a meeting request rate of around 1%, a real point-of-view message can push it to 4%.

While you’re establishing that baseline, also establish a second one at the same time: your signal rate. This is the layer underneath the booked meeting. It answers a different question: When a meeting didn’t get booked, how many people still did something? Checked out the site, poked around, showed up in your first-party data in a way that’s invisible to them but visible to you.

With a mediocre message, that’s usually 20 to 30% of the list. With a message built around a real point of view, we’ve seen it hit 40 to 50%. The better the message, the higher the signal strength, which makes sense, because signal strength is really just a measure of how much the message actually landed. All it takes to generate the signal is a brand name in your email signature. No tracked links, no clever UTM games. Doesn’t matter.

Who Gets a Call (and in What Order)

Signal strength is what determines who gets a phone call. Not opens, not clicks on collateral, not “let’s circle back.” Here’s the order:

• Anybody who has shown interest in a meeting. They replied with the email with some version of, “Yeah, let’s set something up.” Don’t respond with an email. Someone should call immediately. Ten-minute window, sometimes five. This isn’t a discovery call and you’re not making friends. Book the meeting and get off the phone. Don’t qualify the account. Pretend to be your admin and say “this is Frank, I’m so-and-so’s admin, he asked me to call and get this scheduled.” You’re not trying to sell anything. Just schedule the meeting and close the loop.

• Anybody who crossed the signal threshold. They hit the site. They’re checking whether you’re for real. Maybe they’re mildly annoyed that you know something about them. That’s fine, that’s a good sign, as long as you don’t act as if you know it.

Read more

I drew the wrong shape.

When I wrote The Brand Marketer’s Revenge, I handed you a barbell. Revenue Growth Engineers on one end. Brand Marketers on the other. Thin middle. I called it the future of B2B SaaS marketing, and I meant it.

I still mean most of it. But I was missing something. And I didn’t figure out what until a conversation with Yasmine de Aranda, who gave me a term I hadn’t heard before: Zero Churn Architect.

More on that in a moment. First, let me tell you where the barbell breaks down.

What the Barbell Gets Wrong

The barbell was built entirely around acquisition.

Revenue Growth Engineers find prospects experiencing pain right now. Brand Marketers build the reputation that makes you show up when buyers ask an LLM for the best solution in your category. Both sides are real. Both sides matter. But both sides are still fundamentally about getting new customers.

What neither end of the barbell addresses is what happens after you get them.

Here’s the part I’ve been conveniently ignoring: most B2B SaaS companies are running their acquisition motion full speed into a leaky bucket. They’re spending like hell on the left side of the barbell while the right side drains. CAC keeps climbing. CROs cycle out every 18 months — not because they’re bad at their jobs, but because the targets they were handed weren’t mathematically possible to hit in the first place. Nobody asks that question in the boardroom. They just set the number and hire someone to chase it.

A barbell built on acquisition doesn’t fix that. It accelerates it.

Enter Yasmine de Aranda

I met Yasmine at a dinner table full of CXOs. She had just started working with the team at Winning by Design, helping them build a new practice they call Growth Guidance. She also recently called herself — and I’m crediting her here because this is entirely her term — a Zero Churn Architect.

When she said it, I immediately asked if I could use it. She said yes.

Here’s how she defines it, in her own words: it’s not about reducing churn. It’s about identifying who your ideal customer actually is — through product — and then bringing that knowledge back into sales and marketing. That’s the missing piece. Most companies start with sales and marketing and treat product as a separate universe. Zero churn architects start with the full customer lifecycle and work backwards.

The framework that underpins her work is the Winning by Design Bow Tie — if you’re newer to Cannonball, I wrote about it in our SPICED + Cannonball GTM post. The short version: the Bow Tie maps the entire customer lifecycle as two triangles facing each other. The left triangle is everything before a customer signs — awareness, consideration, decision. The right triangle is everything after — onboarding, retention, expansion. Most SaaS companies obsess over the left triangle and sleepwalk through the right.

Winning by Design built a methodology that treats the right side as seriously as the left. Growth Guidance — the practice Yasmine is helping build — goes a level above that. It asks: given where you are right now, given your stage, your product, and your market, what is the mathematically correct way to grow? Is acquisition your lever? Expansion? Retention? And how far are you from your aspirational target if you keep pulling the wrong one?

Yasmine described a CEO who walked into a meeting and said: here’s our North Star, $100 million ARR. Winning by Design ran the math. The gap between where they were trending and where they wanted to go was enormous. More importantly, the path they were on — acquisition-first, always acquisition — wasn’t going to close it. The lever they needed to pull was expansion. Their existing customers were sitting on untapped value. The company didn’t have a pipeline problem. It had a growth design problem.

That’s what a Zero Churn Architect sees that the rest of us miss.

The Diamond

So I’m scrapping the barbell. Here’s the new shape: a diamond.

At the top is the gap. Not a person. Not a role. A number. The delta between where your board wants to go and where you’re actually headed if nothing changes. This is the governing math of the whole system. Everything else in the diamond exists to close it.

On the left is the Revenue Growth Engineer. This is the role I’ve been teaching on Cannonball. They find prospects experiencing pain right now using public data most competitors don’t know exists. They build campaigns so valuable prospects would pay to receive them. They move fast. They use AI to do the work of three people. They’re not support staff for your demand gen team — they ARE your demand gen team.

On the right is the Brand Marketer. This is the role we killed a decade ago and are now desperately trying to resurrect. As LLMs reshape how buyers research, brand presence matters in ways we couldn’t measure before. Getting your executives published matters. Thought leadership that shapes the category conversation matters. Reputation matters. The LLMs are measuring it for you now.

At the bottom is Sales. The human conversion layer. Relationships, qualification, closing. Not going away. Just better supported than before.

In the center — and this is the key — is the Zero Churn Architect. They’re not the thin middle I waved off in the barbell. They’re the governing intelligence of the whole system. They see across product, customer success, sales, and marketing simultaneously. They’re the ones who look at the board’s aspirational target and say: okay, here’s the gap, and here’s which lever we actually need to pull right now to close it. Sometimes that’s acquisition. Sometimes it’s expansion. Sometimes it’s retention. Almost never is it “more of everything.”

The center doesn’t report to the left. It doesn’t report to the right. It holds the math accountable.

Why This Matters Right Now

Three things are converging that make this structure necessary today, not in three years.

First, AI-enabled acquisition is becoming table stakes. Revenue Growth Engineering is real and it works — I’m seeing 4-10X reply rates with clients every week. But markets catch up. The companies doing this in 2026 have an edge. By 2028, everyone will have figured out the basics. When that happens, differentiation shifts. Who you acquire matters less than what you do with them after.

Second, LLMs are reshuffling brand presence in ways that reward the long game. Here’s a data point that stuck with me from Yasmine: Winning by Design is now finding new clients through ChatGPT and Claude referrals, not search. A firm whose entire methodology is about the full customer lifecycle — retention, expansion, compounding growth — is being discovered because AI surfaces holistic thinking. The Brand Marketer’s Revenge is already playing out in real time.

Third, the era of growth at all costs is over and the math is brutal. In a zero interest rate environment, you could burn cash on acquisition indefinitely and call it a strategy. That world is gone. Every quarter you run acquisition-first without a governing model for the whole lifecycle is a quarter your CAC climbs and your expansion revenue sits unrealized.

The companies that win the next five years aren’t just great at finding prospects. They’re great at knowing which lever to pull, when, and how hard. That’s the Zero Churn Architect’s job.

The Honest Admission

I’ve been teaching one corner of the diamond.

Revenue Growth Engineering is real. It works. I’m going to keep teaching it, because there are thousands of growth leaders who still need to modernize their acquisition motion before they can think about anything else.

But I’ve been writing about the org of the future while ignoring two-thirds of it. I called it a barbell because that’s what I could see from where I was standing.

Yasmine de Aranda handed me a better shape. Credit where it’s due.

The diamond is the org of the future. A Revenue Growth Engineer who can find and convert. A Brand Marketer who builds presence before the buyer knows they’re looking. A Sales team that closes. And in the center, someone whose entire job is to see the gap, understand the math, and decide what the organization actually needs to do next.

Figure out which point you’re building. Then figure out who’s sitting in the center.

Yasmine de Aranda is a Zero Churn Architect and growth guidance practitioner working with the team at Winning by Design.

Fifty-seven new customers a year. Or seventy-five. Or a hundred and seventeen.

Same market. Same 16,871 companies. Same product, same team, same budget. Nothing about the company changes across those three numbers. The only thing that moves is one variable, and it isn’t on your dashboard. It has never been on your dashboard. Most growth leaders have never measured it, because most growth leaders don’t know it exists.

This post builds the model that exposes it. Every input is either published research or a number you can count yourself, so when the ladder climbs from 57 to 117, you’ll know exactly which rung did the work.

The map underneath the model

Let’s start with the research everyone quotes and nobody finishes. The 95:5 rule comes from Professor John Dawes at the Ehrenberg-Bass Institute, published in 2021. The math underneath it is simple: businesses buy a given category about once every five years, so roughly 20 percent of a market is in-market in a given year, about 5 percent in a given quarter. Dawes calls it a heuristic. A way of seeing, not a law.

Here’s the part that gets left out. That 95 percent isn’t one territory. Inside it sits a group of companies with the pain your product solves, today, who haven’t started shopping. In my work overhauling go-to-market for vertical SaaS accounts, that group runs about 15 percent of a market at rest. My number, from campaigns, not a journal.

Three territories, then. The 5 percent shopping. The 15 percent hurting but not shopping. The 80 percent fine, for now.

The instrument

The market is Texada’s: software for companies that sell, rent, and service heavy equipment. IBISWorld counts 16,871 heavy equipment rental businesses in the US. Per Dawes, about 845 of them enter market in any quarter, roughly 3,374 buyers a year.

Two rates drive everything, and neither is invented.

The first is capture. In 2022, Bain and Google surveyed 1,208 B2B buyers, industrial equipment buyers included, and published the results in Harvard Business Review: 86 percent of buyers have a day-one list, about three vendors they already intend to consider before formal research begins, and roughly 90 percent buy from that list. Count the credible vendors in this category: call it ten. Three slots, ten vendors. If lists formed at random, you’d sit on 30 percent of them. So capture at parity is 86 percent times 30 percent: 26 percent of in-market buyers will genuinely consider you.

The second is close. The model uses a flat 5 percent MQL-to-close everywhere, the standard blended benchmark. That flat rate is deliberately conservative in ways that will matter later; my accounts close outbound-sourced meetings at 11 percent. Every cohort gets the same 5 so no row can be accused of a thumb on the scale.

One mechanical note. The in-market pool turns over each quarter; the other pools stand. So the model runs quarterly campaign cycles, four passes a year against the standing pools. Now run it three times.

Table 0: the market you’re running

This is the model with no pain segment drawn. Inbound works the shoppers. Everyone else, all 16,871 companies, gets the spray, at the 0.4 percent rate that published data clocks for generic cold outbound.

Look at the pain row. Two customers. Not because those 2,530 companies lack pain. Because they’re being converted at the 80 percent’s rate. Nobody drew a line around them, so they’re blended into the spray and addressed like companies with no problem at all. The segment doesn’t underperform. It’s unaddressed.

Now look at what this table teaches the CMO who lives inside it. Inbound delivers 44 of 57 customers. Seventy-seven percent. The dashboard proves, quarter after quarter, that inbound is the best channel in the company, and the conclusion is completely rational, because the only alternative ever tested was spray, and spray delivered 11. So the 5 percent gets more budget, and the pain segment never gets its own rate, so it never gets its own line item, so it never gets budget, so it never gets a rate. The dashboard confirms a mix that the dashboard itself created.

If you’re over-indexed on the 5 percent, this is why. It isn’t a mistake. It’s the correct reading of an incomplete instrument.

Table 1: the same market, seen

One change. The 15 percent gets drawn as a segment and worked as one: pain-based email, Facebook, calls, LinkedIn where it fits, running quarterly against the pool at the 4 percent rate that pain-based targeting produces. Nothing else moves. Same budget, same market, same everybody.

Eighteen customers appeared. They were in the market the whole time; twenty existed in both tables, and eighteen of them were invisible because the reporting had no place to put them. A 32 percent lift in total acquisition, and the cost was a segmentation exercise, not a budget line.

(Yes, Facebook, in a B2B post, on purpose. Clay now matches a pain-qualified list to Meta ad IDs at 65 to 70 percent rates and roughly ten dollars a lead. When you can target exactly who you need to target, the platform debate is over. You go where your buyers live, and the owner of an equipment rental company lives on Facebook more than he lives on LinkedIn.)

The published data backs the shape of this table, if you want it from somewhere other than me: the same 2026 dataset that clocks generic cold outbound at 0.4 percent puts signal-targeted outreach at 2.5 percent. Six times, explained entirely by targeting precision. My 4 percent sits above their median because the Permissionless Value Proposition is a sharper instrument than an intent signal. Test that claim; don’t take it.

Table 2: the market you might actually be in

Here’s the variable from the opening, the one that isn’t on your dashboard: the share of your market that is past the existential data point.

In this market, the EDP is fleet utilization. Sustained utilization below 60 percent means an equipment lessor is losing money. Not underperforming. Losing money. And right now, Rouse benchmark data has industry utilization sitting in the low 60s, the weakest reading since 2019, against a 72 percent standard. Rouse publishes the average, not the distribution, so nobody can look up what share of companies sit below the line. But when the average is hovering just above it, roughly half the market is at or under. That’s an inference, labeled as one. Run the model on it.

The mix just flipped. The pain motion out-produces the inbound engine, 67 to 44, and total acquisition more than doubles Table 0. Nothing about the company changed between these tables. The market’s condition changed, and the channel mix that was correct in Table 1 became the wrong mix in Table 2.

Run the algebra and the flip point lands near one-third. If one in three companies in your market is past the EDP threshold, the motion nobody funds out-earns the motion everybody perfects. Below that, your inbound-first instinct is right, and this post owes you an apology it will happily pay. The problem is that you don’t know which side of one-third you’re on, your dashboard can’t tell you, and the aggregate data won’t either. The pain rate of a market is invisible in aggregate and findable one company at a time. That is what the existential data point is for. The EDP finds; it does not tell.

The two territories this model doesn’t touch

The 5 percent mix stays exactly as it is: GEO and SEO, brand, events. Twenty years of refinement, and the model just vindicated it. (One signpost: the buyer’s journey increasingly starts inside a language model, not a search bar, which is why GEO belongs in that mix. Its own post is coming.)

And the 80 percent gets no outreach at all, on purpose. Its channels are brand, content, organic and paid social, and its job is that 26 percent capture number, because capture is just day-one presence wearing a suit, and day-one lists are formed long before the pain arrives. Every point of presence above parity compounds through the Looking row forever. Every spray email into the 80 percent compounds too, with the wrong sign: you’re paying to be filed under noise on lists your future buyers haven’t written yet.

Your ideal customer profile can’t see any of this, incidentally. ICP tells you who resembles a buyer, and resemblance doesn’t tell you which territory a company is standing in today. The territory picks the channel. I’ve made the longer argument in The Great ICP Lie.

The Monday morning audit

Three questions, in order.

Does your reporting break out the 15 percent at all? Not “do we do outbound.” Does the segment of your market in acute pain have its own line, its own rate, its own number that someone owns? If no, you are running Table 0 and calling it strategy, and everything your dashboard tells you about channel performance is a self-portrait.

What is your existential data point? The metric that separates a company with a problem from a company with a crisis. If you can’t name it, no channel mix can save you, because you can’t draw a line you can’t see.

What share of your market is past it? Not the average. The share. Because somewhere around one-third, everything you believe about your channel mix inverts, and the only way to know your number is to go find it, one company at a time.

Before you move a dollar on the answers: prove it small. The Two-Week Test Protocol exists for exactly this. One segment, one message, one baseline, then scale.

The channels were never the problem. The map was never the problem, either. You were missing a number.

Sources: John Dawes, “Advertising Effectiveness and the 95-5 Rule,” Ehrenberg-Bass Institute (2021). Bain & Company and Google, “What B2Bs Need to Know About Their Buyers,” Harvard Business Review (2022). IBISWorld, Heavy Equipment Rental in the US (2025). Rouse Services rental utilization benchmarks via MHEDA (Q1 2025); the 50 percent below-threshold figure is an inference from the published average, labeled as such. 2026 outbound benchmark data (0.4 percent generic median, 2.5 percent signal-targeted). Ebsta x Pavilion 2025 GTM Benchmarks and Champify 2025 Impact Report corroborate the win rates cited. The 15 percent resting-state pain figure and the 4 percent pain-based MQL rate are practitioner measurements from Cannonball GTM engagements. Model available on request; every input is sourced or countable, and I want you to check the math.

Follow the wire

You’ve read the data center stories. The gigawatt campuses. The utility deals. The town hall fights. Amazon, Alphabet, Meta, Microsoft, and Oracle plan to spend somewhere between $650 and $700 billion this year, most of it aimed at AI infrastructure.

Every one of those stories stops at the same place. The substation. The megawatts get approved, the reporter files the piece, and the story ends.

Follow the wire instead.

Past the substation, the voltage is still too hot to touch a building. It has to step down one last time, through a gray cylinder on a pole or a green box on a corner. That device is a distribution transformer. Every new house ends at one. Every EV charger. Every data center hall. No box, no power. No power, no story.

Here’s what’s happening to the box. Before 2020, a utility ordered distribution transformers and got them in 8 to 12 weeks. By 2022, per a letter entered into the Senate record, the wait had stretched as long as three years. More than half the transformers on the American grid are older than 33 years. And the load is still coming: the EIA projects record national power demand this year, after fifteen years of the flattest demand curve in modern history.

The utilities know. That’s why 51 investor-owned utilities just proposed at least $1.4 trillion in capital spending through 2030, a number Morningstar DBRS calls a super-cycle, double the investment of the prior decade.

The most invisible product in the economy is missing, the buyers are panicking, and a trillion dollars is rolling downhill toward a gray box.

So we ran demand segmentation on it.

Thirty seconds of re-grounding

If you read The Case for Demand Segmentation, skim this. If you didn’t, here’s the model.

Every market that serves institutions has two curves. Load is what the market demands. Capacity is what the installed infrastructure can process. Factors feed each curve, and the factors are publicly observable. Neither curve matters alone. The condition that matters is the gap: load pulling away from capacity, compounding, refusing to fix itself.

The point where load crosses capacity is the Demand Breakpoint. Before the crossing, the market has pressure. After it, the market must produce a buyer. The segment that forms around a breakpoint is a Demand-Qualified Segment, the DQS, the sibling of the Pain-Qualified Segment you already know from Finding Hidden Customers. Pain finds the buyers of today. Demand finds the buyers of tomorrow, while they’re still forming, while no competitor is looking.

That’s the theory. This post is the theory with its sleeves rolled up.

