What I’m covering in this newsletter:

• How the different sales and marketing data sets commonly disagree

• A live dashboard built with Claude you can click through, built on a full year of actual data

• A roadmap to cleaning the data built into the dashboard

• How to run this against your own HubSpot, Salesforce or other data source in about 45 minutes

• Bonus for paid subscribers: the GTM Reconciliation Kit, which rebuilds this exact dashboard on your own exports.

What if Sales and Marketing were literally on the same page?

What if we could solve the age old problem of sales and marketing leadership not playing in the same revenue generation sandbox?

I’ve lived it. The well-intentioned commentary that sales and marketing are one revenue team and everyone is pushing towards the same goals. When, in reality, once the individual staff meetings come up, there are different metrics and different dashboards driving decision making and forecasting.

Then different data points are sent up the chain in update reports and no one really understands the current state of Go-to-Market (GMT) operations.

Previously, you’d assign an analyst to work through the piles of data to produce one common data viewpoint. Most of that person’s time is spent buried in spreadsheets and queries in CRMs

What if, instead, AI could be set up to manage the spreadsheets and data issues and the Revenue Analyst could spend time turning that data into an actual strategy?

The answer is “Hell yes”.

Let’s walk through how that is possible.

Glad you’re here and thank you for reading Approachable AI by Pat Schaber! In case you missed some recent articles, here’s a recap:

Common Sales and Marketing Differing Viewpoints

After the collaborative overtures early in the year, by mid-year this is some of the commentary you’ll actually hear:

"We sourced 2,238 MQLs this quarter."
"Sure, and we closed $9M. What's your point?"
A huge culprit of marketing and sales working off different data sets. Marketing doesn’t have the deal amounts, so they talk in volume. Salesforce holds no first touch history, and they just talk in dollars because that’s what they’re measured in. Same argument every month.

"Unknown" is one of your top revenue sources and everyone has learned to skip that row.
This is a reporting demolisher. The Lead Source makes it from the MQL to the Lead, but when the opportunity is created, most systems are reliant on human intervention to update that field. It typically doesn’t get updated. Therefore, we have an unknown Lead Source on the opportunity.

"We sent you 2,238 leads."
"We never got half of them."
Unfortunately, both viewpoints are probably accurate. Sync errors, required field issues, broken filters, etc. I often find these syncs are set up by IT and don’t reflect the actual needs of the sync from a sales and marketing standpoint.

“Sales sits on our leads.”
“Marketing sends us garbage.”
Neither claim has ever really sized or proven. This is typically an MQL definition problem and the urgency to fix it is never really there.

We now have a fix at our fingertips to end this back and forth once and for all!

Building a Sales and Marketing GTM Dashboard - One Source of Truth

The AI dashboard solution I’m showing here does not need a data lake or warehouse, no BI licenses, no admin rights, and no blocking an analysts calendar for two weeks a month.

This is where the power of AI needs to be utilized.

I created my dream dashboard that I wish I could have had during all the years I ran go-to-market teams.

Northpath Software is a fake company I made up. I had my friend Claude create a year’s worth of sales and marketing data that would be common outputs from Salesforce (Sales home base) and Hubspot (Marketing home base). It was freakishly realistic. Wow.

I wanted the fake data to surface common data reconciliation issues between sales and marketing. And, it’s common for the data to be coming from two different platforms like Salesforce and Hubspot so I ran with that angle.

I also wanted it to show the actual data issues and give a roadmap to fixing those in the various sources of data.

CLICK HERE TO SEE THE LIVE DASHBOARD

The dashboard is a live Claude Artifact which is a fantastic way to share dashboards like this amongst a team.

Open up the dashboard and browse through it. There are interactive features so feel free to play.

This could easily be customized to any data you want to track as a revenue team. This is a baseline to give you an idea of what is possible.

My favorite is the gap piece about midway down. You can click on one of the eight common gaps in the data and see the results, and then click on one of the results on the right. You’ll be able to see the recommendation on how to fix it.

What this means is you have a roadmap to fixing all the data issues between your two outputs, rather than having to hunt for all the changes that need to be made. This just gives you an easy roadmap to get it done.

How to Build the Dashboard

I know many of you reading this may have AI connection restriction internally or may not have sophisticated Salesforce and HubSpot setups. So this dashboard wasn’t created by connecting these data sources live to Claude (although that is very much possible). This was created using CSV exports from both systems.

How I built this:

• Pulled two exports: HubSpot contacts, and Salesforce leads plus opportunities. Read access only and no admin rights needed.

• Dropped both CSVs in a folder and opened the folder in Claude Cowork (or Claude Code).

• Point Claude at your column names once, in a small config file. This is the only setup step. It helps orient Claude to the data it’s working with.

• Ask it to match the records: email first, then name plus company domain to catch the ones where the email differs.

• Ask it to bucket every disagreement by cause, and to use only what is in the two files.

• It writes a single self-contained HTML file. Open it, or publish it as a link to share (that’s what mine is).

This, of course, means it’s updated manually by doing the CSV export and import. If you’re in an environment where it’s possible to connect Claude to these data sources, this can be something that updates regularly as the data in HubSpot or Salesforce changes in real time.

If you follow a similar methodology to my sales and marketing OS in Claude, you can keep a folder in the workspace with the historical exports so you can start to measure over time.

For paid subscribers, below I have the full setup, including the scripts that build the dashboard, but you can also get started without all that if you are not a paid subscriber. This prompt below will help get you started working with Claude to build out a similar concept.

Example Prompt to Get Started

Here are two CSV exports, one of HubSpot contacts and one of Salesforce leads and opportunities, covering the same twelve months. Match them on email first, then on name plus company domain to catch the records where the email differs, then collapse duplicate HubSpot contacts. Tell me how much closed won revenue traces back to a HubSpot MQL, how much of it Salesforce credits to a marketing lead source, and bucket every record where the two systems disagree by the reason they disagree. Only use what is in the two files.

Paid Subscriber Feature: The Dashboard Kit

A few times a month, I build resources for my paid subscribers that can help them implement some of the strategies that I write about. This weeks resource:

The Sales and Marketing GTM Dashboard

While my examples above ran on a fake company, this kit allows you to run it on yours.

Drop in two CSV exports, point it at your column names once, and you get the same dashboard: your funnel, your gap in dollars, your records bucketed by the eight causes.

Inside: the reconciliation skill, the exact data export field lists, all eight causes with a named fix for each, and the dashboard template.

It ships with the Northpath data preloaded, so it rebuilds this issue’s dashboard on your machine in four seconds before you touch your own.

No warehouse, no BI license, no admin rights. It never connects to either system and never writes anything to either system (an IT dream).

Read more

I write a lot about marketing and sales AI strategies in this newsletter, but I know I have a broader audience that reads each week.

So when Leszek Wisniewski reached out and asked if I’d be interested in a joint article. I jumped at it!

Leszek spent 15 years in engineering and leadership roles in tech. He now works with department heads on making AI adoption measurable. He writes the Agentic Manager Substack newsletter.

I think it’s important to understand how different groups across a company implement and adopt AI. There is a lot that can be learned cross-functionally here.

Not all AI rollouts are equal. Whether you’re in finance, human resources, sales, marketing, or operations, you can gain and learn something from the wins and losses of an engineering rollout.

From Leszek with some commentary from me mixed in

You’re still thinking about mandating an org-wide AI rollout – think again.

I have scaled engineering teams from 3 to 20 people at a high-tech startup selling glasses capable of understanding the world. I don’t have a PhD, and I’m by far not the best engineer.

You’re still thinking about mandating an org-wide AI rollout.

Think again.

What I learned growing that team is the same thing I’m learning now with AI: adoption is a cultural process. You can’t mandate culture - you grow it, one person, one habit, one small win at a time.