The run

The market: distribution transformers. The example brand: ERMCO, a cooperative-owned manufacturer out of Dyersburg, Tennessee, roughly $774 million in revenue, one of the last major American names in the category. Standard disclosure, same as every teardown we publish: no engagement, no inside information, public data only. ERMCO has never heard of us. By the end of this post you’ll know why we picked their market anyway.

One more thing before the stages. We built the process and ran it at the same time, drafting each prompt, executing it, and correcting what broke. What follows is the run as it happened, corrections included. The corrections are the education.

Stage 1: Find the curves

The first prompt asks for the load curve, the capacity curve, and the factors feeding each. It also asks a question that didn’t exist in our pain-based work: when load exceeds capacity, why MUST the owner of this infrastructure respond? Name the forcing mechanism. If there isn’t one, stop, because you don’t have a demand segment. You have a trend.

The transformer market answered with the strongest forcing mechanism we’ve ever seen. It’s called the obligation to serve. An electric utility legally cannot refuse new load. A data center files an interconnection request, a subdivision gets platted, and the utility must build the capacity to serve it. Not should. Must. It’s the law of the franchise.

Sit with that. In the forensic equipment market we studied last time, a county under pressure would eventually respond. Here, the buyer’s response is a legal duty. The Demand Breakpoint doesn’t predict a buyer in this market. The law manufactures one.

Stage 1 also split the product line in two, and the split runs the rest of the post. Single-phase transformers follow residential load: housing starts, new meters, subdivisions. Three-phase transformers follow commercial and industrial load: data centers, chip fabs, EV fleets. Two products, two load curves, two different Americas. The prompt has a rule about this: do not average them. Averaged curves hide both breakpoints.

Stage 2: Find the records

The second prompt catalogs the public data sources behind every factor, then ranks each source on two axes. Does it name a specific buyer, or just describe a geography? Does it appear before the buying event, or after? A source that names a buyer before the buying event is gold. Everything else is supporting cast.

The gold in this market is a document most marketers have never heard of. Rural electric cooperatives that borrow from the USDA must file a Construction Work Plan: a four-year engineering plan listing every substation upgrade, every feeder, every system improvement, with the load forecasts justifying each one. Named utility. Named projects. Filed with the federal government, surfacing through public environmental review, years before a purchase order exists.

Read that again the way we did. The buyer describes their own capacity problem, in their own words, in a federal filing, up to four years before they buy. In the forensic market we found something similar by accident and treated it as a gift. This time the prompt hunted for it on purpose. The lesson generalizes: in regulated markets, there is almost always a document where the buyer confesses. Your job is to learn its name.

The catalog also produced a free national diagnostic. EIA Form 861 is an annual census of every utility in America: sales, customer counts, peak demand, and reliability statistics, by name. Multi-year sales growth against worsening reliability numbers is the load and capacity divergence, visible per utility, in one downloadable file. And the confirmation layer sits on top: Construction Work Plans, interconnection queues, and municipal bonds on MSRB EMMA. Yes, EMMA again. Same database that found the county morgue projects. One public system, two industries, zero dollars.

Stage 3: Rank the demand

The third prompt builds the concentration model, and it carries a scar from the forensic run as a law: volume beats rate. Rank by absolute projected growth, never by percentage. Percentages flatter small markets. A co-op adding 400 meters at 12 percent growth is a story. A co-op adding 9,000 meters at 4 percent growth is a customer.

Run against the grid, the model split down the product line exactly as Stage 1 predicted. Three-phase demand concentrates violently: a few dozen service territories in known corridors, Northern Virginia, central Ohio, Georgia, Texas, Arizona. Single-phase demand does the opposite, spreading across hundreds of growing co-ops and munis in the Sun Belt’s growth rings. Two rankings, two shapes, two sales motions.

And then the overlay: the entities that rank high on both lists. A fast-growing metro co-op that just caught a data center siting is carrying both curves on the same substations. That intersection is the top of the entire stack, and no single-curve model can see it.

(ERMCO, for the record, appears to see it fine. In January they announced their first plant west of the Mississippi: a 566,000 square foot three-phase facility in Maricopa County, Arizona. Look at the three-phase corridor list again. That’s demand segmentation executed in steel.)

Stage 4: Map the formation

The fourth prompt maps the matriculation chain: the ordered public stages a buyer passes through between breakpoint and purchase, per buyer type. Co-ops move through engineering studies, federal work plans, and loan reviews. Munis move through capital plans, council votes, bonds, and statutory public bids. Investor-owned utilities move through resource plans, rate cases, and vendor qualification.

The chain also grows people. That’s what matriculation means here. A board vote creates a project sponsor. An engineering study creates an engineer of record. In all three buyer types, that engineer of record shows up one stage before the specs freeze, holding more influence over the eventual purchase than anyone with “procurement” in their title. Most underrated person in the market.

And then Stage 4 broke something we believed. We assumed every chain ends in an RFP, because in the forensic market it did. In this market, half the time, there is no RFP. Transformers are recurring purchases, and utilities buy them through standing frameworks: master agreements, alliance contracts, approved-vendor lists. The visible buying event is just a release against a framework negotiated years earlier.

Which changes the objective entirely. In project markets, you chase the event. In flow markets, you win the framework slot before the load arrives, and the breakpoint model tells you which frameworks are about to be worth the most. The deadline that matters isn’t the purchase date. It’s the last reachable moment: the point after which the spec is written, the framework is set, and the decision is functionally closed. In a flow market, that moment can arrive years before any money moves. The whole argument for demand segmentation lives inside that sentence.

Stage 5: Sort the geometry

The fifth prompt places every ranked entity relative to its breakpoint, using evidence tests, not vibes. Past the breakpoint: load provably exceeds capacity, and the entity appears somewhere in its matriculation chain. At the breakpoint: the curves are crossing, no chain record yet. Approaching: capacity still holds, but the velocity makes the crossing a matter of time, with an estimated window and the named public filings that will announce each promotion.

The approaching class becomes a watchlist with trigger feeds: which filings to monitor, at what cadence, and what promotion means. Not a list. A standing surveillance program, built from free public data.

And then the sort produced a class we didn’t ask for. Entities that test past their breakpoint, whose reliability numbers are degrading in public, whose load is provably arriving, and who appear nowhere in any matriculation chain. No work plan. No bond. Nothing filed.

Think about what that is. An institution that legally must respond, under the obligation to serve, that has not started responding. Usually a small utility without the engineering staff to begin. Invisible to every competitor, because there’s no filing to see. Inevitable as a buyer, because the law leaves no exit. We named the series Finding Hidden Customers two years ago. This is the most hidden customer we have ever found: the buyer that doesn’t know it’s a buyer yet, in a market where it has no legal choice.

Stage 6: The verdict

The sixth prompt assembles everything into a brief and forces a verdict, with disqualifiers it can’t talk its way around. No forcing mechanism, reject. No gold-tier source, monitor at best. It also runs a screen the demand pipeline doesn’t produce on its own: is there a pain path in this market too?

There is, and it was sitting in the same dataset the whole time. More than half the installed transformer fleet is past 33 years old, against a roughly 40-year design life, and the reliability statistics that measure the decay are filed annually, in public, by utility name. Aging equipment inside existing customers, observable through trailing public data. That’s a textbook pain segment, found by the demand pipeline’s exhaust.

Verdict: proceed, dual-path. Pain path into utilities whose installed fleet is failing today. Demand path into utilities whose load arrives tomorrow. Same verdict the forensic market earned, reached from the opposite direction, which is the strongest evidence yet that the two paths aren’t a coincidence. They’re the architecture.

The commodity trap

Now the objection you’ve been holding since the second section. This industry is backlogged for years. Manufacturers are allocating, not selling. Why would anyone in a shortage need demand intelligence?

Because the shortage is temporary and the market structure isn’t. A distribution transformer is a spec product. When supply normalizes, and $1.8 billion of announced manufacturing expansion says it will, every vendor’s product meets the same IEEE standards, and the category snaps back to what it has always been: a commodity fight, won on price.

Unless you’re already inside. Here’s what the breakpoint model actually buys in a market like this. The vendor who found the forming buyer two years early is in the room when the engineer of record writes the spec. Their framework bid isn’t a bid; it’s a continuation of a relationship the competitors didn’t know existed. The sales cycle runs faster because the discovery happened years ago, for free, in public filings. The cost of sale drops for the same reason. And the price holds, because the differentiator was never the steel in the box. In a commodity market, knowing the buyer before they’re a buyer is the only differentiation left that a competitor can’t copy by Thursday.

Shortages end. Framework positions persist. That’s why a backlogged industry is exactly where you run this.

Your market has a gray box in it

Last time we found buyers forming around county morgues. This time, the national grid. The six stages didn’t change between markets. Not one prompt needed a rewrite for the subject matter, only for what the run taught us about the method itself.

So the test for your market is the same four questions. Are load and capacity publicly observable? Does the gap compound instead of self-correcting? Is the responding buyer an institution that must act? Does its formation run through public records? Four yeses and there are buyers forming in your market right now, in plain sight, on a schedule you can read.

Everything below is what we used to read it.

The prompt progression (free)

The six prompts from this run, frozen at v1.0, each as a Claude artifact you can open and run. They run in sequence; each stage’s output feeds the next. Expect your market to correct you somewhere around Stage 3. Let it. The corrections are where the method learns your market.

Prompt 1: Curve Identification. Start here with your product category and buyer types. Finds both curves, their factors, and the forcing mechanism, or tells you honestly that you don’t have one.

Prompt 2: Data Source Catalog. Turns the factors into named public databases, ranked by whether they identify buyers and how early. Hunts for the document where your buyer confesses.

Prompt 3: Demand Concentration Model. Ranks named entities by volume, gap, and velocity. Volume beats rate. Separate rankings per product line, plus the overlay class.

Prompt 4: Matriculation Chain. Maps the formation stages per buyer type, the records each stage files, the people each stage creates, and the last reachable moment.

Prompt 5: The Geometry Sort. Places entities past, at, or approaching their breakpoint with evidence tests, builds the watchlist, and reports the anomalies. The hidden customers live in the anomaly report.

Prompt 6: The Demand Brief. Assembles the run into a decision: proceed dual-path, proceed demand-only, monitor, or reject, with disqualifiers that keep it honest.

One more thing. This progression is becoming an agent, the way Finding Hidden Customers did. Run it by hand first anyway. The hours you spend running it manually are how you’ll know, later, exactly what the agent should never get wrong.

Start here: The Case for Demand Segmentation | Go deeper: Finding Hidden Customers: The Playbook | See it in action: We Built an Agent for Finding Hidden Customers. Here’s Yours.

The methodology we walked in with

If you’ve been reading Cannonball for a while, you know the play: Pain-Based Segmentation finds companies living with a systemic problem. Not companies that match a firmographic profile. Companies that are actually hurting, right now, in a way you can observe through public data.

The anchor of that methodology is the Existential Data Point. The EDP is the single metric that marks the line between a healthy business and a failing one. It has one rule: it must be systemic, not episodic. A systemic condition compounds over time unless someone intervenes. An episodic one is a bad day that a company can recover from. The citation is episodic. The unsafe condition behind it is systemic. That distinction is the whole game.

Here’s the part I didn’t question. Every EDP we had ever found lived inside a company. Maintenance ratios. Compliance gaps. Revenue leakage. All internal conditions. I thought that was a rule of the methodology. It was just a pattern we hadn’t broken yet.

The run that broke the pattern

A few months ago we ran the Finding Hidden Customers methodology against a market we’d never touched: forensic capital equipment.

We’ve all watched some version of CSI at some point. Think about everything you’d see in a forensics lab. The tables, the coolers, the instruments. Somebody manufactures all of it, and somebody buys it. High ACV. Long sales cycles.

We expected a normal pain-based result. We got one. Pathology labs work with formaldehyde, and OSHA strictly regulates exposure. Facilities running old equipment accumulate exposure problems, and the enforcement data makes them visible. A clean, systemic, inside-the-building EDP. Classic play.

Then the second path showed up, and it didn’t fit the model.

A county’s death volume is driven by population and demographics. When deaths outpace what the local medical examiner facility can process, the county has a structural problem. Bodies don’t wait. Caseloads don’t shrink on their own. And here’s the thing that stopped us: that problem doesn’t live inside any company this manufacturer could sell to today. It lives in the market. And when it gets bad enough, it generates a brand-new buyer: a county that must build or expand a forensic facility, which means buying exactly the equipment this manufacturer sells.

The pain wasn’t inside a company. The market itself was the patient.

Load and capacity

Consumer marketers have understood a version of this forever. Durable goods demand is famously predictable. Appliance makers know roughly how many washing machines will die this year. Automakers model fleet age down to the quarter. Replacement demand is math, and B2C has been running that math for decades.

But B2C only ever knows the number in aggregate. Four million washers will fail this year. Which houses? No idea. Doesn’t matter. Mass channels exist to catch probabilistic demand.

Your market doesn’t work that way, and that turns out to be an advantage. Here’s the model.

Every market that serves institutions has two curves.

Load is what the market demands. For forensic facilities: death volume, population growth, demographic aging. For a water system: population and new housing connections. Load is driven by factors that are publicly observable.

Capacity is what the installed infrastructure can process. Facility age. Storage. Throughput. Rated limits. Also driven by factors that are publicly observable.

Neither curve matters on its own. A 90-year-old facility in a shrinking county is just an old building. High population growth served by a brand-new plant is a healthy market. The condition that matters is the gap: load pulling away from capacity. And that gap behaves exactly like a systemic pain condition. It compounds. It does not fix itself. Counties can’t shrink their death volume. School districts can’t cap enrollment. The gap widens until something structural has to give.

Same systemic test we’ve always run. Different location. The condition exists in the market rather than within a company.

The Demand Breakpoint

The point at which load exceeds capacity is the Demand Breakpoint (or DBP).

The DBP marks the line between latent demand and inevitable demand. Before the crossing, the market is under pressure. Somebody should probably do something. After the crossing, the market must produce a buyer. Not “might.” Must. The caseload physically exceeds what the infrastructure can handle, and an institution is structurally forced to respond.

And just like the EDP, the breakpoint creates the segment structure on its own. Entities past the breakpoint are active buyers forming right now. Entities at the breakpoint are next. Entities approaching it are on your watchlist, and they’re worth more than any intent data feed you’re paying for, because you know they’re coming years in advance.

Which gives Finding Hidden Customers two segment types instead of one:

A Pain-Derived Segment (PDS) is a set of companies currently living with systemic pain, identified through trailing indicators in compliance data.

A Demand-Qualified Segment (DQS) is a market-creating structural demand for buyers who don’t yet exist, identified through leading indicators in demographic, financial, and infrastructure data.

These are segment types, not quality grades. One finds the buyers of today. The other finds tomorrow’s buyers.

The paper trail runs years ahead

Here’s why the DDS is worth the work: it comes down to one fact about institutions: they cannot create demand privately.

A household replaces a refrigerator quietly. Nobody files paperwork. A county does not quietly replace a morgue. It debates the funding in public meetings. It puts a bond on a ballot. It votes. It buys land, files zoning, pulls permits, and issues an RFP. Every stage of that formation leaves a public record, and the records appear in sequence:

Mortality data show the pressure building. The capacity gap becomes visible. The funding debate starts. The bond gets issued (searchable on MSRB EMMA, the municipal bond disclosure system). The land use filing names the site. The building permit starts the clock. The RFP arrives last.

Now look at where everyone else picks up that story. A traditional vendor enters at the RFP, the very end of the chain. Construction data providers like Dodge enter at the permit. This methodology starts with mortality data and computes the gap, putting you 2 to 4 years ahead of the RFP.

We checked whether this was theoretical. It isn’t. A single research pass surfaced 6 county forensic facility projects currently in the funding-to-construction window, every one traceable through public bond records, exactly in the sequence the model predicts. The buyers were forming in plain sight.

The advantage isn’t better data. Everyone can see these records. The advantage is operating upstream of where others are looking.

Not just morgues

The forensic case is vivid, but the structure repeats in four respects: load and capacity are both publicly observable; the gap compounds rather than self-correcting; the responding buyer is an institution that must act; and the buyer’s formation runs through public records.

Run that test against these markets. Water and wastewater: load is population and new connections, capacity is EPA-permitted plant ratings, and a consent decree is literally a codified breakpoint. The electric grid: interconnection queues are a public ledger of demand that hasn’t been served yet. K-12 facilities: enrollment projections against seats, and every portable classroom in a school parking lot is a breached breakpoint you can see from the street. Corrections. Healthcare in Certificate-of-Need states, where the matriculation chain is unusually legible because filings are mandatory.

If you sell equipment, infrastructure software, or services into any market like these, there is a load curve and a capacity curve with your name on them.

What demand is not

One distinction before we close, because the word “demand” carries baggage. This is not demand generation. Nobody here is creating interest. Nobody is warming anyone up. The county’s problem exists whether or not anyone ever markets to it. Demand-driven segmentation finds structural demand that already exists in the market and identifies the buyer it’s about to create. Marketing didn’t make the demand. Demographics did.

Run both paths

The forensic equipment manufacturer got a verdict we’d never issued before: proceed, dual-path.

The pain path targets existing pathology labs with exposure problems, observable through OSHA data. The demand path targets counties approaching or past their breakpoint, observable through mortality data and bond records. Two segments. Two data sources. Two message strategies. One brand. The pain path captures the demand that exists today. The demand path captures the demand that arrives tomorrow, before any competitor knows it’s coming.

So here’s the question to take back to your own market. You already know where the pain is. Now find the two curves. What’s the load in your market? What’s the capacity? And where do they cross?

Somewhere out there, your best buyer doesn’t exist yet. The public record already knows who they’ll be.

New to Cannonball GTM? We help growth leaders find hidden buyers in publicly available data, and figure out which tools, agents, and platforms are actually worth the money. Start here:

• Start here: Finding Hidden Customers: The Playbook

• Go deeper: Beyond the ICP: A Guide to Pain-Based Segmentation

• See it in action: We Built an Agent for Finding Hidden Customers. Here's Yours.

Here’s the thing about that confusion. It’s not your fault either, and it’s not because the words are foreign. It’s because they’re English. Context. Memory. Skills. Tools. Agent. Nobody looks up a word they already know, so the confusion has no exit. You scrolled past MCP three times this week and nobody defined it. Nobody will. In your feed, nobody defines anything, because nobody wants to be the one who asks.

And here’s the confession: those of us who live in this world get confused, too. The vocabulary shifted under all of us, and the people speaking it most fluently learned it about twenty minutes before you did.

So this glossary is for the crinkled brow crowd. Twenty-seven terms. Read it top to bottom and it works like a course, because the entries are sequenced, not alphabetized. Each one starts with a moment you’ve already lived, because you’ve been experiencing these concepts for two years. You just never got handed the words.

A few promises. Nothing here is for sale. There’s no tool list, no vendor pick, no upgrade pitch at the bottom. And this document is dated on purpose: some of these words will mean something different in a year, and when they do, I’ll update this page. That’s the deal.

Read it once. Then open your feed and watch it become legible.

Part One: The Foundation

What you’re actually working with.

Large Language Model (LLM)

What it is. The engine under every AI tool you’ve touched in the last two years. An LLM is software trained on a staggering amount of text until it gets very, very good at one thing: predicting what words come next. Do that well enough and the predictions start to look like understanding. Often they function like understanding. But the middle word of the name is the honest one. Language. Not knowledge.

Where you’ve already felt it. Both halves, daily, for two years. The good half: it drafts in your voice, summarizes a forty-page PDF, explains a term of art like a patient colleague. And then the other half: it hands you a statistic that’s specific, confident, and completely made up. Same tool, same afternoon. You’ve probably wondered how something so fluent can be so wrong. Now you know. It’s fluent because fluency is the whole trick.

Worth knowing. Language, not knowledge. That one distinction explains most of what this glossary covers: why the writing is beautiful, why the facts need checking, why grounding it in your documents changes everything, and why “hallucination” has its own entry later. An LLM isn’t a database that sometimes writes well. It’s a writer that sometimes knows things. Build with that in mind and most of the surprises stop being surprises.

Model

What it is. One specific, versioned build of that engine. Trained at a certain time, at a certain size, with certain strengths. The names and numbers you see (the 4s, the 5s, the words like Opus and Sonnet) aren’t marketing garnish. They’re versions, and the differences between them are real.

Where you’ve already felt it. The dropdown menu at the top of the chat. The one you’ve been ignoring. Switch it and the answers change: faster but shallower, slower but sharper, better at code, better at prose. You’ve also felt it the morning a tool you use got noticeably smarter and no release notes explained why. Nothing about the tool changed. The model underneath it did.

Worth knowing. Every AI product you’ll ever evaluate is built on a model, and usually it’s one of the same handful everyone else builds on. The tool is the car. The model is the engine. Which means “which model is this built on” is one of the most clarifying questions you can ask about any product, and one of the few the demo won’t volunteer. You’re not asking to be technical. You’re asking what you’re actually buying.

Token

What it is. The unit the AI actually reads. Not words: chunks of words, roughly three-quarters of a word each. You write sentences. The model receives tokens. And everything in this world is measured in them. Pricing. Usage limits. The size of the context window you’ll meet in the next entry.

Where you’ve already felt it. The long paste that got cut off. The “message too long” error. The tool that stopped you mid-task and told you you’d hit a limit you never agreed to. Those weren’t arbitrary walls. They were token limits, doing exactly what they were set to do, in a currency nobody told you existed.

Worth knowing. You never need to count tokens. Not once. You just need to know the currency exists, because it explains the shape of everything: why long documents cost more to process, why “unlimited” plans carry asterisks, why the context window has a ceiling. Limits stop feeling like malfunctions once you know what’s being metered. It’s the difference between a dead car and an empty tank. Same stopped car. Different problem.

Context Window

What it is. The AI’s working memory for one conversation. Everything lives inside it: your messages, its answers, every document you paste. The window has a fixed size. Run the conversation long enough and the oldest material starts falling out. No warning. It just knows less than it did an hour ago.

Where you’ve already felt it. You built something good in a chat. Real progress, shared shorthand, the thing finally understood your business. You came back the next morning, opened a new chat, and were talking to a stranger. That’s a context window. Every conversation is its own window, and a new chat is an empty one. You’ve felt it mid-conversation too: brilliant for twenty minutes, then repeating itself, then forgetting a decision you made together. It didn’t get dumber. The window filled up.

Worth knowing. More first-agent frustration traces back to this concept than any other in this glossary, and none of it is your fault. Nobody told you the memory had walls. It’s manageable once you can name it: anything the AI should always know belongs in standing instructions, because chat history is the part that evaporates. And when a vendor says their agent remembers everything, this is the word for what they’re glossing over. Memory features are real. They’re workarounds for the window, not the absence of one.

System Prompt

What it is. The standing instructions an AI carries into every conversation. Who it is, what it’s for, how it should behave, what it should never do. Written once, applied always. The chat is the conversation. The system prompt is the job description.

Where you’ve already felt it. You wrote one the first time you told an AI “you are a B2B copywriter, always write at grade 8, never use jargon” and noticed the whole session held that shape. That was a system prompt, just typed by hand into the wrong place. The packaged version means you stop retyping it. It’s there before you say hello.

Worth knowing. This is the answer to the context window problem from the previous entry, and it’s the first real design decision you’ll make when you build anything. What belongs in the standing instructions, and what belongs in the conversation? Get that split right and the AI feels consistent. Get it wrong and you’re re-explaining your business every morning to a stranger with excellent manners.

AI Agent

What it is. Software that pursues a goal by taking actions, not just answering questions. That’s the honest definition, and it’s worth memorizing, because it’s the yardstick for every product that will be pitched to you this year. A chat answers. An agent does. It plans steps, uses tools, checks its work, and keeps going until the goal is met or it needs you.

Where you’ve already felt it. The first time an AI didn’t just tell you how to build the spreadsheet but built it. Searched, pulled the data, formatted the thing, handed it back. Something shifted in that moment and you probably couldn’t name it. What shifted was the category. You stopped talking to a very well-read intern and started delegating to one.

Worth knowing. Watch the grammar in the pitches. When agent is a noun, it describes software, and you can ask what the software does. When agentic is a floating adjective (agentic mode, agentic workflows, agentic transformation), it usually describes marketing. You’ve scrolled past that adjective a dozen times this month. The next time it shows up in a demo, ask the question you already know how to ask: what do you mean by agentic? The good vendors have an answer. The others have another adjective.

Agentic Workflow

What it is. And here’s the legitimate use of that word, one entry after the warning. An agentic workflow is a chain of AI steps where the output of one becomes the input of the next, moving toward a goal. Research feeds drafting. Drafting feeds review. Review feeds the send.

Where you’ve already felt it. You built one by hand. The prompt whose output you copied into a second prompt, whose output you copied into a third. That copy-paste relay was an agentic workflow with you as the connective tissue. An agent is just the copy-paste, automated.

Worth knowing. This is the most demystifying entry in the glossary, so sit with it. The scariest-sounding term in the category describes something you’ve already done manually, probably this week. Which means the right first question about any agent isn’t “what can it do.” It’s “which of my existing relays should stop running through my hands.” You already know your relays. That knowledge is worth more than any demo.

Part Two: The Plumbing

The acronyms in your feed and in the pitch deck.

MCP (Model Context Protocol)

What it is. A standard way for an AI to plug into other software. Your CRM, your calendar, your data tools. Before the standard, every connection was custom-built, one at a time, priced like custom work. MCP is the common plug shape. Any AI that speaks it connects to any system that speaks it. No custom wiring.

Where you’ve already felt it. Mostly in your feed, at speed. The acronym went from nowhere to everywhere: launches “via MCP,” tools shipping “an MCP,” whole workflows connected through it, nobody defining it once. You may have felt it directly too, the moment an AI reached into your actual calendar instead of asking you to copy-paste your week. That connection was very likely this.

Worth knowing. MCP quietly repriced integration. “We integrate with your stack” used to be a moat: months of engineering, billed accordingly. A standard plug collapses the moat. When connecting to your systems becomes something any tool can do, it stops being a reason to pick a vendor and stops deserving a premium. Your stack will connect to AI whether or not you lift a finger. What’s still yours to decide, and what’s actually worth paying for, is what happens after the plug clicks in.

Skills

What it is. A set of instructions the AI loads on demand for one specific job. Not a conversation. Not a standing rule that applies to everything. A packaged capability: how we write case studies, how we score inbound, how we build a QBR deck. The AI picks it up when the job calls for it and puts it down when the job is done.

Where you’ve already felt it. You’ve almost certainly built one without the word. The document you paste into every campaign brief. The “here’s exactly how we do this” note you keep somewhere and reuse. That’s a skill, minus the packaging. The packaged version means you stop pasting. And if you missed the on-ramp when this concept broke through in early 2026, you’ve felt that too: everyone in your feed building skills, sharing skills, installing skills, and no safe place to ask what one is. This entry is that place.

Worth knowing. A prompt is spent the moment you use it. A skill works every time after. That’s the whole economics of the thing, and it’s why skills are where your team’s process knowledge starts to compound instead of evaporating in chat history. If you build one skill this quarter, make it the process your team explains most often. The one you’re tired of repeating. That’s the one worth packaging.

API

What it is. The doorway software uses to talk to other software. Application Programming Interface, but forget the expansion. It’s the plug, not the appliance. When an agent updates your CRM, it walks through the CRM’s API to do it.

Where you’ve already felt it. Every time you wondered why the chat window can’t see your pipeline but some tool built on the same AI can. The chat has no doorway into your systems. The tool does. Same intelligence, different access.

Worth knowing. This is a quiet entry, and that’s deliberate: you’ll never touch an API, and you don’t need to. You just need the word to stop being fog. When someone says “we’d need API access,” they’re saying “we need the doorway unlocked.” A permissions question, not a technical one. And permissions questions are ones you already know how to have opinions about.

Tool Use (Function Calling)

What it is. The ability of an AI to reach outside the conversation and do something: search the web, run a calculation, update a record, send the draft. Function calling is the engineer’s term for the same idea. Tools are what turn a chat into an agent.

Where you’ve already felt it. The first time the AI paused mid-answer, searched the web, and came back with something published yesterday. Or the moment it actually edited the spreadsheet instead of describing the edits. A small pause, then an action. That pause was the tool call.

Worth knowing. This is the dividing line the whole category rests on: an assistant that knows versus an assistant that does. So when a vendor demos an agent, the load-bearing question is which tools it can use, because the tool list is the honest résumé. An agent with no tools is a chat with a nicer landing page. An agent with the right tools is the thing everyone’s been promising you.

RAG (Retrieval-Augmented Generation)

What it is. A technique that lets an AI look things up in a defined body of documents before answering, instead of answering from general training alone. Retrieval, then generation. An open-book exam instead of a closed-book one.

Where you’ve already felt it. Every time an AI answered from the whole internet’s general wisdom when you needed it to answer from your pricing sheet. And the first time it did the opposite: cited your actual doc, your actual number, your actual policy. The difference in trust was immediate. That difference has a name, and vendors say it constantly.

Worth knowing. RAG is why “trained on your data” is usually the wrong mental model and the wrong thing to ask for. Most tools aren’t retraining anything. They’re looking things up in your documents at the moment of the question, which is cheaper, faster, and easier to keep current. The book gets updated. The student doesn’t need surgery. Keep that image and half the enterprise AI pitches you hear this year will get simpler.

Knowledge Base (Grounding)

What it is. The defined body of information an AI is pointed at. Your docs, your data, your ground truth. Grounding is the practice of making the AI answer from that base instead of from vibes. RAG, one entry up, is the plumbing that does it.

Where you’ve already felt it. The folder of documents you uploaded to a project so the AI would stop guessing about your business. That was you, grounding a model. It’s why the same engine gives strangers generic answers and gives you specific ones. The difference was never the intelligence. It was what the intelligence was standing on.

Worth knowing. Another quiet entry, with one loud implication: the quality of any AI system in your company now has a ceiling, and the ceiling is the quality of the documents you feed it. Stale pricing in the knowledge base means confident answers about stale pricing. Garbage in has a new career. It sounds authoritative now.

Wrapper

What it is. A product built as a thin layer on top of someone else’s model. Interface on top, maybe a system prompt and some plumbing, the same engine underneath that powers the $20-a-month chat you already use.

Where you’ve already felt it. The dawning suspicion, mid-demo, that this $40,000 platform is a system prompt with a login page. You couldn’t articulate why the output felt familiar. It felt familiar because you’ve been generating output like it all year, one floor down, at consumer prices.

Worth knowing. Here’s the honest version, because the word gets thrown around as pure insult: some wrappers are worth every dollar. Deep workflow, real integrations, accumulated domain logic you’d take a year to rebuild. Others are a markup on nothing. The question that separates them is the one from the Model entry, plus one more: what did you build on top of the engine, and what would I lose by doing this myself in a project with a good system prompt? Vendors with real answers enjoy that question. Listen for the enjoyment.

Fine-Tuning

What it is. Actually retraining a model on your examples until its default behavior changes. Real, powerful, occasionally necessary. Expensive in money, time, and maintenance, because a fine-tuned model is a fork you now own.

Where you’ve already felt it. In a pitch, most likely. “We’ll fine-tune a model on your data” has a specific gravity in a sales conversation. It sounds like commitment. It sounds like the serious option. And it’s usually the answer to a question nobody asked.

Worth knowing. You almost certainly don’t need it. Most of what fine-tuning gets pitched for is solved by a good system prompt and a grounded knowledge base, at a hundredth of the cost, changeable by Tuesday. Fine-tuning is how a prompt problem becomes an enterprise price tag. There are legitimate uses, deep and narrow ones. But the burden of proof sits with whoever proposed it, and “we tried instructions and grounding first, here’s where they failed” is what proof looks like. If they can’t show you the failure, they’re selling the price tag.

Prompt Engineering

What it is. The craft of phrasing instructions to get better output from a model. A real skill. Also a shrinking one, because every model generation needs less coaxing than the last.

Where you’ve already felt it. The magic-words phase. The threads promising ten prompts that would change your life. The colleague with the prompt library. You maybe felt guilty for not studying harder. And then you noticed the newer models just sort of got what you meant, and the guilt quietly expired.

Worth knowing. The prompt was never the asset. The thinking behind the prompt was: what to ask for, in what order, judged by what standard. That thinking survives every model upgrade. The phrasing doesn’t. So invest accordingly. Process over prompts, and if you want the durable version of what prompt engineering was reaching for, it’s four entries back. It’s called a skill.

Part Three: The Categories

What agents actually do for your pipeline.

Task Agent

What it is. An agent that does defined work you don’t have time for. Research the account. Enrich the list. Draft the follow-up. Format the report. Clear inputs, clear outputs, a human deciding what happens next.

Where you’ve already felt it. Look at your Tuesday. Somewhere in it is three hours of work that requires your judgment for five minutes of it. The other two fifty-five is collection, formatting, and moving text between boxes. That two fifty-five is the task agent’s whole job description. You’ve been feeling this category every week for years. It just looked like your calendar.

Worth knowing. This is where every growth team should start, and it’s the category vendors are least excited to sell you, because it’s too cheap. There’s no platform fee in “an agent that does your account research.” The pitch economics favor the ambitious categories below. Your actual economics favor this one. Start here, boring and unglamorous, and let the wins fund your education for what comes next.

Interpretation Agent

What it is. An agent that reads what you already have and tells you what it means. Call transcripts, CRM notes, churned-account patterns, win-loss threads. Not more data. Better reading of the data nobody reads.

Where you’ve already felt it. The transcript folder. Every call recorded, dutifully, for two years. Nobody has read them. Nobody was ever going to read them. Somewhere in there is the phrase your best customers keep using and your website doesn’t, and it’s been sitting there the whole time, patient as a fossil.

Worth knowing. The instinct this category corrects is the expensive one: when insight runs low, buy more data. Another intent feed, another enrichment layer. Meanwhile the richest data you’ll ever own is the verbatim voice of your own buyers, pre-paid, sitting in storage. An interpretation agent is how it stops being storage. Before you sign the next data subscription, ask what percentage of the data you already own has ever been read. The answer is a number below five, and it’s the strongest business case in this glossary.

Autonomous Agent

What it is. An agent that acts without a human approving each step. It sends the email. It books the meeting. It makes the change. You find out afterward, by design.

Where you’ve already felt it. The demo where nobody touched the keyboard. Impressive, genuinely. And somewhere in your chest, a small accurate flag: that thing just did something, and no one checked it first. Keep the flag. The flag is calibrated correctly.

Worth knowing. Autonomy isn’t a feature grade, it’s a risk transfer. Every step a human stops reviewing is a category of error you’ve agreed to discover after it ships. Sometimes that trade is right: low stakes, reversible actions, proven agent. But it’s a decision, and vendors present it as a default. The entries in Part Four are the vocabulary for making it a decision again. Approach this category the way you’d approach any hire who wants to skip the review process on week one: slowly, and with a very clear write-up of what happens when they’re wrong.

Copilot vs. Agent

What it is. Two words used interchangeably in pitches that describe structurally different products. A copilot assists a human doing the work: suggests, drafts, sits alongside. An agent does the work: executes, completes, reports back. One is a passenger seat. The other has the keys.

Where you’ve already felt it. The pitch that said agent and demoed a copilot. Something felt off and you couldn’t name it. What felt off was that the ROI math had been done for an agent (replace the hours) while the product was a copilot (improve the hours). Those are different numbers. Sometimes very different.

Worth knowing. Neither category is the wrong one. Copilots are the right buy for judgment-heavy work; agents for defined, repeatable work. What’s wrong is paying agent prices for copilot function, and the confusion of the two words is not always accidental. One question sorts it in any demo: when the work is done, who did it? If the answer involves your team the whole way through, you’re buying a copilot. Price it like one.

Orchestration

What it is. Multiple agents coordinating, handing work to each other, managed by some layer above them. The research agent feeds the drafting agent feeds the review agent. It’s real. It’s also the slide most likely to be describing the vendor’s ambition rather than their product.

Where you’ve already felt it. The diagram with boxes and arrows. It’s in your feed right now: agent-to-agent this, orchestration layer that, entire industries claiming the arrows. You’ve sat in the pitch and felt the pull of it, the aesthetic completeness, the sense that this is what mature looks like.

Worth knowing. For a growth team, orchestration is almost always premature. You don’t orchestrate agents you haven’t built yet, the same way you don’t hire a manager for a team of zero. The pull of the diagram is the old surface area trap in new clothes: acquiring the look of sophistication instead of the substance of one working thing. One task agent, in production, doing real work, beats any diagram. Earn the arrows.

Human in the Loop

What it is. The design choice of where a person reviews, approves, or overrides an agent’s work. The loop is the agent’s cycle. The human is the checkpoint you place inside it.

Where you’ve already felt it. The approval step. The difference between an agent that drafts the email and an agent that sends it, which you understood instantly and viscerally the first time you saw an agent hover near your outbox. Nobody had to explain the stakes to you. Your pulse explained them.

Worth knowing. The phrase often gets written like an apology, a fig leaf for AI that isn’t ready. Read it the other way: it’s the single most important design decision in any agent your team deploys. Not whether a human stays involved. Where. The answer isn’t “everywhere,” which just rebuilds the bottleneck you were escaping. It’s at the moments of consequence: the send, the price, the promise. Put the human where their judgment is worth the most and the agent everywhere else. That sentence is most of agent strategy.

AI SDR

What it is. An agent sold to run outbound prospecting: find the accounts, write the messages, send the sequences, book the meetings. The most heavily funded, most heavily pitched category in this glossary.

Where you’ve already felt it. From both sides. You’ve received the pitch, roughly eleven times this quarter. And you’ve received the output: the email that knew your company’s name and nothing else, the compliment on a post its sender never read. You deleted it in under a second. Your buyers have the same finger.