I’ve lived through what works and what doesn’t. Both lists below.

Quick note from Pat

I can relate to this from the customers I work with. I’ve yet to see a top-down approach that gains traction. Typically, it begins with something simple and small that grows into a culture of AI adoption.

What works

1. Spend the saved time on your people

I automated my weekly status report with Claude Code. That bought back two hours, every week.

Seeing the gains from agentic AI, I didn’t reinvest those two hours into more output. They went into honing my coaching skills and into networking inside my own company - the relationships that make a leader sound.

Volatile markets and AI have made ambiguity the norm for every leader. So this is the best moment to improve your leadership with AI - it hands you second opinions you never had access to before. A leader living the AI adoption, not only mandating it, creates an environment of “togetherness” in the entire team.

Regarding the status update report - yes, it still matters, but its value is in leadership receiving accurate updates, not in me compiling it by hand.

When your people see saved time turn into better conversations instead of a bigger to-do list, they want in. Take that with said “togetherness” - and that’s how adoption spreads.

2. Automate one small task someone knows inside-out

My day-to-day work with Google means that the mixed reality glasses that we develop (kind of like Meta Ray-Bans) are subject to rigorous, automated quality testing that we call “hillclimb”.

One of the most senior engineers on my team computed the same metrics every day, by hand - numbers tracking an algorithm he was working on.

No metrics means no product inclusion. The features that he was working on could’ve been rejected had he not found a quick way to generate them consistently, so that external stakeholders like product managers could inspect them.

He handed the job to an agent that runs unsupervised, with no one watching over its shoulder. The instructions were extremely well-articulated, because he knew the process inside-out. There was little creativity involved, because he knew exactly what kind of tools and data integrations needed to be run.

That’s the pattern:

• Pick a small, repeatable task you could execute blindfolded;

• Compartmentalize the data the agent needs;

• Make sure its quality meets the bar, and make it available.

A call transcript, for example, is near-ground-truth data - with it, you have a significant part of the context for the agentic execution. What’s usually missing is the additional, high quality data to complement the agentic understanding.

3. Let agents sharpen judgment, not just throughput

Better judgement in management means:

• Knowing how to grow talent to be more versatile than “just fast”;

• Deepening the insight into your employees’ motivations to lead more efficiently;

• Selecting the right arguments to influence stakeholders more convincingly;

• Having a sound prioritization function when duties start stacking and overlapping.

If your teams see you going all in on a new technology and actually using it, instead of promoting it in pitch decks and stopping there, they will also start believing in it.

Back to the previous point on automating metrics for product inclusion: the same engineer later used an agent to help craft the pitch to the executive who owns that area of the business.

He had seen how I use agentic skills to elevate my level of articulation, showing the needed amount of empathy when our team was under fire - and he responded to it. What was especially rewarding for me is that he took the skill, modified it and applied it to meet his goal.

I was astounded. I had been coaching him on alignment and communication, and here he was, taking a creative approach to build exactly the right message for exactly the right audience.

That is adoption at its best - a tool amplifying a skill someone was already building.

Subscriber to Leszek's Newsletter

What doesn’t

1. Mandating AI use without a goal

Adopting AI is like managing the performance of a team - it hinges on goal-setting.

My team was part of an organization-wide AI adoption wave. The mandate was seemingly simple: “we will provide you with the tools for programming, and you should try to use them as often as possible”.

On the ground, we struggled with obtaining permissions to run the models, and ran into numerous outages associated with rollouts of bleeding-edge versions.

Goal-setting was missing, which in the end produced fatigue in parts of my team. Frustration first, then flat-out resistance to any deeper adoption. And it made sense, because if the AI adoption didn’t benefit my team, then why should we commit to it?

If I could go back in time, I’d ask myself and the team the following questions:

• With coding done by the agent, what would we want to do with the time won from that?

• Imagine ourselves six months from now, and the AI adoption has worked very well. What will have happened to make that a success?

• How do we flag errors in code brought on by AI, so that we don’t make ourselves vulnerable to outages - and, through that, a drop in trust and satisfaction among our customers?

That would’ve framed AI adoption as an evolution of the team’s cultural process - something that every well-managed team is familiar with.

If you can’t state the expected outcome of the mandate in one sentence, your team can’t either.

2. No time for AI experimentation

We managers love getting more productivity within the same budget.

I understand productivity gain as the capacity freed by automating existing high-effort, manual workflows. In order to consider automation complete, middle managers need to guarantee consistent, high-quality outputs, especially when using novel technology. In my experience, it’s not a straightforward challenge.

The industry confirms my findings - a recent Bain & Company survey of 951 companies found that only 7% of them run fully autonomous agents in production. Let me tell you why it’s not easy to accomplish reliable automation.

The weekly status report I mentioned earlier took me eight (!) hours to produce consistently good outputs. I needed to scrutinize it and fine-tune it on every instance it failed. But when I got it right, the time savings were significant - my peers adopted the skill, multiplying the “two hours” saved across the organization.

Unless the staff is allowed ample experimentation time, it will be impossible to learn how to work with agents to replace existing workflows with a high degree of confidence and get the expected productivity gains.

3. Lack of strategy

A working multi-department roadmap is a product of a good strategy.

A good strategy starts with thoroughly answering the following questions:

• “What is the challenge that we are facing?”

• “How do we use our advantage to overcome it?”

• “What concrete actions are we going to take to accomplish that?”

A top-down mandate usually fails because AI is treated as the “latest” tech to get in on, rather than an answer to the core challenges of the business.

Uber burned through its entire 2026 AI budget in four months - engineers were ranked on internal leaderboards by how much Claude Code they used (Fortune). Microsoft cancelled Claude Code licenses outright when the bill spiraled.

Both had the budget and the talent. What was missing? The answers to the three questions above.

So before you mandate the rollout, write the strategy it deserves. Then grow it the way you’d grow a team - one person, one habit, one small win at a time.

Quick note from Pat

I found some of the best success with multi-department rollouts are when you can identify the hand-offs between the two groups and when both groups are sharing the same context and knowledge in their AI work. Understanding how the teams are connected and how AI can help bridge the gap is a very quick win.

Bottom line - roll out AI with the following in mind:

• one automation goal for a high-effort, manual task;

• one person who knows it cold;

• one visible win where the saved time goes back into people.

Which task on your team does someone know inside-out and still do by hand every week - and who is that someone?

Subscriber to Leszek's Newsletter

What I’m covering in this newsletter:

• Setting up basic AI global memory in ChatGPT and Claude

• Using Projects as a next level of bringing memory to your work with AI

• A third level of AI memory that puts you in charge and gives you a true AI working partner

• Bonus for paid subscribers: a “MemoryOS Kit” to help you get started creating a workspace so you build your own AI memory (third level).

Let’s talk about AI memory.

I think most teams and companies are beyond the “should we test out AI” and they’re more into how to make it actually work for them.

One nagging question and comments that I hear often:

• “Why does it seem like I have to start over every time?”

• “With as much time as I spend re-explaining, I might as well just go and do the task myself.”

• “Why can’t I just pick up where I left off in a different chat?”

All great questions, and we’re going to spend this newsletter giving you three answers to those concerns.

I’m going to outline three layers of AI memory options available in your AI tool. We’ll go from simple to more complex.

AND it doesn’t matter what tool you’re using. Claude and ChatGPT both offer this same functionality.

Let’s dive in!

Glad you’re here and thank you for reading Approachable AI by Pat Schaber! In case you missed some recent articles, here’s a recap:

1. Global memory across all chats and why it’s vague (on purpose)

For a basic account setup in either ChatGPT or Claude, my first stop would be the global instructions or global memory.

These are not going to help you pick up where you left off yesterday in a chat. They are global instructions that will apply to every single chat that you have on the platform.