Worth knowing. Stay calm about this one, because the category invites heat and heat isn’t analysis. The technology sends messages perfectly well. The question it can’t answer is why the recipient should care, and no volume of sending fixes a targeting problem. An AI SDR pointed at a weak segment is the old spray-and-pray with better grammar and worse unit economics, because now the spray scales. If the targeting is right, the sending was never your bottleneck. If the targeting is wrong, automation just gets you to no faster. Solve the who and the why first. The sending is the cheap part. It always was.

Part Four: The Failure Vocabulary

The words for when it goes wrong.

Hallucination

What it is. When a model states something false with complete fluency. Not a bug that slipped through. A structural property of a system built to produce plausible language. Remember the first entry: language, not knowledge. Hallucination is what that trade-off looks like on a bad day.

Where you’ve already felt it. The statistic that was specific, confident, and fabricated. The case study that never happened. The citation to a paper that doesn’t exist, formatted perfectly. What made it unsettling wasn’t the error. It was the confidence. Nothing in the delivery flagged it, because the delivery system has no idea.

Worth knowing. It can be reduced, dramatically: grounding, retrieval, verification steps. It cannot be eliminated, and now you know why, which means you can hear “our AI doesn’t hallucinate” for what it is: a claim the technology can’t back. The honest vendor version is “here’s how we catch them,” and the difference between those two sentences is the difference between a vendor who understands their own product and one who hopes you don’t.

Drift

What it is. The slow degradation of an agent’s performance over time when nobody changed anything. Except something always changed: the model underneath got updated, the data shifted shape, the world the instructions assumed moved on. The agent stood still. Standing still, in a moving system, is a form of change.

Where you’ve already felt it. Week one, the agent was great. Week six, it’s mediocre, and the mediocrity arrived so gradually nobody can date it. Everyone’s first theory is that they imagined the week-one quality. You didn’t imagine it.

Worth knowing. Drift is why an agent is a garden and not a statue. The reasonable-sounding plan (build it, ship it, move on) quietly assumes statue. So the question that matters at deployment isn’t just “does it work.” It’s “who notices when it stops,” because with drift there is no crash, no error message, no alarm. Just a slow slide, and a team that stopped watching in week two.

Guardrails

What it is. Instructions and checks meant to keep an agent inside its lane: don’t discount past ten percent, don’t contact these accounts, don’t promise dates. Some are hard limits enforced by code. Most are strongly worded requests.

Where you’ve already felt it. Every “the AI won’t do that” you’ve heard in a demo, delivered with a confidence the previous three entries should complicate. You’ve also felt the domestic version: the instruction the AI followed for nine responses and forgot on the tenth.

Worth knowing. A guardrail is an instruction, not a guarantee, and the practical question is never whether guardrails exist. It’s what happens when one fails, and how you’d find out. A well-designed system assumes failure and puts a checkpoint after it, which is why this entry and Human in the Loop are the same conversation. Quiet entry, one takeaway: “won’t” usually means “was asked not to.” Price the difference.

Evals (Evaluation)

What it is. The practice of testing an agent’s work against a standard, systematically, before and after it ships. The question you didn’t know existed: how does anyone check this thing? Evals are how. Builders run the agent against known cases and score what comes back, the way you’d check a new hire’s first month.

Where you’ve already felt it. In the negative, mostly. The output that shipped with nobody checking. The agent whose quality everyone assessed by vibe. If you’ve ever asked “how do we know it’s good?” and watched the room go quiet, you were asking for evals. The room just didn’t have the word either.

Worth knowing. This is the last entry because it’s the one that turns the rest into leverage. You now know what can go wrong: hallucination, drift, failed guardrails. Evals are how mature builders catch all three, which makes “how do you evaluate it” the most revealing question you can ask any vendor or any internal team. You’re not asking for the methodology. You’re listening for whether one exists. A real answer has numbers and test cases in it. The other kind of answer has adjectives. And after twenty-seven entries, you know exactly which adjective it’ll be.

Published July 2026. This is a living document. Terms will shift, and when they do, this page updates. If a word showed up in your feed and it isn’t here yet, that’s a signal worth sending me.

There’s a Beck line from 1994 about shaving your face with mace in the dark. It’s been stuck in our heads all week. Sorry if it’s now stuck in your head ( cause you’re a loser, baby, so why don’t you kill me - ahem, sorry ). The whole shaving thing - it’s exactly what most of us are doing with agents right now. Eyes closed. Swinging something dangerous near our own face. Pretty sure it’s fine.

We’ve spent a few posts telling you what to build, what to buy, and what to rebuild for a tenth of what you’re paying. We stand by all of it. But we left a hole in the middle, and it’s a big one.

We never talked about governance.

Neither does anyone else, by the way. Scroll your feed. Everybody’s screaming build, build, build. Ship the agent. Automate the pipeline. Nobody selling you the dream mentions what happens after you turn it on. That part’s not their problem. It’s yours.

We’re not the experts here. We’ll say that plainly. We’re two dudes who build agents to get our own work done. But here’s what we know. And it’s enough to keep you from setting yourself on fire.

The corner you can’t see around

You build the agent. You test it. It works. You feel good.

Then you turn one corner. The corner where it starts running on its own. And everything goes from known to abyss-unknown in about half a second.

No visibility. No observability. Nobody watching. It’s out there now, doing the thing, and you’re trusting it the way you’d trust a new hire you met ten minutes ago and handed the company card.

Doesn’t matter how smart the model is. Doesn’t matter if you’re on Claude or something else. The second it acts without you in the room, you have a governance problem. You just might not know it yet.

What skipping this actually costs

We’re not going to scare you with Fortune 500 horror stories. Those companies have whole teams for this. That’s not your world.

Here’s a smaller one. The kind that happens to people like us.

A friend runs a lean media shop. Two people and him. The whole thing held together with agents he built to kill the grind. Smart guy. Early to all of it. One of his agents deleted a client’s entire operating system. Not a file. The whole instance. Three days rebuilding it by hand. For a client he’d just signed.

The agent didn’t break. It did exactly what it was allowed to do. Nobody ever told it what it wasn’t.

That’s governance. Not a compliance binder. Not a policy department. Just the line between a tool you trust and a liability you haven’t met yet.

And it’s usually quieter than a deletion. One of the sharper people we talked to for this post described the everyday version. An agent hums along at a thousand tokens a month. Then it’s burning five thousand. Nothing exploded. No alarm. But something changed. Maybe its data source moved, and it found a worse one. Maybe it quietly decided to do more than you asked. The token bill was noticed before the human did. A cost spike is often the first sign your agent is off the leash.

You need a constitution

Take one thing from this post. Every agent starts with a constitution and a contract.

Before you build, you tell the thing what it’s for and what counts as it going wrong. Not a legal document. A short, honest list of what scares you. Don’t delete records. Don’t spend money. Don’t touch anything outside this one system. If you’re not sure, stop and ask.

You won’t think of everything. Fine. The point isn’t to predict every failure. The point is to make the agent hold the question instead of you holding your breath. Every dumb thing you’ve already done becomes an anti-pattern you write into the next one. That’s the discipline. You’re not designing for what you want it to do. You’re designing for what you never want it to do.

A handful of habits make that real. None of them requires you to be an engineer.

Give it the least power that still gets the job done. Broad read, narrow write. If it just hands you a draft, it does not need permission to write to your CRM. Most people grant way more than the task needs and never think about it again.

Keep the rules in the code, not the model. A prompt that says “please don’t” is not a control. It’s a suggestion. The real guardrails live in the container and the scoped tools. The stuff the model can’t talk itself out of.

Credential it like a person. Scoped access. Token caps. API limits. Give it what you’d give a brand-new hire on day one. Not one key more.

Make it tell you what it’s doing, in plain English, as it goes. If you can’t read what it did, you can’t catch what it did wrong.

Watch the bill. The cost spike is the smoke alarm. Wire an alert. Set a cap. Cheapest early-warning system you’ve got.

Match the armor to the danger

You do not need any of this for an agent that reads transcripts and hands you a summary. Read-only and harmless needs almost no governance at all.

But an agent with write access to your CRM and your inbox, running on its own, deciding who gets contacted and what gets changed? That one needs all of it. The skill isn’t bolting maximum governance onto everything. It’s knowing what kind of thing you just built.

Whose job is this really?

Here’s where we land. Who’s most responsible right now for sloppy code doing sloppy things in the wild? Honestly? The platforms. They handed a few million of us a brilliant brain and the keys to everything, and not much in the way of brakes. Someday they will build the guardrails at the source, and a lot of this will get easier.

That day is not today. Until it comes, the trust is your job. Govern what you build. Because a thing you can’t govern is a thing you shouldn’t have built, no matter how cheap it was to make.

We’re still standing at the edge of this one ourselves. Next time, we bring in someone who builds governance for a living to show you the how, not just the why. We’ll do the part we’re good at, making sense of the map. They’ll handle the terrain.

For now, go look at what you’ve already built. Ask it what it’s allowed to do.

You might not love the answer.

New to Cannonball GTM? We help growth leaders find hidden buyers in publicly available data, and figure out which tools, agents, and platforms are actually worth the money. Start here:

• The Agent Landscape for Growth Leaders — the rubric and the table. Fifteen use cases mapped against build it yourself, build it on a platform, or buy it.

• We Built an Agent for Finding Hidden Customers. Here’s Yours. — a working agent you can run yourself, free.

• The Cannonball GTM Two-Week Test Protocol — how to test before you scale, so you learn instead of burning budget.

Real estate has more public data than almost any other industry in America.

(And so does every competing vendor in your category.)

Assessor records, deed histories, mortgage amounts, equity positions, foreclosure notices… that stack of data has been packaged and sold for twenty-some years. Pulling property records to find motivated sellers is a starting point, not a strategy.

What those records don’t have: the specific dollar penalty a building owner will owe starting in 2030 if they do not retrofit. Or the maturity date on the CMBS loan at that same address. Or the open code violations accumulating across their portfolio that remain unresolved. Or even the eviction filing patterns, indicating that their multifamily operation has been under strain since the last quarter. None of that is in the property record.

But, every one of those sources is public. Most are free. And almost no one selling into commercial real estate is using them to find the hidden customers.

Get out the KitchenAid. Let’s mix things up.

The Inventory

We have ranked these public data sources by how much of an edge they give you over every other vendor fighting for attention in your prospect’s inbox.

County Assessor and Recorder Data (attomdata.com, regrid.com, corelogic.com, propstream.com)

Every county records property transfers, deeds, mortgages, and liens. Every assessor maintains ownership, characteristics, and assessed value. Aggregators like ATTOM, CoreLogic, Regrid, and PropStream cover roughly 155 to 160 million properties at about 99 percent U.S. population coverage, refreshed daily via API or bulk:

• Owner name and mailing address

• Sale history and price

• Mortgage lender, amount, origination, and recording date

• Liens and transfers

• Property characteristics and assessed value

• Persistent property identifier for joining to other sources

Treat this as the spine, not the edge. Every source in this inventory joins to it. The residential motivated-seller motion built on this data is one of the most saturated GTM plays anywhere. Arrive with something attached to the join key that no competitor thought to cross.

ATTOM Property and Mortgage Data · Regrid · CoreLogic · PropStream

Municipal and State Building Energy and Emissions Disclosure Data

A growing set of city and state laws requires large buildings to publicly benchmark energy and water use and, increasingly, to meet hard emissions caps with penalties attached. New York City is the model: Local Law 84 requires annual benchmarking, Local Law 97 sets emissions caps at $268 per ton of CO2 over the limit, and Local Law 33 assigns public energy grades from A to F. Submissions go through the EPA’s Energy Star Portfolio Manager, and the city publishes the results. Roughly 57% of covered buildings are projected to exceed the 2030 to 2034 caps. These data sources include:

• Building address and covered-buildings-list status

• Benchmarked energy use intensity (EUI)

• Water use

• Emissions versus the applicable cap

• Projected annual penalty in dollars

• Public energy grade (A through F)

Energy efficiency, retrofit, decarbonization, HVAC, ESG compliance, and capital planning vendors all sell from this source. The entry point is the projected penalty: a specific dollar figure on a specific building, years before the deadline, pulled from a filing the owner submitted themselves.

NYC Benchmarking and LL97 · EPA Energy Star Portfolio Manager · LL97 Accelerator

Foreclosure and Default Records (attomdata.com/data/foreclosure-data)

Notices of default, lis pendens (pending lawsuits), notices of trustee’s sale, and bank-owned records. Each stage is dated and retrievable. For commercial and operator targets, distress is a specific, timed signal that stacks well with maintenance and violation data from the same property. Fields include:

• Notice type: NOD, lis pendens, trustee’s sale, or REO

• Filing date and auction date

• Opening bid

• Foreclosing lender and servicer

Workout and restructuring firms, opportunistic acquirers, property services, and capital advisory all sell from this source. The residential motivated-seller version of this motion is the saturated lane; aim it at commercial properties and operator portfolios.

ATTOM Foreclosure Data

CMBS and Agency Loan Data (sec.gov/search-filings, trepp.com)

Most commercial loans are private. Commercial mortgage-backed securities (CMBS) and agency loans (Fannie Mae, Freddie Mac, HUD) are not. Loan amounts, origination and maturity dates, performance status, and servicer watchlist flags are all trackable for the securitized slice of the market through SEC EDGAR asset-backed securities filings. Deeper performance data runs through Trepp and Moody’s Analytics. Data sets include:

• Loan amount, origination date, and maturity date

• Property address and type

• Performance and delinquency status

• Servicer watchlist flags

• Securitization deal identifier

CRE debt and lending technology, restructuring and advisory, investment platforms, and brokers with access to capital all sell from this source. The honest limit: bank and life-insurance-company commercial loans are private, so this covers only the CMBS and agency slices.

SEC EDGAR ABS Filings · Trepp

Building Permits (shovels.ai)

Renovation and repositioning activity at the asset level. A major permit for a specific building signals an ownership decision about what that asset will become, made before any broker has the listing and before the leasing sign goes up. Covered in depth in the construction guide, accessed at scale via Shovels:

• Permit type and scope

• Valuation

• Contractor

• Filing and issue dates

• Address and parcel identifier

Construction and project management, FF&E and procurement, property management software, and leasing and marketing platforms all sell from this source. The window is between permit issuance and project completion, when downstream vendor decisions are being made.

Shovels

Planning, Zoning, and Entitlement Filings (shovels.ai/decisions)

Pre-development approvals, rezonings, and planning-commission discussions. The earliest public leading signal of new development is often months before a permit and sometimes years before groundbreaking. Fragmented at the city level, partially aggregated through Shovels Decisions:

• Planning application type and status

• Rezoning and variance requests

• Project description and address

• Commission hearing dates

Developers, land brokers, construction-phase vendors, and anyone selling into the pipeline before a permit is filed all benefit from this source. The window here is the widest in the vertical and the most forward-looking.

Shovels Decisions

Code Violations and Housing-Code Enforcement

City building and housing-code violation records: open violations, complaints, and orders to repair. Available through cities with open-data portals, like New York City’s HPD and DOB records, are the most accessible and field-rich examples. Data sets will have:

• Property address and violation type

• Class and severity

• Complaint and issue dates

• Status: open or resolved

• Issuing agency

Property management software, maintenance and work-order platforms, and compliance tools all sell from this source. A portfolio with a pile of open violations is broadcasting operational pain in a public record. Almost no one selling into property management is reading it.

NYC Open Data · Check Sunlight Foundation’s list of Code Enforcement Violations data sets by city/municipality

Eviction and Housing-Court Filings

Court records of eviction filings and outcomes. Turnover and management strain signal for multifamily operators. The data is public but fragmented, court by court, and the assembly cost is the barrier, not the access.

• Filing date and case type

• Property address

• Outcome and timeline

Multifamily property management platforms, tenant-experience products, and collections tools all sell from this source. Pair it with code violations on the same portfolio, and the picture of operational strain becomes very specific.

Search eviction and housing-court filings + the city.

HUD Multifamily and LIHTC Data (huduser.gov/portal/datasets/lihtc)

HUD-insured and assisted property data plus the Low-Income Housing Tax Credit database: LIHTC properties, unit counts, placed-in-service dates, and compliance period expiration dates. Free and public from HUD:

• Property and unit count

• Placed-in-service date

• LIHTC compliance period and expiration

• HUD assistance type and contract details

For the affordable-housing segment, the expiration of the compliance period is a precise temporal trigger. A LIHTC property approaching the end of its compliance window faces a defined set of decisions about recapitalization, repositioning, or disposition, and the date of that decision is publicly available in a database.

HUD LIHTC Database

SEC EDGAR (REITs) and MSRB EMMA (sec.gov, emma.msrb.org)

Public REITs disclose strategy, portfolio composition, and risk on EDGAR. Municipal bond and bond-financed real estate projects file on EMMA. Free and public. Rich for large institutional owners and for securitized project debt. Data points include:

• REIT portfolio holdings, strategy, and risk disclosures

• Lease expiration schedules and debt maturity profiles

• Bond-financed project details and covenant conditions

Institutional sales teams, advisory firms, and capital markets vendors that sell to public owners or to bond-financed projects use this source. Smaller and more specific than the rest of the inventory, but the information is exact.

SEC EDGAR · MSRB EMMA

Job Postings

A surge in property management, leasing, asset management, or compliance hiring is an operational strain in motion. Best used as a confirmation layer on a structured signal. Pair it with a permit, and the repositioning timeline sharpens; pair it with code violations, and the management capacity picture becomes concrete.

• Job title and department

• Posting date and location

• Urgency indicators in the listing text

Recruiting and HR platforms, operational software vendors, and anyone selling into the transition window between operating teams all read this signal. Its value is in the combination, not in isolation.

Check job sites like Indeed and LinkedIn Jobs.

The Real Estate Data Cookbook

Here are five recipes ranked by the asymmetry they create. Each one combines public sources into a specific dish: a pain-based segment plus a message angle no competitor can match, because none of your competitors has combined the same ingredients.

(Note - the messages are purely samples meant to give you an idea of what’s possible at the most basic level and are not up to PVP/PQS standards.)

Recipe 1: The Emissions Cliff

The Dish: A commercial building owner facing a dated, dollar-quantified emissions penalty before a fixed compliance deadline.

Ingredients: Building energy benchmarking and LL97-style emissions data (a building projected to breach the 2030 cap, with a calculable annual penalty) + assessor record (owner name and mailing address) + public energy grade (optional garnish).

Read more

There are 525k+ active interstate motor carriers in the United States, and every one of them has a USDOT number and an MC number filed with the federal government. Every freight broker knows this. Every vendor selling into trucking knows this.

It is also the most comprehensive directory of people in logistics and transportation, and they are being spammed to death.

That is the crowded lane. This article is about the other lane.

The Federal Motor Carrier Safety Administration (FMCSA), the same agency that publishes the carrier list, also publishes monthly safety scores, crash records, out-of-service rates, and insurance filing histories for those same half-million carriers. Very few selling into this vertical are watching those records for buying signals. Even fewer are tracking when a carrier’s safety score starts sliding, or when their insurance lapses, or when a hazmat incident report drops with their name on it. The directory is picked clean. The distress data is ripe for the picking.

What follows is a ranked inventory of public data sources for logistics and transportation, plus five recipes for combining them into outbound that your competitors cannot replicate.

Grab the bushel basket.

The Inventory

We’ve ranked these public data sources by how much of an edge they give you over every other vendor fighting for attention in your prospect’s inbox.

FMCSA SAFER, SMS, and the Monthly Safety Datasets (safer.fmcsa.dot.gov)

This is the crown jewel of the vertical, and it is goes almost entirely unused for GTM. The Federal Motor Carrier Safety Administration maintains a safety and performance record for every interstate carrier, updated monthly, available in bulk, for free. The Company Snapshot gives you a quick per-carrier view; the Safety Measurement System organizes it into the BASICs, which are percentile-ranked safety categories that function like a carrier’s credit score. What you get per carrier:

• USDOT number, legal name, DBA, address, phone, and contact information

• Power units, drivers, and annual mileage

• Operation type, commodity, and hazmat flag

• Official safety rating

• Out-of-service rate (vehicles and drivers)

• Crash count and crash rate

• BASIC percentile rankings: Unsafe Driving, Hours-of-Service Compliance, Driver Fitness, Controlled Substances and Alcohol, Vehicle Maintenance, Hazardous Materials Compliance

One caveat worth understanding: the Crash Indicator and Hazmat Compliance BASICs are hidden from public display for property carriers under the 2015 FAST Act. The rest are public. Build your recipes on the visible BASICs.

A carrier with a deteriorating Unsafe Driving percentile and a rising out-of-service rate is not a static data point. That is a quantitative pain threshold moving in real time, the trucking equivalent of a bank slipping below its capital ratio. Safety software vendors, telematics companies, ELD providers, and insurance underwriters should be watching this monthly, not querying the list once and calling it research.

FMCSA SAFER Company Snapshot · Safety Measurement System · SMS bulk downloads · FMCSA data dissemination program

FMCSA Licensing and Insurance (li-public.fmcsa.dot.gov)

Every carrier, broker, and freight forwarder with active operating authority has an insurance filing on record with the FMCSA. When that insurance lapses (or when an authority is at risk) it shows up here. The Carrier All With History dataset includes authority type and status, insurance type and amount, and policy cancellation method and effective date. Fields include:

• USDOT and MC docket number

• Authority type and current status (active, pending, revoked)

• Insurance type and coverage amount on file

• Policy cancellation method and cancellation date

An insurance lapse is an existential event for a motor carrier. You cannot legally operate without it. The moment that cancellation is filed, it is public record. That is the sharpest distress signal in this entire vertical, and goes essentially untouched as a GTM trigger.

FMCSA Licensing and Insurance · Carrier All With History dataset

Import Bills of Lading in US Customs Data (importyeti.com)

Every sea shipment into the United States generates a bill of lading, and customs records are public. ImportYeti has compiled roughly 70 million of them. Panjiva and similar vendors sell deeper access and historical queries. What you get per record:

• Importer (purchaser) name

• Supplier name and country

• Origin and destination ports

• Vessel and shipment date

• Weight

• Product HS code

The bill-of-lading directory is the most GTM-mature source in this vertical. Freight brokers have been mining it for years. The edge is not in the list of importers; it is in the deltas. A company whose import volume has grown 40% in six months and is opening a new port of entry is not the same prospect as a company whose import volume has been flat. The former is about to need more freight capacity, a new customs broker, and probably a TMS upgrade. You see that before they put it out to bid.

ImportYeti · Panjiva

PHMSA Hazmat Incident Data (phmsa.dot.gov)

The Pipeline and Hazardous Materials Safety Administration publishes hazmat incident reports and pipeline incident, inspection, and enforcement data. This free downloadable data set includes:

• Hazmat substance and quantity involved

• Transportation mode (highway, rail, air, pipeline)

• Incident consequences (injuries, fatalities, evacuations, property damage)

• Date and location

• Carrier or operator name

For hazmat carriers, pipeline operators, and the compliance, safety, and specialty insurance vendors who serve them, a recent incident is specific, dated evidence of a problem. That is the whole game.

PHMSA hazmat incident statistics · Pipeline enforcement data

OSHA Inspections and Citations for Warehouse and Distribution (osha.gov/data)

Warehousing has become one of the highest-injury segments in American industry. OSHA’s inspection and citation database is searchable by establishment, industry code, and date. A warehouse or distribution center that has been cited (especially for serious or willful violations) is a quantified safety problem with a compliance clock running. This data set includes:

• Establishment name and address

• Inspection date and type

• Citation type and standard cited

• Penalty amounts

• Abatement dates

Safety software, ergonomics, training, and insurance vendors selling into warehousing and distribution should treat the OSHA dataset the way construction vendors treat building permits: not as background research, but as a trigger list.

OSHA establishment search · Injury and illness data · Severe injury reports

Bureau of Transportation Statistics (bts.gov)

Macro freight flow, modal data, border-crossing volumes, and on-time performance. Free, public. Creates little asymmetry on its own — its job is timing and context. Which lanes are growing, which crossings are heating up, which modes are taking share. Use it to validate timing, not to build target lists.

Bureau of Transportation Statistics

FAA Registry and Service Difficulty Reports (registry.faa.gov)

Aircraft registration and reported mechanical issues. For logistics vendors selling into air cargo or aviation operators, it identifies the fleet and flags recurring problems. Narrow audience, but precise when it applies.

FAA aircraft registry · FAA service difficulty reporting system

State DOT Oversize and Overweight Permits

Permits to move heavy or specialized loads on state roads. Usually public, varies by state. Reveals specialized and heavy-haul movement — useful for carriers and vendors in the overdimensional, equipment transport, or wind energy logistics niches. Access varies; start with your target states’ DOT websites.

UCC Equipment Liens for Tractors, Trailers, Reefers (nass.org/business-services/ucc-filings)

When a carrier finances equipment, the lender files a UCC financing statement. Free to search at state Secretary of State portals. What you get:

• Debtor name (the carrier)

• Secured party (the lender)

• Collateral description (the equipment — often specific make, model, and VIN)

• Filing date and lapse date

A carrier filing a new lien on three tractors and a refrigerated trailer is adding capacity. That is a clean, dated leading indicator of fleet expansion, plus it pairs naturally with the FMCSA Census record, which shows the same trend in its power-unit count.

NASS UCC Filings Portal

SEC EDGAR (sec.gov/search-filings)

Public trucking, rail, parcel, and third-party logistics companies disclose strategy, risk factors, and segment results. Only covers the public slice, but rich for large named operators. Free, full-text searchable.

SEC EDGAR full-text search · EDGAR filings search

Job Postings for Drivers, Dispatchers, Safety and Compliance

Carriers and warehouses broadcast capacity strain in their hiring. A surge in driver, dispatcher, or safety-manager requisitions is pain in motion. High-asymmetry unstructured source; best used as confirmation layered on top of a structured signal from FMCSA or UCC data, not as a standalone trigger.

Job listings on Indeed or LinkedIn Jobs.

The Logistics and Transportation Industries Cookbook

The inventory tells you what exists. These recipes tell you what to do with it. Every recipe combines two or three sources into a signal that is specific, dated, and actionable. The kind of outbound that makes a prospect ask how you knew.

(Note - the messages are purely samples meant to give you an idea of what’s possible at the most basic level and are not up to PVP/PQS standards.)

Recipe 1: The Safety Spiral

The Dish: A carrier whose safety scores are worsening, heading toward federal intervention, and rising insurance costs.

Ingredients: FMCSA SMS filtered for deteriorating BASIC percentiles (especially Unsafe Driving or Hours-of-Service) and a rising out-of-service rate + FMCSA crash data showing recent incidents + job postings with an active search for a safety manager or compliance director.

Indicator Type: Trailing on the violations. Leading on the consequences.

Target Segment: Safety and compliance software, ELD and telematics, driver training, insurance.

The Asymmetry: When a carrier’s Unsafe Driving BASIC climbs into the upper quartile, their out-of-service rate is trending up, and they are simultaneously posting for a safety director, you know their exact compliance trend and the intervention threshold they are approaching before your competitors who didn’t go beyond just the carrier list. You can name the specific BASIC, the trend direction, and the timeline. No one else calling that carrier can do that.

Read more

Financial services is one of the most heavily regulated industries in the United States, so it makes sense that it’s the most data-rich regulated source as well.

Every bank and credit union in the United States files detailed quarterly financials publicly. Every registered investment adviser discloses its assets, clients, and disciplinary history publicly. Every consumer complaint against every financial company is published daily. Every enforcement action lands on the public record the day it is issued.

Data is not hidden or locked up or fragmented. It’s sitting out on federal agency sites with free APIs, bulk downloads, often with no authentication required.

Which is why it’s kinda insane when GTM teams completely ignore all. The standard buy list, filter job title, add intent data, run ABM, somehow survives. Someone says, “Hey, you guys! A bank might be in-market for compliance software! Let’s sequence them!”

NARRATOR: THE BANK WAS NOT IN-MARKET FOR COMPLIANCE SOFTWARE

The Inventory

We’ve ranked these public data sources by how much of an edge they give you over other vendors competing for the same budget in the inboxes of finserv leaders.

Bank Call Reports and UBPR (cdr.ffiec.gov, banks.data.fdic.gov)

Every national bank, state member bank, and insured nonmember bank files a Report of Condition and Income called the Call Report every quarter. The Uniform Bank Performance Report turns that into performance ratios with peer comparisons. This is the financial X-ray of every bank in America, filed publicly and quarterly, going all the way back to 1992 via the FDIC BankFind Suite’s free API. Data include:

• Balance sheet and income statements by quarter

• Loan portfolio by category and delinquency bucket

• Capital ratios, net interest margin, and efficiency ratio

• Deposit composition and funding mix

• Branch-level deposits via the annual FDIC Summary of Deposits (openings, closings, deposit flows)

The pain is in the numbers. The numbers are public. You just have to go get them:

FFIEC CDR · FDIC BankFind Suite · API docs · Bulk downloads

Credit Union 5300 Call Reports, FPR, and Profile (ncua.gov)

The credit union parallel to the bank Call Report, and in one way, it goes further: the Profile (Form 4501A) hands you the non-financial layer for free. Branch locations, member services, and the names of senior management and volunteer officials, all in the same quarterly package.

That combination of a distress signal and a named executive contact in a single public source is rare. In almost every other data source in this vertical, you get the signal or the contact. The NCUA gives you both. Data include:

• Net worth ratio, delinquency, ROA, and loan and share growth

• Membership trajectory

• Named senior management and volunteer officials

• Branch locations and member service offerings

NCUA Credit Union Call Report Data

Regulatory Enforcement Actions and Consent Orders

When a regulator finds a problem, it issues a public enforcement action. A consent order is not a soft signal. It is a legal mandate to remediate a specific problem by a specific deadline, and it is public the day it lands.

The OCC, Federal Reserve, FDIC, NCUA, and CFPB all maintain searchable public databases of their actions. This is the strongest forced-buying trigger in the entire vertical. An institution subject to a BSA/AML consent order must fix the problem. The question is not whether they are buying. It is who reaches them first.

That said, it is a trailing indicator; The pain has already materialized. Which means you do not have to convince anyone they have a problem. You just have to prove you already know what it is. Data per action:

• Institution name and regulator

• Action type (consent order, civil money penalty, cease-and-desist)

• The conduct cited, in plain language

• Dates, penalty amounts, and remediation terms

OCC enforcement actions · FDIC enforcement decisions · NCUA enforcement actions · CFPB enforcement actions

CFPB Consumer Complaint Database (consumerfinance.gov)

A live, public record of consumer complaints against financial companies. Over 13.8M complaints since 2011, updated daily. When a consumer submits a complaint, it is published after the company responds or after 15 days, whichever comes first. The consumer can include their own narrative. That narrative is public.

This is the closest thing to real-time consumer pain in any regulated industry. A spike in complaints about one product at one institution is a dated, quotable, public pain signal. GTM teams should be watching it.

A note for 2026: the CFPB’s enforcement posture has shifted under current leadership, which can affect how actively the database is maintained and how quickly complaints move through the pipeline. The data is public so treat its cadence as a variable, not a constant. Complaint details include:

• Date received

• Company name

• Product and sub-product

• Issue and sub-issue

• Consumer narrative text

• The company’s public response

• State and ZIP code

• Timely-response flag and disputed flag

Complaint database · Public API (no key required)

HMDA Loan-Level Mortgage Data (ffiec.cfpb.gov)

Under the Home Mortgage Disclosure Act, roughly 5,000 financial institutions disclose loan-level mortgage data annually in their Loan Application Registers. In practice, this is every reporting lender’s entire mortgage book, origins, denials, withdrawals, amounts, purposes, and applicant characteristics, filed publicly by March 2 of each year. A lender whose origination volume collapsed year over year, or whose denial rate spiked, is showing distress in their own public filing. You do not have to guess. You can read it. Per loan:

• Loan type and purpose

• Loan amount

• Action taken (originated, denied, withdrawn, incomplete)

• Pre-approval status

• Property type and lien status

• Applicant demographics

• The reporting lender

HMDA Data Browser · Downloads and documentation

SEC Form ADV / IAPD (adviserinfo.sec.gov)

Every SEC-registered and state-registered investment adviser files Form ADV through the Investment Adviser Public Disclosure site. Part 1 carries the structured business data. Part 2 is the plain-language brochure the firm is required to hand clients. Both are public.

Year-over-year deltas are where this source earns its place: an AUM jump, a new custodian, a shift in client type, a fresh disciplinary disclosure. Change is the signal. Per firm:

• Assets under management

• Number and type of clients

• Employee count

• Custody arrangements and fee structure

• Ownership and control persons

• Disciplinary disclosures

IAPD search · Bulk Form ADV data files

FINRA BrokerCheck / CRD (brokercheck.finra.org)

The broker-dealer is parallel to IAPD. BrokerCheck, built on FINRA’s Central Registration Depository, covers brokerage firms and registered representatives with registration history, disclosures, and disciplinary record. Many professionals are dually registered, so BrokerCheck and IAPD are read together rather than separately.

Registration churn and disciplinary disclosures at a firm are pain and change signals for compliance, supervision, and wealthtech offerings. Most useful when velocity is the signal: not one disclosure, but a pattern of them over a twelve-month window.

BrokerCheck

Merger Applications and Holding-Company Filings (Federal Reserve / FDIC / NCUA)

Bank and credit union mergers require public applications to regulators before they close. Bank holding companies file the FR Y-9C with the Federal Reserve. Both reveal structural change before it is complete (and before any RFP is written).

A pending merger means a core conversion, a system integration, and a compliance harmonization are all coming, on a known timeline. That is a bundle of fresh, datable pressure. And unlike most triggers, it comes with a public document describing exactly what is about to happen. Per merger application:

• Acquirer and target

• Transaction structure

• Regulatory approval status and timeline

Fed National Information Center · FDIC merger applications

State Data Breach Notifications

Most states require breach notifications to be filed with the state attorney general when a company suffers a qualifying incident, and several also require publication. Financial firms are among the most frequent filers.

A breach disclosure is a dated, public, high-urgency pain signal for anyone selling security, compliance, incident response, or cyber insurance to financial institutions. Format and availability vary by state. Some states, like California, publish a searchable list; others require records requests. Mark as verify-by-state before building a workflow around it.

Search for data breach notifications + state.

Example: California AG breach list

NMLS Consumer Access (nmlsconsumeraccess.org)

The Nationwide Multistate Licensing System publishes mortgage company and loan-originator licensing through its Consumer Access site. License classification, status, and branch locations are all visible.

On its own, it is a denominator, not a trigger. Many licensed lenders have not originated a loan in months. Use it to validate and classify. Do not use it to trigger. Best paired with HMDA data to separate active lenders from dormant licensees.

NMLS Consumer Access

NAIC and State Insurance Filings and Complaint Data

Insurers file statutory financial statements with state regulators and the NAIC, and states run market-conduct examinations and publish consumer complaint data through the NAIC complaint index.

For insurtech, a complaint-index outlier or a market-conduct finding is a real pain signal. The data is harder to assemble uniformly than the banking equivalents (availability, format, and refresh rate vary by state and line of business) which is why it sits lower on the list. Verify coverage before promising it.

NAIC Consumer Insurance Search

FinServ Firm Job Postings

Banks, credit unions, and fintechs broadcast their pressure in their hiring. A burst of BSA/AML analyst, fraud, compliance officer, or risk management reqs is pain in motion. The job descriptions often specify the system they use, the gap they aim to fill, and the volume they manage.

This is the highest-asymmetry unstructured source in the vertical and a strong leading indicator. Most valuable when it confirms a structured signal: an enforcement action plus a sudden BSA-officer search is a richer combination than either alone.

Indeed and LinkedIn Jobs

SEC EDGAR (sec.gov)

Public banks, insurers, fintechs, and payment companies disclose strategy, risk factors, capital plans, and segment results in 10-Ks, 10-Qs, and 8-Ks. Only the public slice of the market files, and Call Reports already cover banking financials in greater detail. Most useful for the large public names and for risk-factor language, you can quote back in an opening line.

Free, full-text searchable, API-accessible.

SEC EDGAR

The Financial Services Data Cookbook

In financial services, the asymmetry is in plain sight, and almost nobody selling into the industry is using it.

Here are five recipes, ranked by the edge they create. Each one combines public sources into a specific dish: a pain-based segment plus a message angle no competitor can match, because no competitor has crossed the same data.

(Note - the messages are purely samples meant to give you an idea of what’s possible at the most basic level and are not up to PVP/PQS standards.)

Apron on. Let’s cook.

Recipe 1: The Consent Order Clock

The Dish: An institution under a fresh regulatory order with a remediation deadline, scrambling to comply.

Ingredients: A regulatory enforcement action or consent order (the conduct, the deadline, the penalty) + the institution’s Call Report or 5300 (its size and the specific deficient area) + job postings (a sudden BSA officer or compliance-staff hiring burst).

Indicator Type: Trailing on the order, leading on the remediation.

Target Segment: AML/BSA and compliance regtech, risk consulting, audit and monitoring software, managed services for remediation.

The Asymmetry: You know the exact order, the deadline, and the conduct cited so you can see they are already trying to hire their way out of it (and before any vendor in your category has picked up the phone. The institution is not “evaluating options.” It is legally required to fix a named problem by a named date.

The Message: “I saw the OCC consent order from [date] and the BSA/AML findings and the [X]-month remediation window. I also noticed you’ve posted three BSA analyst roles in the last six weeks. We work with institutions in exactly this situation. Worth a conversation?”

Recipe 2: The Complaint Spike

The Dish: An institution with a public, escalating consumer-pain problem in one specific product line.

Read more

Every healthcare GTM team touts its list of target companies and NPI numbers - the National Provider Identifier, which means the provider exists and is billing Medicare. It’s the healthcare industry’s version of a business license.