You want things in here that are true everywhere. If you load up the memory here with information on a singular project you’re working on this week, it will apply that to every single chat that you have moving forward, whether it’s about that project or a recipe you want to make for dinner this evening.

Since I mainly use Claude and ChatGPT for my business, I have things in here like my brand colors, my general positioning of my business, and high-level voice instructions.

For instance, if you never want your writing to have em dashes or negative parallelism that AI is so famous for, this is where you could put that. This would apply towards ANY writing drafts AI does for you.

Here is where you can find these settings in both Claude and ChatGPT.

Pro Tip: If there is something you realize you want your AI tool to remember while you’re in the middle of a chat, simply tell it you’d like to update your global memory instructions to remember that, and it will add that to that file. But be careful here, because what you just added will apply to every single thing you do in the tool moving forward

ChatGPT Global Instructions

Claude Global Instructions

2. Project level memory and what to watch out for

Projects are available in both ChatGPT and Claude and are a great way to start taking more advantage of memory in your AI tool of choice.

Projects draw a boundary around one workstream or task. They’re meant to do one thing very well. Think of this like a repeatable prompt you don’t have to keep rewriting, along with context you don’t have to spell out every time.

Also important to note that the global memory you set up previously will also flow into every chat you do within a Project.

Here’s what to watch out for. Say, for instance, you have a document that spells out who your ideal customer profile (ICP) is, and you create a Project and upload that as context. That’s perfect for that point in time.

But in your 2027 strategy meeting your company updates details of your ICP. Now you need to go back through all of your Projects and make sure you update that ICP context document. Otherwise, you’ll be working in a Project with outdated information.

Here are some very brief, high-level instructions on how to get started with projects.

ChatGPT Projects

Claude Projects

3. Memory you own and manage making AI a true working partner

Do you ever wish you could open up Claude or ChatGPT and type in:

“Let’s continue on <project>. Remind me where we’re at so we can keep going.”

And all of a sudden it just knows what to do and you continue working on the project.

Or

“Let’s update Q2 financials and get a view of how the business is doing.”

And you get a beautiful chart showing exactly where you are.

Believe it or not, this is not that far off for you. Of the three stages we’ve laid out, this takes the most work, but it sets you up the best to have a true AI working partner. When I say more work, I’m talking about maybe an afternoon to get started.

Basically we’re talking about plain text files in a folder on your computer or in your Drive. Your assistant reads them at the start of a session and updates them at the end, and you can open any one of them in any editor and see exactly what it sees (and edit it too).

Both ChatGPT and Claude now ship something that can do this. Claude Cowork and Claude Code from Anthropic and ChatGPT's Work and Codex from OpenAI. I run mine in Cowork and Claude Code with folders backed up in iCloud and Google Drive.

Below is a snapshot of what my personal operating system looks like. Every session starts by asking the Conductor (CLAUDE.md) where in the workspace it should go. The Conductor points in the right direction and says, “Before getting started read the context files and the memory files.”

Then, we pick up where we left off.

When I’m done with a session, I let Claude know, and Claude scans the conversation for decisions, progress, corrections, and anything worth keeping. Then it asks for my approval of what it thinks it should keep, and then it goes and updates the files.

And guess what? You don’t need to know a single line of code or anything special to get this started.

Think of it this way: Levels 1 and 2 reside within a vendor’s ecosystem.

Level three lives in files, which means you can read, correct, edit, or use those files to move to a new platform next year if a better one comes out.

Anywhere you start from Level 1 to Level 3 is good progress. These should give you options to implement memory at whatever stage of AI adoption you’re at.

For paid subscribers, I have a kit (below) they can download and start using right now. If you’d rather have a system like this built for you, click on the box below for more details.

Paid Subscriber Feature - the MemoryOS Kit

A few times a month, I build resources for my paid subscribers that can help them implement some of the strategies that I write about. This weeks resource:

The MemoryOS Kit

I packaged the whole thing up so you can get started.

MemoryOS is the starter kit for level three. It’s basically the same structure I run, stripped down to the parts that are easy to transfer:

1. The routing file (CLAUDE.md). The one small file that loads first and keeps everything else out of the way.

2. The index and project templates. Where active work lives and how a project folder is laid out.

3. The memory taxonomy. Four types (decisions, preferences, project context, references) so things land somewhere findable instead of in one giant notes file.

4. Five prompt examples. Cold start, the wrap-up, correcting the record, promoting a correction to a rule, and a quarterly audit that catches contradictions before they cause problems.

It also ships with an example project already filled in so you can watch the system run before you type anything.

There’s a one-page Quick Start in the folder and I’d estimate setup to be about twenty minutes.

Read more

What I’m covering in this newsletter:

• How to start recovering prospects that were previously closed-lost opportunities

• Taking large Total Addressable Markets and breaking them down to actionable call lists and intelligence for sales and marketing teams.

• Better prediction engines for when current customers are ready to expand their buying with your company.

• Bonus for paid subscribers: a “Lost Opportunity Recovery Kit” to help you get started bringing back those prospects.

I’m loving some of the trends I’m seeing right now with customers and their thought process around how to use AI in sales and marketing.

I’m seeing teams start to think in multiple ways about the use of AI.

On one hand, teams are wanting to figure out how to use tools like Claude or ChatGPT to augment their current work and to make their processes more efficient and scalable.

On the other hand, there’s more of a thought process around how to use AI to accomplish big ideas and initiatives that used to take months and months of research but can be done quickly now with this technology.

Both are equally important, But it is really fun to see people start to think outside the box about ways to use AI in Go-To-Market (GTM) initiatives.

I wanted to take this newsletter and outline a couple of my new favorite ways that I’ve been working with customers to put AI Agents, data and automations to work on producing big outcomes contributing towards larger GTM goals and objectives.

Let’s dive in!

Glad you’re here and thank you for reading Approachable AI by Pat Schaber! In case you missed some recent articles, here’s a recap:

Lost Opportunity Follow-up

Sitting buried in CRMs and recorded call transcripts is a treasure trove of data that can give teams much-needed insight on opportunities that were lost and what a strategy might be to follow up with them again.

Historically, I would have asked a product marketing team in one of my organizations to dive into lost opportunities for the product lines, figure out what trends were there, and start to prep a list that we can work with sales on for follow-up.

This would have involved mass export of data from the CRM and days to figure out how to make sense of the data in pivot tables.

Then, analysis by listening to a few calls because there’s not enough time in the day to listen to all of them, and then research on anything related to the customer in the lost opportunity.

Wow, that seems like such an ancient way to do it now. Let’s look at a better way.

Building a Lost Opportunity Follow-up Engine

I told someone on a call the other day that I’m having the most fun in AI right now building out solutions for things that caused problems for my teams in the past.

One of the strengths of AI is its ability to analyze and crunch extremely large amounts of data very quickly. This is a prime example. By feeding it the close lost opportunity information and possibly call recordings, an AI machine can be built that takes care of all of the analysis and building a reactivation campaign.

The graphic speaks for itself, so I won’t reiterate a bunch of it. If you’ve already built out your company context and the AI machine knows what buying signals look like for your ideal customer profiles, then it’s just a matter of having AI analyze all that information, communicate with the buying signal and the context information, and then kick out the list that preps your salespeople to make valuable calls.

Large TAM Broken Down to Actual Call Lists

I’ve been that guy. The guy that stands up in front of the organization and touts the size of the total addressable market. If we can just get 5% of that market, the company will exponentially grow!

There’s truth in what’s presented, but there is also a reality in how hard it is to align sales and marketing to get that 5%.

This is one of those problems I want to fix.

The problem, historically, is that there is just not an easy way to break down all the companies in that TAM into companies that fit your ideal customer profile, and then further take that down to companies that are ready to buy, with actual call lists of who to contact.