But a list of NPIs is not an insight by itself. It’s just the starting point.

What the list does not include: the operating margin for every hospital on it, from each hospital’s own filings. The specific survey deficiencies at every nursing home, the F-tag, the severity level, and the remediation deadline. The breach disclosed to HHS last month: the type, scale, and date. The Certificate of Need filing one of your prospects submitted six months ago for a service line that hasn’t been built yet.

Healthcare is one of the most data-rich industries in America, and almost nobody selling into it is actually reading the data.

But you will.

The Inventory

We’ve ranked these public data sources by how much of an edge they give you over every other vendor fighting for attention in your prospect’s inbox.

Medicare Cost Reports (cms.gov/cost-reports)

Every Medicare-certified institutional provider files an annual cost report with CMS, the Centers for Medicare & Medicaid Services which is the federal agency in the Department of Health and Human Services that administers critical public health programs, including Medicare, Medicaid, the Children’s Health Insurance Program (CHIP), and the Health Insurance markets of the Affordable Care Act. This data source is the financial X-ray of each provider and it’s wild that GTM teams are still ignoring it:

• Operating margin and net income

• Cost per discharge

• Payer mix: Medicare %, Medicaid %, commercial %, self-pay %

• Bed size, occupancy rate, admissions, patient days

• Ownership type, teaching status, urban/rural designation

The data runs 12–18 months old, but a hospital with a declining margin across three consecutive reports is a documented trend you can name before the first call. Revenue-cycle vendors, cost-reduction platforms, advisory firms: anyone whose pitch improves a margin now has a number to put on it.

CMS cost reports · HCRIS data downloads

CMS Care Compare and the Provider Data Catalog (medicare.gov/care-compare)

Quality, safety, and survey data for every Medicare-certified hospital, nursing home, home health agency, and dialysis facility. For nursing homes specifically, you get the exact deficiency findings from every survey inspection: the F-tag, the scope and severity level, the correction deadline. That is healthcare’s OSHA equivalent, a named legal obligation, on a clock, in public record.

• Star ratings: overall, health inspection, staffing, quality measures

• F-tag citations: specific deficiency, scope/severity (A–L), correction deadline

• CMS-2567 findings: surveyor narrative describing what was found

• Staffing: RN hours per resident day, total nurse hours

• Quality measures: falls, pressure ulcers, hospitalizations, antipsychotic use

Compliance platforms, clinical documentation software, staffing vendors, and EHR companies: the entry point is the specific item cited.

CMS Care Compare · Provider Data Catalog

HHS OCR Breach Portal “the Wall of Shame” (ocrportal.hhs.gov/ocr/breach)

Every reported breach of protected health information affecting 500 or more individuals, posted publicly within about two weeks of HHS receipt. Roughly 700 large breaches per year. The breach is specific, dated, and named:

• Covered entity name and state

• Business associate involved: yes/no

• Individuals affected

• Date of breach incident and date reported to HHS

• Breach type: hacking/IT incident, unauthorized access, theft, loss, improper disposal

• Location of breached PHI: network server, email, EMR, portable device, paper

Most of the vendor field hasn’t noticed the posting yet. Nobody has had time to run a procurement process. The buying window is open. Cybersecurity vendors, HIPAA and GRC platforms, incident-response firms, cyber insurers: if your product prevents breaches or manages their aftermath, this is your trigger.

HHS OCR Breach Portal

CMS Open Payments/Sunshine Act (cms.gov/open-payments)

Every payment from a drug or device manufacturer to a physician or teaching hospital is published annually. Before you call on a physician, you can see whether they’re already on a competitor’s speaker bureau. Before you approach a teaching hospital for a research partnership, you can see which manufacturers are already embedded:

• Paying the manufacturer and product

• Recipient physician or teaching hospital

• Amount, date, and nature of payment: consulting, speaking, travel, research, royalty, meals

For device and pharma commercial teams, this maps the competitive landscape before the first call. Cross-reference with OIG exclusions or Care Compare deficiencies, and it becomes a compliance exposure signal as well.

CMS Open Payments data explorer

Medicare Provider Utilization and Payment Data (data.cms.gov)

Per-provider, per-procedure data on what every Medicare provider actually does at what volume: not what they say they do, what they bill. Joint replacements, oncology infusions, cardiac catheterizations, all of it keyed to the NPI, by year.

• Provider NPI, specialty, location

• Procedure code (HCPCS/CPT) and description

• Beneficiary and service counts

• Submitted charges and Medicare payment per service

• Place of service (facility or non-facility)

Capital equipment sellers, service-line growth tools, anyone whose pitch depends on knowing a provider’s actual procedure mix.

CMS provider utilization and payment data

Hospital Price Transparency Machine-Readable Files (cms.gov/price-transparency)

Since 2021, hospitals must publicly post every standard charge available both in machine readable formats and for consumers. About 91% of hospitals were posted by the end of 2023. Enforcement tightened in April 2026:

• Item or service description and code

• Service setting: inpatient or outpatient

• Gross charge

• Payer-specific negotiated rate by payer name and plan name

• De-identified minimum and maximum negotiated rates

• Discounted cash price

There are two commercial angles for two different products.

1. Rate intelligence: A hospital’s negotiated rate by payer and plan is now a public record, and a payer-contracting tool that benchmarks those rates against regional peers opens a CFO conversation with the prospect’s own numbers.

2. Compliance exposure: A missing or non-conforming file is a civil monetary penalty waiting to happen, and the enforcement clock just got shorter. Payer-contracting tools own the first angle. Price-transparency compliance vendors own the second.

CMS price transparency rule · MRF schema on GitHub

openFDA (open.fda.gov)

The FDA’s open data: device adverse events (MAUDE), recalls classified I through III by severity, inspection outcomes (Form 483s and Warning Letters), and device registration. All API-accessible and downloadable.

• Adverse events: device or drug involved, event type, patient outcomes, reporter type, date

• Recalls: recalling firm, product description, recall classification (I/II/III), reason, distribution scope

• Inspections: facility, district, classification (Official Action Indicated, Voluntary Action Indicated, No Action Indicated)

• Registration: establishment name, registration number, operation type

A Class I recall is the FDA’s highest severity classification. A Warning Letter means significant violations were found. Both create remediation obligations with known dates. For device, diagnostics, and pharma-adjacent sellers, this is your consent order: a specific, dated, public forced-buying trigger. For medical device sellers, quality management platforms, and post-market surveillance tools, the recall is your entry point.

openFDA · MAUDE adverse events · Recalls database

NPPES NPI Registry (npiregistry.cms.hhs.gov)

The national registry of every provider and organization with a National Provider Identifier. On its own, just a denominator.

• NPI number (Type 1: individual provider; Type 2: organization)

• Provider name and credentials

• Taxonomy codes: specialty and subspecialty classification

• Primary practice address, phone, fax

• Authorized official name and title (for organizations)

• Enumeration date and deactivation date

NPPES NPI Registry · Bulk download

State Licensing and Certificate of Need Filings

A Certificate of Need (CON) filing is one of the few true leading indicators in healthcare: it announces an expansion before the building is built, before the RFP goes out, before any vendor has been called. In the roughly 35 states that maintain CON programs, providers must file a public application before adding beds, building facilities, or launching major new service lines:

• Proposed project type and service line

• Scope, projected volumes, and cost

• Need justification

• Approval/denial status and conditions

Coverage and portal quality vary significantly by state. But in the states where it works, this is the rarest signal in the vertical: a named expansion with months of lead time. Capital equipment, EHR modules, staffing, patient-acquisition platforms: anyone who sells into the build-out of a new service line should be watching these filings.

State CON program directories vary. Verify by state before building workflows.

The Healthcare Data Cookbook

Here are five recipes ranked by the asymmetry they create. Each one combines public sources into a specific dish: a pain-based segment plus a message angle no competitor can match, because none of your competitors has combined the same ingredients (yet).

Recipe 1: The Breach Aftermath

The Dish: A provider that just disclosed a large PHI breach and is scrambling to respond.

Ingredients: OCR Breach Portal (filter last 30–60 days, breach type = hacking/IT incident, 500+ individuals affected) + NPPES or Care Compare (facility profile) + (optional garnish) job postings for security, privacy officer, or CISO roles at the same entity in the same window.

Indicator Type: Trailing on the breach; leading on the remediation buying cycle.

Target Segment: Cybersecurity vendors, HIPAA and GRC platforms, incident-response firms, cyber insurers.

Read more

When we were kids, the rules of life were hard and fast. Don’t run with scissors. Don’t go swimming after you eat. Don’t stick metal things in the wall socket. Don’t put forks in the microwave. And don’t build your own software.

Hold our beer.

We’re about to run with the scissors straight into that wall socket. Because one of those rules just stopped being true, and most growth leaders haven’t noticed yet.

Start with the list

Before you read another word, do one thing. Open a spreadsheet. List every tool in your stack and what you pay for it. Every subscription. Every seat. Every “platform fee” you forgot you signed up for. Estimate if you have to. Nobody’s grading the math.

We call it the Grail list. It’s the most uncomfortable hour you’ll spend this quarter, and it’s the most useful.

Here’s why it works. You cannot make a single smart decision about build, buy, or anything else until you see the whole bill in one place. Most leaders never do. The spend is scattered across departments, buried in annual contracts, hidden behind names nobody remembers approving. You feel the pain at budget review. You never see the shape of it.

Make the list. Now cross off the three you can’t touch yet. Your CRM, because the whole company runs on it. Your marketing automation platform, because unwinding it is a year of your life. Your data provider, the ZoomInfo line because you haven’t fully deployed Cannonball GTM yet (clears throat).

What’s the next biggest number?

For most revenue teams, it’s the conversation intelligence line. It’s Gong. Ouchie.

What you’re actually paying

Gong doesn’t publish pricing. There’s no self-serve plan. No sandbox. Every deal runs through a sales cycle, and what shows up at the end is a three-part bill most buyers never saw coming.

A per-user license. A platform fee that doesn’t care how big your team is. And an implementation charge.

Run the math for a 50-person revenue team. Licenses land around $76,000. The platform fee adds up to $50,000 on its own. Onboarding, another $15,000. Year one comes in around $140,000. Go for the full stack with Forecast and Engage bundled in, and you’re north of $200,000.

That’s $235 per user per month, before anyone has booked a single meeting off it. And the contract locks you in for two to three years, with an auto-renewal uplift baked in so the number only goes up.

That’s what you’re paying. Now look at what it does.

What Gong actually is

Strip away the category language and the revenue-intelligence branding. At its heart, Gong is three things.

It records and transcribes your calls. It dumps those transcripts into a repository. And it reads them against a sales methodology to tell your reps what they missed.

That’s the product. A transcription service, a place to put the transcripts, and a coaching layer on top.

Here’s the part that should make you angry. The one thing Gong should be brilliant at, asking the whole pile of conversations a hard question, is the thing it’s worst at. Try to query a Gong database for the common pains your buyers raised last quarter. Good luck. The metadata is thin. The trends are buried. You’re paying $140,000 for a system that captures everything and tells you almost nothing about the pattern.

So we asked the obvious question. What would it cost to build it ourselves?

The rebuild

We’re going to walk this in three steps, lazy to sophisticated. Watch the cost.

The lazy version. Granola for transcription, fourteen dollars a month. Claude to do the reading. Your existing CRM to hold the output. You build a small coaching component that takes the transcript, checks it against whatever methodology you run, MEDDIC, BANT, SPICED, whatever, and tells the rep what they should have said. Three tools. It runs in perpetuity. It’s janky. For a small team, it works.

The real version. Now you add a database so you’re not capped, Supabase or Mongo, somewhere to actually keep the transcripts at volume. You add a workflow step that moves the data automatically. And you have Claude read across every transcript to build pain footprints and keyword maps, the trend layer Gong never gave you.

The sophisticated version. Add a vector layer, Pinecone or similar, so you can ask the whole corpus real questions and get real answers. Add structured analysis that gets better over time instead of starting cold every run. This is the part that beats Gong at the exact thing Gong is bad at.

And here’s the kicker. You don’t need a frontier model for any of this. “Did this rep follow the methodology” is a Haiku-class question. The cheapest, dumbest model on the menu does it fine. A guy named Bernie in a back room could run it.

So what does it cost? For a normal team, somewhere around $700 a month. Push it to the most fantastical scenario, big team, sloppy token use, nothing optimized, pure jank, and maybe you hit $1,000 to $1,500 a month.

Fifteen hundred a month against a hundred and forty thousand a year. That’s the gap. That’s the whole post.

Now the part nobody selling you a weekend build will say

We are not telling you to go build Gong. Read that twice.

The cost of building collapsed. The cost of maintaining did not. The moment you build that system, you own it. When it breaks, you fix it. When the API changes, you fix it. When the person who built it leaves, and it lived in their head, you’re stuck holding a system nobody understands. We can build almost anything now. The question that still matters is the old one: how well does it actually work for the people who have to use it every day?

And there’s a real reason to buy Gong, a reason that has nothing to do with the technology. Governance. Security. Support. A vendor on the hook when it fails. A system that doesn’t walk out the door when one person quits. For a lot of companies, that’s worth six figures. For a lot of others, it absolutely is not.

That’s the decision. Not “can I build it.” You can. It’s “should I own this, or should I rent the governance.”

The three questions that fall out of this

You don’t need a Gong teardown for every tool on your Grail list. You need the pattern. Three questions answer most of it.

How complex is the motion? Simple, repeatable motions, the kind packaged software automates, are the easiest to rebuild. The more sophisticated and specific your motion, the more the math tips toward owning it, because no vendor flexes to fit you.

What’s the gap between replacement cost and replaced cost? If you can rebuild it for a fraction of the annual license, the vendor isn’t selling you technology. They’re selling you a pricing model. Gong fails this test by a mile.

How good is your source of truth? This is the one that humbles everyone. Clean data and a real source of truth raise your ceiling, you can build almost anything. A fragile, duct-taped source of truth caps you, and forces you to either buy your way out or accept lag you build around. Most teams are worse off here than they think, and it’s the thing that decides what’s actually possible.

Run the list

The rule we grew up with, don’t build your own software, was true for thirty years. It isn’t anymore. Building got cheap. Deciding got hard. The value moved from the building to knowing what’s worth owning in the first place.

So make your Grail list. Find your Gong. Do the math on the gap. And before you build a thing, ask the question that actually matters now, the one about governance, because that’s where the next decision lives.

That’s the next post.

—-------------------------------------------------------------------

New to Cannonball GTM? We help growth leaders find hidden buyers using publicly available data and figure out which tools, agents, and platforms are actually worth your money. Start with our most-read posts:

• The Beginner’s Guide to Cannonball GTM — Start Here

• Finding Hidden Customers: The Playbook

• The Outbound Campaign Testing Bible

Keep reading

This post is part one of a series on building, buying, and surviving the agent era.

Start at the beginning with The Agent Landscape for Growth Leaders, the rubric, and the decision table that maps fifteen use cases against build it yourself, build it on a platform, or buy it.

Next up: The Pragmatist’s Guide to Governance, where we get into the real reason to buy instead of build, and how to keep a system you built from walking out the door when the person who built it quits. [Coming soon]

Nonprofit and education are two of the most financially transparent sectors in the country. That should make selling into them easier. Should.

Every tax-exempt organization files a public annual return with the IRS. Every college that participates in federal student aid reports enrollment, finance, staffing, and outcomes to the federal government. The data is public, structured, and unusually complete.

That is the trap in this market. The data is already heavily mined, but mostly for the wrong motion. The donor-prospect-research industry uses filings, wealth markers, foundation giving, and board relationships to find donors.

But nobody is reading those same filings, trying to find institutions in operational or financial pain to sell software and services to. The opportunity is huge.

This is your guide to breaking it open.

The Inventory

We’ve ranked these public data sources by how much of an edge they give you over every other vendor fighting for attention in your prospect’s inbox.

IRS Form 990 and the EO Business Master File (projects.propublica.org/nonprofits)

Every tax-exempt organization in the US files an annual information return with the IRS: for most organizations, it’s the Form 990. Smaller orgs may file Form 990-EZ and then Form 990-PF for private foundations. About 1.8 million organizations reside in this universe, keyed by EIN and classified in the Exempt Organizations Business Master File. ProPublica’s Nonprofit Explorer makes the filings searchable and exposes a clean API on top of the IRS’s bulk XML. This is your financial X-ray of the entire nonprofit sector. For each organization, you get:

• Revenue, expenses, and net assets (with year-over-year trends readable across filings)

• Program-service revenue breakdown and grant-dependence percentage

• Compensation for executives and key staff, plus names and titles

• Named officers, directors, and board members

• Governance answers: audits conducted, conflicts of interest, and board oversight structure

• Foundation grants paid with recipient names and amounts (990-PF)

The officer and compensation data are often underappreciated. The 990 is the only source in the sector that gives you named decision-makers, their titles, and their pay in a single free filing. Before going about buying a contact list, check here first.

ProPublica Nonprofit Explorer · ProPublica API · IRS EO BMF · IRS Tax-Exempt Organization Search

IPEDS (nces.ed.gov/ipeds)

The Integrated Postsecondary Education Data System collects mandatory annual data from every college, university, and vocational institution that participates in federal student aid. It covers roughly 7,000 institutions, with downloadable data going back decades. It is publicly and freely available data, updated annually, with built-in peer comparisons. Specific fields include:

• Fall and 12-month enrollment by level, status, race, and gender

• Institutional finance: revenues by source, expenses by function, endowment value

• Completions by program and award level

• Human resources data, like staffing counts by type and function

• Student financial aid: count and amount by program

• Graduation and transfer-out rates

• Admissions rates and test-score ranges

The finance-and-enrollment combination is the recipe base for most of what follows. Three years of enrollment decline, high tuition dependence, and thin reserves is not a research note. It is a sales trigger. No call is needed to establish the problem. The data is already there.

IPEDS Use the Data · IPEDS Data Center

Federal Student Aid Financial Flags (studentaid.gov/data-center/school)

The Department of Education publishes three financial-oversight signals for higher-education institutions: the financial-responsibility composite score, the Heightened Cash Monitoring list, and the cohort default rate.

A school on the Heightened Cash Monitoring list has been flagged by the federal government for financial or compliance concerns. Sometimes this requires a letter of credit before they can receive federal aid. That is a public, dated, forced-action distress signal. Fields include:

• Financial-responsibility composite score (below 1.5 triggers heightened scrutiny)

• Heightened Cash Monitoring tier: HCM1 or HCM2, with HCM2 the more severe designation

• Cohort default rate by year

A school on HCM2 is not a prospect that might have a problem. The Department of Education has already made that case, in writing, with oversight conditions attached. This is the higher-ed cousin of an OSHA consent order or a bank enforcement action.

Federal Student Aid Data Center

Single Audits — Federal Audit Clearinghouse (fac.gov)

Any organization that spends $750,000 or more in federal funds in a year must file a single audit. The findings are public: material weaknesses, significant deficiencies, questioned costs, and the federal programs involved. The Federal Audit Clearinghouse holds filings from fiscal year 2015 forward, with ProPublica linking many of them.

A finding tells you exactly what broke, where it broke, and whether it was fixed last year. Per filing:

• Auditee name, EIN, and fiscal year

• Federal programs audited and dollar amounts

• Findings: material weaknesses, significant deficiencies, or qualified opinions

• Questioned costs with amounts

• Prior-year findings showing whether the same problems appeared before

That last column is the sharpest cut in the dataset. If the same material weakness appeared in the prior year, the organization has not fixed it. That is not just pain. That is pain with a documented failure to remediate, a new audit window approaching, and a federal program relationship at risk. Anyone selling grant management, financial controls, or compliance remediation should be monitoring the clearinghouse.

Federal Audit Clearinghouse

NCES Common Core of Data and F-33 Survey (nces.ed.gov/ccd)

The Common Core of Data covers every public school and district in the country. The Census F-33 survey adds per-pupil revenues and expenditures by district. This is the K-12 universe, with finance and staffing attached, and should serve as the foundation for any district-level targeting. The data set includes:

• District and school enrollment by grade, race, and gender

• Staff counts: teachers, administrators, instructional aides

• Locale type: city, suburban, town, or rural

• Per-pupil revenues by source: local, state, and federal

• Per-pupil expenditures by function: instruction, administration, support services

The per-pupil expenditure breakdown by function is the one to watch. A district with flat instructional spending and rising administrative costs is not the same prospect as one losing federal revenue. That is a finance conversation dressed as a budget line, and it is publicly reported annually.

NCES Common Core of Data

College Scorecard (collegescorecard.ed.gov/data)

College Scorecard adds outcome and debt data that IPEDS does not carry: graduate earnings, federal loan amounts, and debt-to-earnings comparisons, built from IPEDS, federal aid records, and IRS tax data. Free download and API. Fields include:

• Median earnings at one, five, and ten years post-enrollment, by program

• Median federal loan amounts and monthly payments

• Earnings relative to high school graduates

• Program-level outcomes in more recent Scorecard releases

The GTM angle: any vendor making a value or ROI argument to enrollment management, student-success, or career-services teams now has a public data point for that argument, per institution, per program, before the first call. Scorecard turns “we help improve outcomes” into “your graduates in this program are earning X, against a peer median of Y.”

College Scorecard Data

Federal and Foundation Grant Data (usaspending.gov / candid.org)

Federal grant awards via USAspending and Grants.gov; foundation grantmaking via Candid and 990-PF filings. Together, they reveal who just won funding and whose funding is about to end.

A grant landing is a moment of scaling. A grant ending is a funding cliff. Both are temporal triggers that are readable in advance, from public data, before the organization has communicated any of them to the market. Fields across sources include:

• Awarding agency and program name

• Recipient name, EIN, and address

• Award amount and period of performance (start and end dates)

• Place of performance

• Foundation grants: recipient, purpose, amount, and year (from 990-PF)

The period-of-performance end date is the signal most teams miss. A major federal grant ending in seven months at an organization where that revenue represents 40% of the budget is not a future problem. It is a current decision window. Grant dependence is readable from the 990. The end date is in USAspending. Crossing them is the recipe.

USAspending · Grants.gov · Candid

State Charity Registration and AG Filings

Most states require charitable organizations to register annually and renew their registrations, and state attorneys general's offices publish enforcement actions and lapsed registrations. Each state maintains its own portal. A lapsed registration is a compliance signal. An AG enforcement action is louder. Fields vary by state but typically include:

• Organization name, EIN, and address

• Registration status and expiration

• Enforcement actions and findings where published

Vendors with strong state-specific coverage or a compliance-oriented product have an information edge that no one else is competing for, because no one wants to do the grunt work of assembling the data sets.

IPEDS Human Resources Data and Job Postings (nces.ed.gov/ipeds)

IPEDS staffing data shows the count and type of employees per institution over time. Year-over-year trends surface layoffs, hiring freezes, and staff transitions before any press release appears. Job postings add the real-time, unstructured layer on top. You should be extracting:

• Role, location, and posted date

• Full description text: tools named, pain language used, certifications required

• Posting velocity over time, i.e., how fast the institution is adding or contracting headcount in a given function

Posting velocity is the signal inside the signal. A single admissions coordinator posting is noise. Four enrollment-marketing postings across twelve weeks at a college with a three-year declining cohort is a pattern. That pattern is a trigger.

MSRB EMMA Database of Municipal Bond Disclosures (emma.msrb.org)

Public colleges, many K-12 districts, and nonprofit hospitals that issue municipal bonds must file continuing disclosures on EMMA, the Municipal Securities Rulemaking Board’s Electronic Municipal Market Access system, including financial statements, material event notices, and rating changes.

The 990 does not apply to public districts and state colleges because they are government entities, not tax-exempt organizations. EMMA does. Fields include:

• Issuer name and bond series

• Official statements and annual financial disclosures

• Material event notices: rating changes, payment defaults, covenant violations

• Continuing disclosure filings with financials attached

MSRB EMMA

A note on 2026 availability. Federal education data programs have undergone restructuring, and the continuity of some NCES and Department of Education datasets is in flux. The IRS 990 data, the Federal Audit Clearinghouse, and USAspending are on firmer footing. Before building a workflow around a specific federal education dataset, confirm current availability.

The Nonprofit and Education Data Cookbook

Five recipes, ranked by the asymmetry they create. Both sectors, tagged. The data is mostly free. The combination is the work.

Grab the whisk.

Recipe 1: The Enrollment Cliff [Higher Ed]

The Dish: A tuition-dependent college where enrollment has been falling for years and the financial consequences are now visible in the public filings.

Ingredients: IPEDS enrollment (3+ year undergraduate decline) + IPEDS finance (tuition as a percentage of total revenue above 75%, thin operating reserves) + optional garnish: a job posting for enrollment marketing or a hiring freeze across administrative roles in the same window.

Read more

I have spent the last two months building an agent.

Not writing about agents. Not teaching agent development. Building one, for one reason: I needed to move faster with clients.

Here is the problem it solves. The Finding Hidden Customers playbook works. The four-prompt chain finds pain segments your competitors cannot see. But running it by hand takes time. Research in one tool. EDP analysis in another thread. Segment scoring. Data mapping. Then the loop between scoring and data until the rankings settle.

I do this four to five times a week. So I built a thing that does it for me.

What It Does

You give it a brand. It does the rest.

It researches the company. It maps the regulatory environment and the public databases that reveal buyer pain. It finds the Existential Data Point. It builds pain segments and ranks them gold, silver, and bronze. It scores each one on urgency, observability, value prop fit, buyer resonance, and market density. Then it hands you a verdict and an execution roadmap your go-to-market engineer can run.

It does all of this stage by stage. At every stage it stops and lets you push back. Correct the buyer. Adjust the deal size. Tell it the website is wrong about who actually buys. It revises and keeps going.

That last part matters. The agent is not a black box that spits out an answer. It is a working session. You stay in the loop because your judgment is the thing that makes the output real. I learned that the hard way watching growth leaders use early versions. The ones who handed over context up front got segments they could defend. The ones who skipped it got segments they rejected.

The agent is built to keep you in the chair.

Why I Am Giving It to You

I am not going into the software business. I spent thirty years in and around software. I know exactly what it costs to host an application, support it, fix it, and keep it running. That is not the business I am building.

The business I am building is this. I learn the methodology by doing the work. I build tools that make the work faster. And I hand those tools to the people who pay to learn alongside me.

Think of it like a meal kit. There are people who will teach you knife skills. That is a real and valuable thing. I am handing you the kit, tested and portioned, with the recipe that works. You still cook. But you are not starting from an empty kitchen.

Agent development is going to be part of every growth leader’s job. I am doing that work first so you understand what good looks like. As I build each agent for my practice, I will hand it to you. This is the first. There will be more, one for each playbook.

How to Get It

The agent runs on your own infrastructure. Your API key. Your password. Your data. Nothing shared with me or anyone else.

There are two ways to get it running. Pick the one that fits you.

Path 1: One Click on Railway (hosted, easiest)

Railway hosts the agent for you at your own private URL. You need three things:

1. A GitHub account, connected to Railway. This is the step people miss, and it is the number one reason deploys fail. Railway pulls the agent’s code from GitHub when it deploys. If your Railway account is not connected to a GitHub account, the deploy hangs or fails with “GitHub Repo not found.” A free GitHub account works fine. Connect it in Railway under Account Settings, then Integrations, then GitHub. One time, about a minute.

2. A Railway account. Free to start, about five dollars a month for hosting.

3. An Anthropic API key. About three dollars per brand analysis, billed to your own account.

Click the button (above) or here. Paste your API key. Choose a password. Deploy. Two minutes, no terminal, no code. The README walks you through every step, including how to get your Anthropic key.

If your deploy hangs or fails: nine times out of ten it is the GitHub connection. Go to Railway, then Account Settings, then Integrations, and confirm GitHub is connected. Then delete the failed deploy and click the deploy button again fresh. If it still will not go, use Path 2. It is the same agent.

Path 2: Run It on Your Own Machine (local, no hosting fee)

If you are comfortable spending ten minutes in a terminal, you can skip Railway entirely and run the agent on your own computer. No hosting account, no monthly fee. Your only cost is the API usage, about three dollars per brand.

1. Download the agent files here: https://github.com/marketadvocate-sudo/Finding-Hidden-Customers-Public-Agent/archive/refs/heads/main.zip

2. Unzip it. Open the SETUP file inside. It walks you through every step for Mac and Windows: installing Python, setting your API key, and starting the agent. First time takes about ten minutes. After that it is one command to launch.

3. Open your browser to the address the agent gives you, enter your password, and run a brand.

This is exactly how I run my own copy. Local is not the consolation prize. It is the version with the fewest moving parts.

One Honest Word About Beta

This agent is lab-tested. I have run it across more than seventy brands and refined it with feedback from working growth leaders who used it on their own clients. It produces consistently strong results. I trust it enough to use it in live engagements.

It is still beta. That means it is good and getting better, and your feedback is the fuel. When something does not fit your industry or your situation, tell me. That is how the next version gets sharper.

Where to Start

If you have not run the four prompts by hand yet, start there. The prompt library is the methodology. It teaches you how the work thinks, and you will get far more out of the agent once you understand what it is doing under the hood.

If you already know the chain, deploy the agent and run a brand you know well. Compare what it finds to what you already believe. That is the fastest way to trust it.

The prompts teach you to find hidden customers. The agent finds them for you.

Now go find buyers your competitors cannot see.

Construction is one of the best industries for the CannonballGTM method: it’s a huge, fragmented market driven by deadlines and chock-full of public data that most outbound teams never bother to consider. But too many GTM teams stop at the first signal: A permit gets filed. Someone says, “Hey, you guys! A project is starting! Let’s sell something to the contractor!”

[long DMV sigh]

The permit is not the insight. The permit plus something else is the insight.

This is the guide to finding that something else.

The Inventory

We’ve ranked these public data sources by how much of an edge they give you over every other vendor fighting for attention in your prospect’s inbox.

Shovels (shovels.ai)

Building permits are the first physical signal that something is getting built. Shovels provides a clean, unified layer on top of the most fragmented public dataset in construction: local permits filed across 48 states and thousands of jurisdictions. Data include:

• Permit records: type, status, dates, job value, and work classification across 22 categories

• Contractor profiles: firmographics, revenue estimates, service area, and full permit history

• Contact data: employee names, titles, emails, and phones for 2.3M contractors

• Derived metrics: inspection pass rates, average project duration, monthly activity trends, and work-type tallies

• Pre-permit decisions data: city council and planning approvals that surface projects before a permit is ever filed

• State-level contractor group IDs that consolidate related entities under one roof

Shovels.ai · Permit database · API docs · Construction leads workflow · Shovels/Clay integration

OSHA Enforcement and Injury Data (osha.gov/data)

OSHA provides a federal record of who was inspected, who was cited, who was fined, and who reported a severe injury. This is the most underused buying signal in construction for anyone selling safety, compliance, training, EHS software, or insurance. Keep in mind it is a trailing indicator, so the pain has already materialized or the incident has already happened, which means the message does not have to convince anyone they have a problem. It just has to prove you already know what happened. Data set includes:

• Establishment search: inspections, citations, and penalties searchable by NAICS or SIC code

• ITA injury data: Form 300A summaries plus 300 and 301 case detail, with recordable cases, days-away counts, fatalities, and hours worked for rate normalization

• Severe Injury Reports: amputations, hospitalizations, and eye loss since 2015

• Fatality and Catastrophe investigations: the full record of the worst outcomes

• Accident Investigation narratives: the OSHA-170 text describing exactly what happened, in plain language

That last one is underappreciated. The narrative text is where the specificity lives. It is not just a citation number and a dollar amount. It is a description of the moment the pain became real.

Establishment-specific injury and illness data · Data downloads and FOIA

State-Level UCC-1 Filings (nass.org/business-services/ucc-filings)

A contractor finances a crane. The lender files a UCC-1 with the Secretary of State. Most GTM teams in construction have never looked at one. Which means this is still wide-open asymmetry. The five-year lapse cycle, with continuations and terminations also on the record, creates a natural hook for re-engagement that nobody is using. Data includes:

• Debtor name and address

• Secured party (the lender, which often tells you what kind of financing it was)

• Collateral description, often naming the specific equipment

• Filing date, lapse date, continuations, and terminations

The collateral description is the signal. “One Caterpillar 336 hydraulic excavator” is not just a legal formality. It is a window into what the contractor is building the capacity to do.

The National Association of Secretaries of State has a shared landing page that links to the UCC portal for any state:

NASS UCC Filings Portal

Mechanic’s Liens

When a contractor, sub, or supplier does not get paid, they file a mechanic’s lien against the improved property. It is a public statement that money is stuck and a project has friction. It points straight at payment, billing, and project-financing pain. A lien is one of the most specific cash-flow-stress signals in any industry. Data sets include:

• Claimant who is owed money

• Property owner

• The improved property’s specific address

• The unpaid amount

• The project

• Filing date and statutory deadlines

Every field is useful. The property address lets you cross-reference against permits. The unpaid amount indicates how severe the cash-flow problem is. The statutory deadlines tell you how urgent the situation is right now.

Mechanic’s liens are filed with the county recorder, so access is fragmented, but sites like Levelset aggregate them to provide a clean, queryable data source.

Levelset Mechanic’s Liens data sets.

Public Construction Contracts and Bid Opportunities (SAM.gov)

Federal opportunities and awards via SAM.gov. State and local procurement, such as DoTs, school districts, water and sewer authorities, and municipalities, via a mix of free portals and paid aggregators.

A recent award indicates that a contractor has just taken on funded work with a deadline, a scope, and usually bonding and compliance obligations attached. That is a bundle of fresh, datable pain. And unlike a permit, it comes with a dollar amount attached from day one. Data points include:

• Opportunity notices: pre-solicitation, solicitation, award, and sole-source announcements

• Award records: amount, recipient, NAICS code, place of performance, agency, and set-aside type

• Entity registration and small-business status

The place-of-performance field is underused. It tells you exactly where the funded work is happening, which lets you build geography-first outbound around specific metros or corridors.

SAM.gov opportunities · USAspending prime awards

Additional Source: ConstructionBids.ai vs GovWin

Construction Firm Job Postings

A burst of requisitions for superintendents, estimators, or project managers is a backlog-and-capacity signal. The job description itself often names the tools they use, the volume they manage, and the strain they feel. This is the highest-asymmetry unstructured source in the vertical. No list-buyer is parsing this text. You should be extracting:

• Role and location

• Posted date

• Full description text: tools named, volume described, certifications required, pain language used

• Posting velocity over time, or how fast the company is adding headcount in a given function

That last one is the signal inside the signal. A single estimator posting is noise. Three estimator postings in sixty days is a pattern. Posting velocity is what turns a job board into a growth-and-strain detector.

State DoT Oversized and Overweight Load Permits

To move heavy equipment on public roads, a contractor files for an oversize or overweight permit with the state Department of Transportation. That filing tells you what equipment is moving, from where to where, and when.

Equipment in motion is the clearest possible signal that a project is starting or an asset is being repositioned. Format and availability vary by state, so confirm coverage for your specific territory before building a workflow around it. Data sets include:

• Carrier or contractor name

• Equipment dimensions and type

• Origin and destination

• Route

• Permit dates

That is enough to infer a project start, an asset repositioning, or a contractor expanding into a new geography.

Contractor Licensing Boards

Every state licensing board publishes its roster of active license holders, with classification, status, and often disciplinary history. California’s CSLB is the canonical example. On its own, it is a denominator, not a signal. Many licensed contractors have not pulled a permit in months. Use it to validate and classify. Do not use it to trigger.

EPA Enforcement and Environmental Permits (ECHO.epa.gov)

The EPA’s Enforcement and Compliance History Online or ECHO carries environmental permits, inspections, violations, and penalties, including construction stormwater records under NPDES legislation governing water pollution. For anyone selling environmental compliance, stormwater management, or site services, a violation or an active construction stormwater permit is a precise, dated pain signal.

Free, public, API-accessible.

PACER Court Records (pacer.uscourts.gov)

Federal bankruptcy and litigation records via PACER, plus state court dockets, canreveal financial distress, contract disputes, and construction-defect litigation. Powerful signals.

Slower to refresh and heavier to work with than the sources above. High value for credit, surety, and legal-adjacent plays.

EDGAR SEC Filings Database (sec.gov/search-filings)

Public construction firms, large GCs, and building-materials suppliers disclose strategy, risk, capital plans, and segment performance in their quarterly and annual filings.

This is only the public slice of the market files. But for that slice, the disclosures are rich. Most useful for selling into large public builders and material suppliers.

Free, full-text searchable, API-accessible.

US Import Records

Customs bill-of-lading data showing what materials and equipment a company is importing, in what volume, and from whom. Useful for supply-chain and materials plays.

Better as a corroborating ingredient than a standalone trigger.

Practically accessed through vendors like S&P’s Panjiva database or ImportYeti.

Census and BLS Construction Statistics

Construction spending, housing starts, employment, job openings. Everyone has this. It is aggregate, so it creates almost no asymmetry on its own. Its job is context, which regions and segments are heating up, not targeting.

Use it to frame the macro. Never use it to find the account.

The Construction Data Cookbook

Here are five recipes ranked by the asymmetry they create. Each one combines public sources into a specific dish: a pain-based segment plus a message angle no competitor can match, because none of your competitors has combined the same ingredients.

Go get the apron and put on the chef’s hat:

Recipe 1: The Stalled Project

The Dish: Funded, authorized projects that started and then quietly… stopped.

Ingredients: Shovels permits filtered for an “approved” status 18 months ago that never reached a “final” inspection + the contractor profile + (optional garnish) a fresh mechanic’s lien filed for the same address.

Read more

We have both stood where you are standing. Staring at a dashboard, trying to work out why the company is missing its number. We have done this for a combined sixty years (Do not ask who is carrying which share of that sixty).