NOT ANYMORE!

The Large Market to Call List Machine

The last thing you want is your salespeople spending hours researching companies in that addressable market that may be interested in buying your product. Their value comes in working the sales process and engaging with the customers, so let’s give them a head start on that

These AI machines are fun to build. They’re multi-step, multi-agent AI machines that

• Go out and do the deep analysis

• Do the research,

• Compare it to your ideal customer profile

• Then enrich, get contact information, and prep the salespeople to call

This happens in days and not months.

Predicting Upsell and Cross-sell for Current Customer Opportunities

Last but not least, I want to talk about a better prediction model for selling to current customers.

I’ve seen many Customer Success groups find out in quarterly business reviews (QBRs) that a customer is ready to churn or that they’ve been ready for a new product offering for a while. When an Account Manager is managing so many accounts, it’s hard to spend time outside of those QBRs anticipating some of these circumstances.

This is where AI can fill a gap in most customer management processes. This is a gap that most teams just don’t have the time for, and where AI can jump in and give those customer success teams more of an opportunity to have a greater impact on the growth of a client.

The Customer Expansion Machine

I believe there are hidden opportunities all the time with current clients, but the systems just aren’t always there to offer the signals up that someone is ready to buy an additional product or expand what they’ve already purchased.

Yes, SaaS businesses and businesses that primarily sell through a website have an advantage in picking up on some of these signals, but not every B2B company out there has that opportunity.

An AI machine like this can be continuously running and offer the customer reps opportunities between those QBRs to pick up on chances to sell additional product or even anticipate when someone’s about to churn.

Paid Subscriber Feature

A few times a month, I build resources for my paid subscribers that can help them implement some of the strategies that I write about. This weeks resource:

The Lost Opportunity Recover Kit

The Lost Opportunity Recovery Kit scans your closed-lost sales opportunities and scores each one on whether now is actually a good time to reach back out, based on real signals like a champion changing jobs, new funding, a competitor stumble, or enough time simply passing.

For every deal worth pursuing, it drafts a reactivation email that opens with the specific reason it's worth revisiting instead of a generic check-in. It works with whatever you already have on hand, no CRM export cleanup required, and comes in two forms: a copy-paste prompt for a one-time run, or a skill that plugs into your GTM OS and personalizes the output to your actual business and voice.

Read more

What I’m covering in this newsletter:

• What a connector actually is, and what “allow” really grants

• Where your data goes the moment you start using one

• The model training question, and what changes between the consumer and business versions

• Who else at your company can see what, once it’s connected

• Bonus for free and paid subscribers: my “Before You Connect” checklist, an AI data security list of actions before making any connections.

Quick story to start:

I’m a dedicated HubSpot user, and I rely heavily on Tasks in HubSpot to remind me whom to follow up with and when. Whether it’s an open opportunity or a networking contact, tasks are my go-to place to make sure that I follow up when I say I will.

Well, I once accidentally deleted 85 tasks through the Claude to Hubspot connection. It wasn’t Claude’s fault. It was mine for not understanding at the time what guardrails should be in place so that doesn’t happen.

That was a rough day!

First, I do AI Consulting and a lot of go-to-market (GTM) operations for a living. I’m not your security officer, and when the data is genuinely sensitive, your IT or security team should be in the room (I make sure they are). The information I’m providing here is a great overview for that discussion with your team.

Here’s the honest starting point. Connecting an AI tool to your data is normal. Companies do it safely every day. The click that deserves your attention is the one labeled “allow,” because it carries the same weight as handing any vendor access to your systems. Give it that level of care and most of the worry goes away.

Glad you’re here and thank you for reading Approachable AI by Pat Schaber! In case you missed some recent articles, here’s a recap:

What a connector actually is

When you connect Claude or ChatGPT to Google Drive, HubSpot, email, etc., you open a permissioned bridge to an account you already have. The AI reaches your data through that bridge. It works from the access you carry, and it can’t step past it.

Here’s the most reassuring part. A connector inherits your permissions. Anthropic puts it plainly: Claude “inherits each person’s permissions from the connected service.” OpenAI works the same way. If you can’t see a file, a channel, or a record in the source system, the AI can’t reach it either.

The “allow” screen is where you set the terms. Good connectors let you decide, per action, whether the AI can read and summarize, or whether it can also write, send, and delete. The default I’d set: let it read and draft, and make it ask before it sends a message or changes a record. (yes, I learned this the hard way with my Hubspot Task issue)

One thing to watch in 2026. Connectors increasingly support “actions,” which means the tool can do things, not only read them. ChatGPT added Google actions in June that can touch Drive files directly. Reading your calendar is low stakes. Acting on your behalf is not. Know which one you’re granting.

In relation to my story above, here’s the HubSpot connector approval screen. There are about 20+ actions or tool permissions that the connection allows, and for each one you can set it to the options in the drop down.

Where your data goes

When you ask a connected AI a question, it retrieves what it needs, uses it, and hands back an answer. Two things are worth knowing.

Retrieval happens on demand, and it stays minimal. Claude’s own documentation says it “only accesses your data when you explicitly ask a question or request an action requiring this information, and retrieves the minimum information needed.” It doesn’t copy your whole Drive into a warehouse somewhere. It pulls the file you asked about, at the moment you ask.

Retrieved business data stays out of model training. Both Anthropic and OpenAI keep connector data out of model training on their business tiers, and Anthropic says directly that it doesn’t train on your Gmail, Drive, Calendar, etc. connector data. There’s one exception, and it’s in the next section.

The model training question

This is what people actually mean when they ask if their data is safe. Is the AI learning from our stuff? It depends entirely on which version you use, and the gap between consumer and business is wide.

On the consumer plans, the individual carries the burden.

• ChatGPT Free, Plus, and Pro train on your conversations by default. You turn it off in Settings, under Data Controls, by switching off “Improve the model for everyone.”

• Claude Free, Pro, and Max changed this in August 2025. Your chats can now be used for training and kept for up to five years when the setting is on. Anthropic presents it as a choice you have to make, and the training option sits on unless you change it.

On the business plans, the company carries it, and the default flips.

• Claude Team, Enterprise, and the API aren’t used to train models by default.

• ChatGPT Business (formerly Team), Enterprise, and Edu aren’t trained on by default either. The API trains only if you deliberately opt in.

Two traps, whatever tier you’re on.

1. Clicking thumbs up or thumbs down on a response can hand that whole conversation over for training, even on plans that otherwise leave your data alone.

2. And on a consumer plan with training on, if you copy and paste content out of a connected tool into the chat, that pasted text can be used to improve the model, even though the raw connector data wouldn’t be.

The short version. On consumer plans, assume you’re training the model until you’ve turned it off. On business plans, assume you’re not, and confirm it in the admin settings.

Who else can see it

Here’s the part leaders most often get backwards. Moving to a business or enterprise plan makes your data more private from the AI vendor. It can also make your data more visible inside your own company.

On the lighter business tiers, Claude Team and ChatGPT Business, admins see usage metrics, adoption, and activity. They don’t get a routine window into individual chats.

On the enterprise tiers, that changes. Both Anthropic and OpenAI offer a compliance capability built for audits and legal discovery. It lets a company’s security or legal team pull actual conversation content and files. That’s the right feature for a regulated company. It also means a work AI account should be treated like work email. If your team is connecting company tools, tell them plainly that their work conversations can be reviewed.

Connectors widen this. Once a shared connector or shared project is in play, what one person pulls in can surface to others in that space. Access still follows each person’s own permissions, and shared workspaces just broaden who sees the result.

One honest caveat, because this area keeps moving. (Yes, everything I’m writing here could have a shelf like of a week. Who knows?)

A deletion promise comes with an asterisk, because a legal hold can override it. For much of 2025, a court order required OpenAI to preserve consumer and API chats that would normally have been deleted. That order was lifted in late September 2025 and OpenAI went back to standard 30-day deletion, though the underlying case is still going as I write this in July 2026. The specifics will keep changing. The habit that protects you won’t: when data handling really matters, read the current policy on the vendor’s own page instead of trusting a summary, including this one.

Try this

Before you trust a connection, spend five minutes on it. Open the permission screen and actually read what you’re granting, with an eye on anything that lets the AI write or send. Then check one setting: on a personal account, confirm whether training is on or off. You can also just ask the tool what it can and can’t reach. The answer tells you how seriously it takes the boundary.

Connecting AI to your data is a normal, safe move when you do it with your eyes open. The tools borrow the access you already have, they pull only what they need, and on a business plan they keep your data out of their training. The click that matters is “allow.” Give it the same few seconds of attention you’d give a contract, and you’ve handled the part most leaders worry about.

“Before You Connect” Checklist for Free and Paid Subscribers

Below is a link to a PDF checklist you can use today that will help guide you through decisions on what to do before you make a connection from your AI tool to your data:

The checklist is only available for subscribers to this newsletter. The good news for you is that it only takes a free subscription to gain access! Give the newsletter a try for a few weeks and if it’s not your thing, you can always unsubscribe.

(but, I’m hoping you’ll stick around)

Subscribe now

AI Tech I’m Currently Using

I get quite a few questions on Substack, LinkedIn, and through my website on what AI tools I use for what use cases. I’ll try to share a few in each newsletter so it will give you some ideas of tools you can try for specific purposes. Here are a few for this week:

Everyday LLM - ChatGPT and Claude: I cover a lot of this above so won’t dive in too far here.

Call and Meeting Transcriptions - Granola: I like Granola for a couple of reasons. It takes the transcription and rolls it up into very good notes and action items but doesn’t need to be added to a meeting as an attendee. Also, it syncs great with my Hubspot CRM which makes it very easy to send notes to the contact record.

Presentations / Branded Documents - Gamma.app: I use this 3-4 times a week for anything I do with client presentations or professional document creation. I cover why I’m a heavy Gamma user in this post:

Video Recording / Editing - Descript: I’m trying to work in time to my weekly routine for video creation. I feel it’s important to give people a more personal connection. Descript is a huge timesaver. Imagine being able to erase or add text from a script and have it be reflected in the actual video in seconds. Unbelievable.

Sales Prospecting - Apollo.io and Clay: I don’t do a ton of outbound prospecting for my business but I’ve been using the free versions of Apollo.io and Clay to find companies that I could get to know through LinkedIn that may be interested in my services. I’m only touching on a sliver of the capabilities of these tools.

What I’m covering in this newsletter:

• Narrowing a total addressable market down to actual companies you can sell to

• Understanding which of those companies are ready to buy

• Finding the titles and messaging that will get you in the door.

• Bonus for Paid Subscribers: The Market Narrowing System added to my Substack Subscriber Library. It includes the four Claude skills that run every step of this walkthrough for you.

How many of you can relate to this story?

It’s the January sales kickoff meeting where marketing and sales sit down for three days to lay out the plan for the year.

At some point during the meeting, one of the executives stands up with a (too) high-level presentation that references a market size that is so incredible that even a 1% gain in the market would mean huge things for the company.

Raise your hand if you’ve been there.

The market size report was either downloaded for free or purchased for a few hundred dollars on a somewhat reputable site. Ugh.

All of a sudden, a 5% gain in that market is the new goal for the year, and everybody has dollar signs flashing in front of them. The executive is patted on the back for uncovering a market that the company should be heading into, and the managers start talking endlessly about the amazing opportunity ahead.

3 months later there is zero progress and the questions start swirling as to why.

The casualty is typically an uncoordinated Go-To-Market (GTM) plan.

With the tools and technology at our disposable today, there is no longer any excuse for this to happen.

Here’s the 4-step process to fix this problem that takes a day. Not the two months of works this used to take.

Glad you’re here and thank you for reading Approachable AI by Pat Schaber! In case you missed some recent articles, here’s a recap:

Narrowing Down the Market

This used to take weeks of research. With Claude Cowork or ChatGPT Work (new) this is an hour of work. If you have your TeamOS set up with the right context and skills, it will be quicker and more accurate. What we don’t want:

❌ a massive list of random contacts that someone purchases going into the CRM

❌ someone purchasing LinkedIn Navigator or ZoomInfo and saying we’re good to go

Start by cutting the market down to the accounts that match how you actually win. The ones that look like the customers you already close, rather than everyone who could theoretically write a check.

We’re going to use AI for this analysis. Here’s what we need.

Closed Won Deal Analysis

Pull your closed-won deals from the CRM and find what they share. Maybe your product lands best at companies between two hundred and eight hundred employees, in services and light manufacturing, that already run a specific ERP. AI does this fast. You give it your win list, it names the firmographic patterns you have been running on instinct.

Won Deal Call Transcripts

The transcripts tell you why they bought: the trigger that started their search, the words they used for the problem, who was in the room, what almost killed the deal. That behavioral layer catches criteria a firmographic filter misses, like a specific event or a phrase that only shows up when you read the conversation.

If you’re connected by MCP to your CRM and call recordings from your AI tool, just ask it to go look for this data. Otherwise, export and upload to Claude or ChatGPT.

Once you have the data, the link below is to a Google Doc that contains the prompt you can use to pull everything together and get a set of criteria you can use in a platform like Clay to start to narrow down your company list.

AI prompt for narrowing down your market criteria

Now you take your set of criteria over to Clay, select that market that has such a huge opportunity, and pair that with the data you just got from your prompt above.

This will take that giant market down to a list of companies that make sense to reach out to, but we’re not done narrowing yet.

Find the Signals That Say “NOW”

Let’s not go using our credits to buy contact information in Clay before we’ve taken this next step.

While a massive list of companies will make everyone feel good, it also has the potential to waste the salespeople’s time, so let’s take that list of companies and let’s figure out who’s ready to buy now.

These are called buying signals and AI is perfect for helping us quickly identify these. Here are examples:

• A new executive in the last ninety days.

• A recent funding round.

• Headcount growth past a threshold.

• Open job postings for certain roles at the company

Try this. Have AI map the buying signals for your market before you go looking for them (throw in your data from Step 1 for more clarity):

“I sell [product] to [buyer]. Below is the firmagraphic data that identifies our ideal customer profile. List the public buying signals that a company is likely to be in the market in the next 90 days. For each one, tell me what it means and where I would find it (news, job boards, filings, LinkedIn). Rank them by how well they predict a real buying window.”

Let’s pretend, for example’s sake, the buying signal are

• Positive headcount growth in the last twelve months

• An executive hire in the last twelve months.

Now we take that data over to clay, where you can see below I’ve prepared an example table of large public companies in the Twin Cities just for this example. But pretend this is a table of 10,000 companies.

I’ve used the AI capabilities within Clay to go out and find headcount from the last twelve months, as well as any executive hires in the last twelve months.

Then, in the column on the far right, if headcount growth is positive and an executive has been hired, it takes that data and sends to the ready to buy tab at the bottom.

Now we’ve taken that massive market number and narrowed it down to companies that are likely to buy what we offer and are showing signs of being ready to buy.

Technically, this would be called a Serviceable Addressable Market (SAM) which basically means it’s a consolidated target within that larger market that you can actually go after and win right now.

Map the Title to the Signal

Next, let’s make sure we’re using our clay credits wisely. We want to only find the contacts that are going to map to the buying signal.

Assuming your LLM of choice knows all the context around you, including your ICP, then a simple prompt like this can help you determine the right titles for the signals to make sure that they align with your context.