Here is the first thing six decades teaches you: Growth is a trailing indicator.

It is the thing the board stares at, the number on the slide, the symptom that brought you into the room. But you cannot treat growth directly. You can only treat the things upstream of it. And the most important thing upstream of growth, the one you can actually measure and fix, is the meeting.

Not just any meeting. A meeting that can become revenue. So this is a post about diagnosing one thing: why you are not generating enough of those.

Symptoms

A growth leader rarely walks in and says, “I am not booking enough qualified meetings.” They say, “We are not hitting our number.” Fair. That is the chief complaint, and it is real. But it is too big to fix. Please don’t step in that bear trap. You have to find the underlying symptom you can put your hands on.

That symptom is the BANT-qualified meeting. Budget, authority, need, timing. A meeting that has those can close. Everything we diagnose here lives upstream of that meeting. What happens after qualification is a different examination for a different day.

Which means we are going to ignore almost everything your tools are showing you. Open rate. Reply rate. Positive reply rate. Click-through rate. We know that’s uncomfortable. Those numbers feel like progress. But a campaign can post a 20 percent reply rate and still be dying, because “sounds interesting, send me more” is not a meeting. It is a polite goodbye. The dashboard is not the patient. The meeting is.

So we borrow one habit from pathology and leave the rest of the metaphor at the door. The habit is this: look at the most common and most visible cause first, rule it out before you go deeper, and never invent a cause you cannot prove. There are two causes you can measure. There is a third you cannot yet. We’ll be honest about which is which.

Start With the List

The first cause is targeting. This is also where most campaigns die, before the message is ever read.

The test is one number: Your meeting booked rate. Meetings booked divided by total activities. You can do it on a napkin or use the calculator from the Meeting Booked Rate post. Then read it against two reference points from the field (see below).

The TAM spam ceiling is about 0.25 percent. Untargeted volume, working through a territory or title filter, tops out at a low level. Across B2B, cold outreach-to-meeting averages near 0.1 percent, while SaaS lands between 0.3 and 0.8 percent. That is not a failure number. It is simply the best that “praying and spraying” can do.

The pain-based standard is 1 percent. Well-targeted campaigns are benchmarked at 1-3%. The gap between the ceiling and the standard is neither effort nor volume. It is relevance. And, relevance is pain. The field backs this up plainly: tightly targeted lists with under 50 contacts average a 5.8 percent reply rate, compared to 2.1 percent for large sends and below 3 percent. The problem is almost always relevance, not the subject line.

So here is your first finding. At or below 0.25 percent, the list is the cause. You are pointed at accounts that look right on paper and hoping to land on someone in pain. Do not touch the message. Fix the list. Between 0.25 and 1 percent, targeting is partly working, and you go deeper. At or above 1 percent, the list is sound; the cause is elsewhere.

One note worth sitting with. Meeting booked rate is the metric almost no benchmark report publishes, much to our dismay. We had to assemble those numbers from the few sources that bother. That gap is the whole point. The metric that matters most is the one nearly everyone skips.

Your Meeting Rate Cannot Diagnose Itself

Here is the trap, and it catches smart people. Your meeting booked rate is a composite. A one percent rate could be excellent targeting but could carry a weak message. It could be a loose list rescued by a brilliant message. It could be a real pain segment with a forgettable message, leaving meetings on the table. The number says the patient is sick. It does not say why.

You cannot pull one number apart by staring at it harder. But you can measure a second thing and compare the two. You already have the meeting rate. If you measure the message on its own, the composite comes apart, and you can see which one is the problem. That is the entire move. Two measured things, one diagnosis.

Grade the Message

So grade the message on its own, without looking at your results. Messages fall into four grades, and we describe them by what they do to the person who reads them.

Grade One (-) hurts your case. It misreads the pain, dumps features, fakes a personal touch, or sounds like every other vendor in the inbox. The reader trusts you less than if you had said nothing.

Grade Two (~) does no damage. It is competent, inoffensive, and forgettable. It does not move anyone. It is simply ignored.

Read more

Here’s what you’re getting in this post: A rubric. Three categories of agents with examples, so you know what belongs where. Then a table. The table is the thing you bookmark. It maps use cases against your options: build it yourself, build it on a platform, buy it. Green, amber, red. It comes from thousands of hours and dozens of engagements where we’ve been inside RevTech stacks, watching what works, what breaks, and what never was going to work.

Let’s be honest about something before we go any further. This is an exciting time to work in go-to-market. It’s also one of the hardest. Every week, there’s a new post telling you to go learn Claude Code or build your own agent swarm or automate your entire pipeline by Friday. Most of that advice is written by people who don’t have your job. You’re trying to hit a number, keep a team together, and make sense of a landscape that changes every week. We’ve had your job. We know. And we’re pissed about how much noise is being thrown at you right now. This post is the opposite of that noise.

Now here’s why you need it.

The RevTech stack is a house of cards on fire, and someone is pointing a high-powered fan at it while your revenue team screams.

You’ve got 45 tools. A CRM that was supposed to be the source of truth but became a junk drawer. A marketing automation platform nobody fully owns. A sequencing tool. An enrichment vendor. If you’re lucky, your intent signals live on the same platform. Maybe HubSpot is holding it together. If you’re unlucky, you’re running the Frankenstein monster that is Sales and Marketing Cloud from Salesforce, and half your team has stopped logging in.

You’re looking at this through the lens of your stack. Wrong lens.

We don’t see many organizations thinking about this correctly. Arrogance alert. Sorry, but it’s what’s happening. If we could wave a magic wand, here’s what we’d be seeing: data layer first. A single source of truth that feeds a go-to-market orchestration layer. The CRM becomes an input, not the center. An orchestration layer acts on the data, not alongside 40+ applications that don’t agree on what constitutes a qualified lead. We’ll cover the GTM orchestration layer in a future post. It will be chock-a-block full of recommendations. That’s not this post.

That’s the future: GTM Orchestration. Full stop. And it’s also a massive lift. Organizational alignment. Political capital. Budget. The kind of change that gets people fired when it goes sideways. If you’ve spent any time in a company north of $25M ARR, you know how much entropy lives between a good architecture slide and an actual implementation.

Most growth leaders don’t have the runway for that play. Not right now. We’re busy hearing about how Peter Thiel-backed Monaco just farted out $10M in ARR in three months. And we hate to say it, but it’s sorta the proof point of where things are headed.

But there’s another path. Faster. Cheaper. You can start this week. Many of you already have.

Agents.

Not the garbage AI SDR that’s going to run your pipeline autonomously. It doesn’t. Adoption of those platforms has gotten growth leaders fired. You can’t throw AI on top of a shit ICP and pray that personalization makes the difference. Something simpler. Task agents that do the work you don’t have time for. Interpretation agents that help you make better decisions with the data you already have. Autonomous agents: approach carefully, and only after you’ve earned the right to trust their outputs.

Three categories. We’re going to walk you through each one. Where to start. Where to stay away.

With your permission, Tyson Parody and Doug Bell. We’re your guides. Tyson is a Revenue Systems Architect who has been inside more broken stacks than either of us wants to count. He sees where agents break. Doug sees where the methodology breaks. Together, that’s the whole picture.

Think of this as a punk rock set list. Three chords. No filler. Just enough to play the song.

The Rubric

People hear “agent” and don’t know if they’re talking about a script that pulls data or a system that sends emails on its own. Those are not the same thing.

We know there are dozens of rubrics out there now. We don’t need you to like this one. It’s how we’re navigating, what we’re building, and why. Scott Brinker has a baller rubric you should check out. We aren’t Scott. Our knuckles drag a little too close to the ground.

Three chords:

Do This For Me. A task agent. You give it a job, and it does the job. Research a competitor. Collect public data on a set of companies. Summarize a document. A very fast intern with perfect memory and no opinions. Low risk. High value. Start here.

Help Me Decide. An interpretation agent. It takes what the task agent collected and surfaces patterns, comparisons, and rankings you wouldn’t have spotted. But you make the call. The agent does the heavy lifting. You bring the judgment.

Just Handle It. An autonomous agent. Runs without you. Sends emails. Books meetings. Decides who to contact and when. This is where AI SDRs live. This is where most deployments fail. Not because the technology can’t do it. Because nobody tested the outputs before they shipped.

The rubric isn’t about sophistication. It’s about knowing which category each use case belongs to before you build anything. The agent gives you an answer. The answer is only as good as the human who evaluates it.

One thing before the table: P0

There’s a move that comes before everything in this table. Call it P0. Before you build a single agent or buy a single tool, you teach the machine who you are. Connect your drives. Connect your email if you’re senior enough for it to matter. Feed it your transcripts, your wins, your losses, your ICP, your positioning. Then let it spend real time reading it all and telling you where your own story contradicts itself. You’re building a single source of truth the model can work from.

Here’s why this matters. You cannot orchestrate what you cannot read. Bad input equals bad output. That used to be forgivable because the models were weak. The models aren’t weak anymore. They’re sharp. Which means they’ll take your disorganized, contradictory, half-true data and confidently build on top of it, faster than ever. Garbage in, garbage out, at machine speed.

Most organizations skip this. The data layer is the sacred cow nobody wants to touch. It’s not sexy, it’s expensive, it’s boring, and it has no owner. So everyone builds on a foundation that was never poured.

If you’re just starting, this is good news. You pour the foundation before the mess accumulates. Congratulations. You’re ahead of companies ten times your size.

We’re not going to fully unpack P0 here. That’s a future post, with Tyson walking through the architecture in depth. For now, know this: every use case in the table below works better when P0 is in place. Some barely work without it.

The Tables

Before you look at the tables, a quick note on the columns. ‘Build It Yourself’ means you and Claude Code (or Cursor, or whatever you’re using - or having a team member who can do this - whatevs) in a terminal, building the thing from scratch. No vendor. No platform. Just you and the model. ‘Build It on a Platform’ means you’re using an orchestration layer like LangChain, Relevance, or one of the dozen others that let you assemble agents with less raw code but more dependencies. ‘Buy It’ means you’re paying a vendor for a finished product that promises to do the job out of the box. Each column carries different tradeoffs in cost, control, and risk. The tables maps those tradeoffs against the rubric so you can see where each option makes sense and where it doesn’t.

Reading this table: Green means go — this approach works today for this use case. Yellow means it works, but you need to know what you’re getting into. Red means stop — either it doesn’t exist, it’s too complex, or the risks outweigh the benefits right now. This landscape is changing fast. What’s red today may be green in six months.

You’ll be seeing this table quite a bit in the future. As we see the market evolve, our recommendations will shift. The rubric, not so much.

What Comes Next

Agents are more capable than most growth leaders realize. Most agent deployments fail because no one checked the output before it shipped. Both true.

The teams winning right now are not sprinting to Just Handle It. They know which category each use case belongs in. Do This For Me is cheap, fast, and real. Help Me Decide is where the value lives. Just Handle It is where the risk lives.

Post one of three. Next question: where do these agents actually live? Build versus buy. Platform versus DIY. That’s where decisions get expensive.

We pulled two cross-references out of the tables to keep it clean. Both are worth your time if the cells above raised a question.

On why personalization keeps failing: most teams treat personalization as the fix when it’s actually the symptom. The differentiator was never the tool. Read: Why Personalization is DOA.

On testing before you scale: the single most skipped step in outbound, and the one that separates teams who learn from teams who burn budget. Read: The Cannonball GTM 2-Week Test Protocol.

Welcome to Live Fire, the series I wanted to call Fuck the Noise, Do the Thing, because that’s the entire point: stop reading about go-to-market and go produce a result.

Your reply rates used to be fine. Now they’re not.

You’re sending the same volume. Maybe more. The sequences look right, the copy reads clean, but the meetings aren’t landing the way they used to. And somewhere in the back of your head a quiet thought has been getting louder. I don’t think these are the right people.

You’re probably right.

You’re not crazy

Here’s the part nobody says out loud. Targeting is where most outbound quietly bleeds out. Not the subject line. Not the CTA. The list. You can write the best email of your career, send it to the wrong 2,000 people, and it will do nothing, and then you’ll spend three weeks blaming the email.

(If you want the long version of why your ICP keeps lying to you, start here: Beyond the ICP: A Guide to Pain-Based Segmentation. Otherwise stay with me. This costs you an hour, and it’s worth it.)

A 20-second gut check before you spend the hour

When outbound stops working, it’s almost always one of three things.

1. Bad list. You’re reaching people who don’t have the problem you solve, or don’t have it badly enough to do anything about it.

2. Bad message. You’re reaching the right people, but you’re leading with you instead of them.

3. Too slow. Right people, right message, but you showed up after the moment already passed.

Now read those again. If your message hasn’t changed and your speed hasn’t changed, but the results have, it isn’t three and it isn’t two. It’s the list. The market moved and your targeting stayed put.

That’s the whole diagnosis. You don’t need a forty-question audit. You found this post because you already suspected it. At this point it doesn’t matter if you’re 100% sure. You’ve made the call. It costs you an hour.

Why the right list is so hard to find

Here’s the thing underneath the thing.

Companies switch their providers, software, banking, legal, telecom, roughly once every five years. Do that math and only about 20% of your market is in play in any given year, and only about 5% in any given quarter. Professor John Dawes at the Ehrenberg-Bass Institute put the number on it, and once you see it you can’t unsee it.

Five percent. That’s the pond everyone is fishing in.

If your market is 10,000 accounts, roughly 500 are actively shopping right now. Your competitors are chasing the same 500. Your marketing team is chasing the same 500. Your agency is chasing the same 500. This is where inbound works beautifully and where everybody is already optimized to win. It’s also a ceiling.

The opportunity isn’t in the 5%. It’s in the roughly 15% that has the problem you solve right now and just hasn’t started shopping yet. Real pain, quiet inbox. Nobody’s emailing them about it because nobody’s bothered to find them. You can’t buy that list. You have to build it, by finding the companies where a specific painful condition is already true, using data sitting in public view.

That’s pain-based segmentation. (The full argument, including why those early meetings are structurally different from inbound, is in The 5% Problem. Worth your time after this.)

In the next hour, you’re going to build your first one.

What you’ll walk away with

Let’s be binary about it.

By the end of this hour, you will have a clear yes or no answer to one question: are there pain segments in your market that will out-produce your current ICP?

And if the answer is yes, you’ll have those segments scored and ranked by how likely they are to book you meetings. Not a vibe. A list you can hand to your SDRs on Monday.

What you’ll need (5 minutes to gather)

Grab these before you start. Nothing here requires digging.

• What you sell, in two or three plain sentences.

• Your current ICP. Job title, company size, industry. The filter you use today.

• Your average deal size. Rough is fine.

• The problem you solve. This is the gun to your head problem. The one problem your product or service solves no matter what.

• An hour and a clear head. An AI assistant helps with the research, but the thinking here is yours.

That’s it. Let’s go.

The exercise: four moves

Here’s the exercise. In the next hour, you’ll take one company, yours, and find the single piece of public data that proves a buyer has the problem you solve, right now. You’ll turn that into three or four sharp pain segments, score them, and walk away knowing whether your targeting is the reason your outbound is slipping. No tools, no prompts, no theory, just you, one company, and an answer you didn’t have this morning.

We’re not going to hand you a tool for this. We’re going to teach you to think the way the methodology thinks, on one company, by hand. That’s the point. The tools come later, and they’re only worth anything because this part is sound.

Four moves. Work them in order. We’ll show you each one on a real company, Texada, an equipment rental software business, so you can see what “good” looks like before you do your own.

Move 1 — Name the pain you actually solve

Forget your ICP for a second. Forget titles and company size. Answer one question in plain words: what painful problem does your product make go away, and what has to be true inside a company for that pain to be sharp right now, not someday? Write down the top three conditions.

Texada: The pain isn’t “they need better software.” It’s “their equipment is sitting idle, and every idle day is revenue they will never get back.” That’s the pain. The software is just the fix.

Move 2 — Find your line in the sand

Now find the single number that separates a company that’s fine from a company that’s in trouble. The metric that flips you from nice-to-have to must-have the moment they cross it. We call it the Existential Data Point. There’s almost always one that matters more than all the others, and when you find it you’ll know, because it comes with a real consequence. A deadline. A fine. Lost revenue. Someone’s job.

Texada: The EDP is 60% equipment utilization. Above it, healthy. Below it, bleeding cash. One number. Everything else hangs off it.

Move 3 — Split your market by where they sit on that line

Here’s where it gets good. Most people have one ICP. You’re about to have three or four sharper targets. Sort your market by where companies fall relative to your EDP. Each group is in a different kind of pain, which means each one needs a different conversation.

Texada: The 60% line creates three. Below 60%, bleeding cash, act now. Between 60 and 70%, the Goldilocks zone, optimize. Above 70%, capacity constrained, a completely different problem (they need to expand, not trim). Three segments. Three messages. One number made all of them.

Move 4 — Score them, then face the data

For each segment, gut-check three things. How bad is the pain. How likely are they to take a meeting. Is the deal big enough to be worth the effort. High, medium, low is plenty.

Then the question that decides everything: can you actually find these companies in public data? Not buy a list. Find them, one real company at a time, because something public proves the pain is true for that specific company. A permit. An enforcement action. A filing. A recall. A job posting. That proof has a name in the methodology. It’s a Data Key. A segment you can’t find is a daydream. A segment with a Data Key is a campaign.

Now read your verdict

Look at your ranked list. The best segments are high pain, likely to take a meeting, and worth the deal size, with a Data Key you can actually go get. Strong on one of those is a maybe. Strong on all of them is where you point your team next week.

If even one pain segment beats your current ICP and has a Data Key behind it, you have your answer. Targeting was the problem, and you just found the fix. If nothing beats your ICP, that’s a real answer too, and it’s worth knowing before you blame another email.

What you just did

That hour was the whole idea behind Cannonball, done by hand on a napkin. You threw out the ICP, found an Existential Data Point, built pain segments off it, and pressure-tested them against public data. No personas. No guessing. Pain you can prove.

Doing it once, on one company, by hand, is the lesson. Doing it on every vertical you sell into, systematically, with a Data Key that proves the pain company by company, is the system.

That system is the Finding Hidden Customers prompt library, the implementation layer behind everything you just did. It’s the same four-move logic, productized into a tested chain and run on real data. And tested is the word. We’ve put it through hundreds of brands over the past year, very often live and on stage with the clock running. (Last week alone, roughly twenty brands in a single week across Copenhagen and Berlin. Stressful every time. It holds up every time.) Paid subscribers get the full chain. The Finding Hidden Customers agent runs the whole thing in one pass.

Process over prompts. You just proved it. The prompts only matter because the process underneath them is sound, and you now own the process.

Want to see it worked end to end on a real company, with the economics attached? Read Pain-Based Segmentation: The $135M Opportunity for DISCO.

You found the right people. The next question is what on earth to say to them so they actually reply. That’s the next exercise.

New here? Start with the Cannonball GTM methodology · Finding Hidden Customers: The Playbook · The Outbound Campaign Testing Bible