AI builds this mapping once and applies it to every account, then finds the actual named person and their contact details for each one.

Try this. Turn your signals into a who-to-call map:

“For each of these signals [list them], tell me the one or two job titles I should reach out to and why that person owns the problem when the signal fires. I sell [product]. Keep it to the titles most likely to actually respond.”

Now we can take this data and the intelligence behind the titles over to Clay and find those contacts for the specific companies that are ready to buy.

Let’s just say, for instance, that the Chief Financial Officer was the main target, and we can simply narrow down in Clay to the companies ready to buy and people holding that title at those companies.

There are, of course, hundreds of other options, but I’m just showing, for simplicity’s sake, what the workflow can look like.

At this point, we’re safe getting more enrichment information and contact information for our targeted contacts at our ready-to-buy companies, which we can do in clay. From there, we can export to our CRM of choice and be ready for our salespeople to start calling.

Match Positioning and Message to the Signal

Now the message gets easy, because the signal already tells you what the person cares about.

AI drafts the positioning for each signal and the message that goes with it, so the rep is not staring at a blank page wondering why this account, why now.

Try this. Let it write the opener for a single signal so you can hear the difference:

“Write a two-sentence opening line for a first message to a [title] at a company that just [signal]. I sell [product]. Speak to what that person cares about because of that signal. No hype, no feature list, just why the timing makes sense for them.”

The Marketing to Sales Deliverable

Stack those four steps and you get the thing marketing should hand to sales. Every row on the list says:

• The company, and why it fits

• The signal that fired, and when

• The person to call, and their contact info

• The one line of why-now that opens the conversation

That is what moves the dial on the market growth initiative. They open the list and there is no guesswork about who to call, why, or what to say. The narrowing already happened. The campaign has somewhere to go.

Every market looks big when you zoom out far enough. The value is in the zoom in, and that work is doable now in a way it was not two years ago. A team of one can do what used to take a research analyst and a month.

If you run marketing, this is the deliverable worth being proud of. A deck full of market size is the easy part. The hard and valuable part is the list that puts your reps in front of the right people, at the right time, with the right thing to say.

The Market Narrowing Claude Kit

Paid subscribers this week get the Market Narrowing System added to my Substack Subscriber Library. It includes the four Claude skills that run every step of this walkthrough for you.

• TAM Narrow turns your closed-won data into a countable list of fit companies

• Signal Map finds the ones ready to buy now

• Signal-to-Title finds the right person at each

• Enriched List Builder hands you a finished call list.

It comes as a drop-in kit with the prompts, a one-page quick start, and setup that takes a few minutes.

Read more

What I’m covering in this newsletter:

• How I create design inspirations with Claude Design and connect to Canva

• Setting up Canva to actually work with Claude (or ChatGPT)

• Building an automation that pairs Claude, Canva, and Asana to get real work done.

Before I jump in, let’s set some context.

I’m not a graphic designer and I’ve always be in awe of what true graphic design talent in my marketing organizations over the years have produced.

I’m going to share an AI plus Canva workflow I’ve been working on, but in no way am I suggesting that AI and/or Canva replaces graphic design talent. No way.

But I’m a solopreneur. I also work with a lot of small marketing teams without dedicated graphic design talent, and Canva paired with AI workflows is a game-changer for quick graphics.

What I’m gonna share in this newsletter is a workflow I’ve been working on that helps me speed up and scale the content and graphic creation in my content processes.

Let’s dive in!

Glad you’re here and thank you for reading Approachable AI by Pat Schaber! In case you missed some recent articles, here’s a recap:

Let’s Start at the End

Before I dive into how to create this workflow, I want to show you the end result of how the workflow actually operates.

For this workflow, here are the tools I’m using:

• Claude Cowork: the operator of the workflow

• Asana: where I manage my content calendar

• Canva: where my graphics are created from my Brand Templates

I also touch on Claude Design below which is a new part of my process.

In the graphic below would be a typical chat that I put into Claude Cowork.

You can see I’ve selected my PatOS, which has all the information Claude Cowork needs to adapt my request to my business and point of view.

I then move on and work on other projects while Claude gets to work. A few minutes later, I’ll have this sitting in my Asana content calendar.

What I’ll have ready for me in Asana:

• A link to the Canva design with my LinkedIn Post Carousel

• Suggested hooks and angles based on my ICP and Point-of-View

• 3 variations of a LinkedIn post

I do all of my own writing for the most part, but I do lean heavily on the suggestions that Claude puts out there. It’s gotten to know my tone of voice and point of view so well that it does come pretty close to putting out something that I would post.

But I like the writing process, and I like to have things sound very much like me, so I end up considering all the suggestions and then writing my own post.

Now that you’ve seen the end result, let’s walk through how I built this.

Claude Design for Inspiration

As I mentioned above, I’m not a graphic designer, and I don’t want to put my energy towards that right now. We need to create some designs that we can load into Canva to use for our posts so I head to Claude Design.

I’ve never been that impressed with Canva’s AI tool, so I don’t even try that one anymore.

I give Claude Design an example of something I’ve used in the past, and I ask it to create some variations.

I have gone through the process of setting up a design system for my business in Claude design, so I don’t need to explain a lot about what everything should look like.

After some back and forth, I have a grouping of simple LinkedIn post designs.

Send Designs to Canva

I have my Canva account connected to Claude Design, so I can simply have Claude Design send these images to Canva.

A few seconds later, those Claude Design images are now sitting in Canva.

Save Designs as Canva Brand Templates

In order for Claude to be able to use Canva effectively, you’ll want to set up brand templates. To be able to have brand templates and everything else related to your brand design in Canva, you’ll need a paid account. This is a no-brainer spend in my opinion.

After making a couple small tweaks to each image sent over from Claude Design, I save each image as a brand template.

When I’m all done, you’ll see I have LinkedIn post image templates for single images and LinkedIn carousel post images.

Give Claude Context and Instructions

A key part of this whole workflow is, of course, my PatOS, which is my Claude operating system. Claude already has all the information it needs to know what my point of view will be and what to recommend from a LinkedIn post and graphic.

Plus, it has a LinkedIn Post skill and LinkedIn Post Image skill that work together to complete the task

The skills will:

• Research the topic and draft post variations with recommendations for hooks and angles (in my voice, style, and tone)

• Find my brand template in Canva and make a copy for the post design

• Will use main themes from the content to create the images

• Rename the image in a format I’ve given it and then put that design in the right Canva folder to make finding it easier in the future

• Create the Asana Task in the Content Calendar and place the drafts and a link to the design in the Description field of the Task

• Schedule it for the day I tell it to.

Now, when I have an idea for a LinkedIn post, I just give it a prompt like the one below in Claude Cowork and it gets to work.

Once in Asana this will also now populate in my Claude Dashboard where I manage my daily work.

The whole idea here is to remove things from my workflow that either I’m not great at or I don’t want to be spending the time on right now. In my opinion, this is one of the main functions of AI right now in business workflows.

I’m spending more time on evaluating what types of content my ideal customer profile would like to read and what additional types of content I can provide that give value to my readers.

AI Tech I’m Currently Using

I get quite a few questions on Substack, LinkedIn, and through my website on what AI tools I use for what use cases. I’ll try to share a few in each newsletter so it will give you some ideas of tools you can try for specific purposes. Here are a few for this week:

Everyday LLM - ChatGPT and Claude: I cover a lot of this above so won’t dive in too far here.

Call and Meeting Transcriptions - Granola: I like Granola for a couple of reasons. It takes the transcription and rolls it up into very good notes and action items but doesn’t need to be added to a meeting as an attendee. Also, it syncs great with my Hubspot CRM which makes it very easy to send notes to the contact record.

Presentations / Branded Documents - Gamma.app: I use this 3-4 times a week for anything I do with client presentations or professional document creation. I cover why I’m a heavy Gamma user in this post:

Video Recording / Editing - Descript: I’m trying to work in time to my weekly routine for video creation. I feel it’s important to give people a more personal connection. Descript is a huge timesaver. Imagine being able to erase or add text from a script and have it be reflected in the actual video in seconds. Unbelievable.

Sales Prospecting - Apollo.io and Clay: I don’t do a ton of outbound prospecting for my business but I’ve been using the free versions of Apollo.io and Clay to find companies that I could get to know through LinkedIn that may be interested in my services. I’m only touching on a sliver of the capabilities of these tools.

What I’m covering in this newsletter:

• Why call transcripts and AI are such a powerful pairing

• 18 best uses of call transcripts paired with AI

• 3 detailed use cases to put into action right now

• Starter bundle of call recording Skills for all free and paid subscribers to download.

• Bonus for Paid Subscribers: I built an 18-skill Claude plug-in and individual skill files for all 18 use cases shown below for paid subscribers to access

I spoke to someone recently that blocks 3 to 4 hours a week to listen to sales call and client review recordings.

And they find it so valuable they wish they could spend more time.

3-4 hours.

When I talk about slam-dunk use cases for AI, call recordings are the lowest of the lowest-hanging fruit.

I don’t know who has the time anymore to sit at a desk and listen to three to four hours of calls. I’m going to make the argument in this newsletter that your time can be better spent in other places when there are actually better ways to gather information from these calls.

First, if you’re not recording prospect or customer calls, you’re missing an opportunity to gain extremely valuable data at a minimal cost. But too many people sign up for call recording software, check the box, and then the information sits in a folder untouched.

DON’T WASTE THIS DATA!

Tools like Granola, Fathom, Fireflies, Otter, Zoom, tldv are fantastic and all have one-click cloud connectors to Claude, ChatGPT, etc. making it very easy to use the transcripts in many ways.

In the examples I outlined below, I’m using Granola, which is my favorite right now.

Glad you’re here and thank you for reading Approachable AI by Pat Schaber! In case you missed some recent articles, here’s a recap:

18 Real Examples of How to Use Call Recordings

Before we jump into a couple use cases and how to actually hook this stuff up, let’s get your mind spinning with 18 ideas below of how to use call transcripts paired with the power of AI.

These are just the tip of the iceberg. There’s more out there once you start thinking about it.

Okay, now let’s see some practical use cases in action.

First, the part that used to be hard

Getting a transcript in front of AI used to mean copy-paste or a real integration (think IT admin work). That is the part that stopped people. It is not hard anymore.

Most recorders now have a cloud connector. Granola, Fathom, Fireflies, Otter, Zoom, and tl;dv all offer one. You connect it once inside Claude or ChatGPT, sign in, and you are done. No API keys, no Zapier, no code. After that you just ask for a call by name and the transcript comes to you.

Example: “Can you refresh my memory on my call with ABC Corp last week?”

Full recap in seconds.

If you only do one setup step from this issue, do that one. Everything below assumes it is done.

The examples below start from easy to a bit more complex.

Turning Calls into a Follow-up Email Draft

What it does: takes the call you just finished and writes a draft of the follow-up email. The “next steps” are real because they came from the call, not a template.

The flow: connect your recorder, then ask for the call and the draft in one go.

The prompt:

Pull my call with Acme from today in Granola. Draft a follow-up email in my voice. Open with the one outcome they care about most, list the next steps we agreed on with owners and dates, and keep it under 150 words. Plain, direct, no filler.

What you get back: a send-ready draft that references what was actually said. Read it, fix a line or two, send. The first time you do this and the email is already 90 percent right, the habit sticks.

Tip: make this part of your AI operating system that already includes your voice, tone, positioning, messaging, et cetera. The better the context, the closer the draft will be to what you want to send.

(if interested, have a starter AI operating system kit built for you 👉)

Turning Calls into Actual Task Lists (and taking action)

What it does: turns the commitments made on a call into tasks with owners and dates, so nothing falls through between the conversation and your to-do list.

The flow: connect your recorder, pull the call, ask for the action items.

The prompt:

Pull my call with Acme from today. Turn every commitment and action item into a task list. For each one give me the owner, the due date if a date was mentioned, and one line of context so I remember why it matters.

What you get back: a clean list of who owns what by when. Drop it into your project tool or your CRM. The calls where you said “I’ll send that over” and then forgot stop happening.

Tip: if your CRM has a connector too, ask it to log the tasks and update the deal in the same step.

I use Asana for all of my project management and tasks, so my action item lists automatically flow into Asana and create tasks with due dates and any associated notes that I’ll need to reference later. This will then flow into my Daily Work Dashboard.

Turning Calls into an Insights Database

This is my favorite. By far.

I’ve written about how to make use of data with AI previously so now you can start building that data.

What it does: It creates a database of valuable information that can be used across your organization well into the future. Examples of what can be plucked from each call:

• Competitor mentions

• Price or product objections

• Feature requests

• Possible testimonials

• Ideas for new content

• Positive or negative sentiment towards product or brand

• Etc.

Sure, there can be some value in a single run of this for a single call, but imagine the data and insights gathered over fifty to a hundred calls?

When you’re connecting through ChatGPT or Claude, a third step in this can be the population of an Airtable, Google Sheets, or Supabase database, where this information will compile over time.

3-Set Call Recording Skill Bundle

I’ve bundled up the three skills I just described here in a zip file that you can access from the Google Drive folder below. This is for all free and paid subscribers. there is a readme file included that will show you how to install the skills into your Claude account.

Access the zip file here

(The link is to a standard Google Drive folder. You can download the zip file from there)

If you want to get started with an operating system that makes automations like this more effective, check out my GTM Starter Kit.

18 Skill Plugin and Skill Pack

I took this a step further and created an 18-skill plugin and skill pack of the 18 use cases in the image above.

If you’re on the paid version of Claude you can simply upload the plugin and customize for your uses and environment.

I also did a simple Skill pack of the 18 in case you’re on the free version and can only upload Skills.

👇 Both of these are now located in my AI Resource Library. 👇

Read more

What I’m covering in this newsletter:

• The 4 AI adoption phases that lead up to the “Operational Layer”

• What an AI Operational Layer looks like and what it means for how you work.

• How to get an AI Starter System

• Bonus for Paid Subscribers: I’m rolling out new Paid Tiers and new member benefits including an AI Resource Library and details on how to get a custom-built AI Start Kit

I’m going to take us on a little journey and evolution for this newsletter that will help address a couple questions I’m hearing often lately.

The questions are in response to some phrasing and lingo that I use often in my content here on Substack and LinkedIn. I’m also hearing the questions on customer and prospect calls.

I attribute this to people getting more curious to better understand how to dive a little deeper into their AI use.

The questions:

• What do you mean by there being an AI operational layer to my work or my team’s work?

• What do you mean by AI should be folded in so it’s invisible?

When individuals and teams are truly implementing AI, there is an invisible operational layer that gets created that makes AI almost invisible. It’s just part of the processes that drive the work of a team, individual, or group.

But, what does that actually look like?

In this newsletter I’m going to walk through the phases of AI adoption that lead up to that operational layer. At the end you’ll have a clearer picture of what that means and how you can get there.

Glad you’re here and thank you for reading Approachable AI by Pat Schaber! In case you missed some recent articles, here’s a recap:

We’re going to be laying out a very simple and practical example of someone wanting to use AI for content creation. This is a common use case that is applicable at all phases of AI adoption and across all platforms.

The Paste and Pray Phase

Many of us started here and there is a good percentage of AI users are still at this phase (and that’s okay).

“Can you give me a blog post about (topic)”

It can be frustrating because the output sounds like everyone else and it takes a while to get it right with your AI tool because it sounds nothing like you or your company.

You might as well have just written it from scratch.

Capturing Your Voice

You grew tired of going back and forth with ChatGPT or Claude until it actually got close to sounding like you….and then you probably edited it again.

So, you started your “context” journey and let the LLM know what you expect from voice and tone in its output.

Yes, you’re still explaining in great detail what you want from the content output but this is starting to sound less like AI and more like you or your company’s brand.

Context Lives in the Skill or GPT

You’re starting to see the power of AI.

You’ve discovered a CustomGPT, Gemini Gem, or Claude Project where you can repurpose your prompts and tie in a deeper layer of intelligence and context.

This is when the AI juices start flowing. you see how powerful AI can be, even when you’re throwing multiple data points at it to improve your content output.

Instead of explaining what great looks like or what’s coming up on your content calendar, it’s reading it automatically along with your brand voice and style. You’re starting to create a content assistant that knows what it’s doing.

AI Reaches Out and Pulls In

Boom! You’ve connected up external data sources and realized AI can go out and gather on-going data sources to use in the process.

You’re starting to create an operation layer that is educating itself on how best to work with you.

You’ve combined your intelligence layer with arms and legs to go out and get the necessary information so you don’t have to. You’re not explaining what the AI assistant should do here….it knows.

The key element here is that you’ve taught the AI assistant to update the intelligence layer with data gathered.

Your prompt could look like this now:

”Draft for me one week’s worth of content and put the drafts in a Google Doc. Incorporate any intelligence you can gather from our recent deals won and make sure to update context files with anything new you learn. Ask me for approval before updating the files.”

You’re now focused on strategy around the content and not staring at a blank page waiting for words.

The Operational Layer Takes Shape

We carry on with the content creation example so the image below is heavily marketing focused, but this model works for a team of one or a team of 100.

You build a few workflows with AI, connected a few data sources and build out the intelligence layer. Now, you’re ready to experience the Operational Layer of AI.

Now your prompt looks like this:

“Let’s plan and draft this week’s content”

That’s all you need. You’ve built a system that knows what to do with that, knows where to pull and push from, and has all the background information.

You’ve spent your time on refining the strategy around the content. The system executes for you.

This is what it means to fold AI into your processes so it’s almost invisible.

How to Get Started and New Subscriber Benefits

I’ve launched a new service to help people get started with building out their system. It’s currently built for Claude but upon request we adapt to other LLMs.

There is a hand-built component to this so I’m charging a small fee. You can get more details at the link below.

For Paid Subscribers, there are details below about a new Resource Library and how to get your starter kit.

New Paid Membership Benefits

I don’t do much promotion of the paid tiers of this newsletter, but I have made some changes, and I wanted to make both paid members and those considering a paid membership aware of the updates.

Important note: all current paid members will continue to pay their normal price but receive the additional benefits.

Below you’ll find access to the Resource Library and your discount code for the AI Starter Kit mentioned above. For operator members, you’ll be receiving an email with information on how to start your Starter Kit build.

Here are the new membership tiers:

Read more

What I’m covering in this newsletter:

• What is a Connector

• What is an MCP Connection

• How to make a Connection

• Making Connections safe

• My favorite Connectors and how I make them work together

I want to dive into the topic of connectors in tools like Claude and ChatGPT because this is the next step for many users that can truly extend what they’re getting out of this technology. (and you don’t have to be technical)

What is an AI Connector

I like to think of Connectors as the arms and legs of whatever you’re doing with AI. Most people are still using AI without arms and legs. (that’s an odd comparison, I know, but stick with me)

Connectors give you mobility. A connector is a secure link between your AI assistant and a tool you already use. Gmail, Google Drive, Slack, OneDrive, your CRM, your calendar. Once a tool is connected, the assistant can pull in the relevant information on its own instead of waiting for you to feed it.

The shift is simple but it changes how you work.

Instead of “here’s the email thread, here’s the contract, here’s the context,” you ask the question and the assistant goes and gets what it needs. “Summarize my unread emails from this client.” “Pull the latest version of the proposal from Drive and check it against their requirements.” “What did we agree to in the Slack thread last week?”

There are two things a connector can do, and the difference matters. Some only read. They look things up and bring information back. Others can also act. They send the email, update the record, create the calendar invite. Reading is low risk. Acting is where you want to pay attention, and we’ll get to that.

Glad you’re here and thank you for reading Approachable AI by Pat Schaber! In case you missed some recent articles, here’s a recap:

What is an MCP connection

You’ll start seeing the term MCP, so it’s worth understanding what it means.

MCP stands for Model Context Protocol (sounds techie but don’t worry about that). Anthropic introduced MCP for connecting AI assistants to outside tools and data. ChatGPT and others followed their lead. Before MCP, every company that wanted its software to talk to an AI assistant had to build a custom integration for each one. Slow, fragmented, and it meant most tools never connected at all.

For you, the practical takeaway is this. The connector ecosystem is growing fast because nobody has to reinvent the bridge each time. The tools you depend on are far more likely to have a connector now than they were even six months ago, and the list keeps expanding.

In my odd parallel above, MCP is what gives you the ability to use the arms and legs.

How to Make a Connection

Both ChatGPT and Claude have made it fairly easy to make connections to your data sources and applications outside of AI. Below are a couple of diagrams that walk you through where to start to find your connections.

Once you find your desired connection, you’ll be asked to authenticate with the external applications such as Gmail, HubSpot, OneDrive, etc.

A Word of Caution (you’re welcome IT friends)

Okay, now that I have you excited about connections, I’m going to ask you to pause.

Here's the part the excited demos skip over. A connector means the assistant is reaching into your actual accounts. Your real inbox. Your real customer records. Your real files. That's exactly what makes it useful, and it's also what makes it worth slowing down for.

A read-only connector is hard to get in trouble with. The worst case is the assistant looks at something it didn't need to. An acting connector is different. When an assistant can send messages, change records, or delete things, a misunderstanding can become an email that went to the wrong client or a CRM field that got overwritten.

Here’s some advice:

• Connect what you need, not everything you can. Each connector is a door into a system. Open the ones that earn their place and leave the rest closed.

• Keep approval on for anything that acts (more below). Most platforms let you decide, per connector, whether the assistant can act on its own or has to stop and ask first. For anything that sends, writes, or deletes, make it ask. Reading on its own is fine.

• Be extra careful on shared accounts. A connector on a shared inbox or a team CRM acts with your full access.

None of this is a reason to avoid connectors! It's the same caution you'd apply before giving a new hire the keys to your systems. You'd start them with read access and a clear scope, then expand as trust builds. Same idea here.

Where to Review a Connections Settings

You have the ability to make sure the connection asks for your approval before it acts on your behalf. This is important.

My advice is that any connection that has WRITE or DELETE access should be prompted to ask for your permission before it takes action.

In ChatGPT, the Permissions settings look like this for a connector:

I will always typically err on the side of caution and just have it ask me before it makes any changes. I’d rather decide what’s important than turn over the keys of what’s important to the LLM.

In Claude, this is what the Approval Permissions look like:

I usually leave read-only tools as always allow, and anything that has WRITE or DELETE access to my accounts, I make sure it asks for my approval before acting.

Another quick note here: I will always lean towards the connectors that are built by the company you’re connecting to. For instance, Google building the MCP connector to Gmail. A third-party MCP connection deserves much more scrutiny.

My Favorite Connections

Ok, now that we’re done with the security stuff, let’s talk about the fun side of connections.

Where I get the most use out of connections is when I am able to use them together. For instance, taking information from one connection and using it to create something in a different connection.

I’m going to list out the connections below that I’m using the most right now, and then I’m going to put some prompts below that show how I use them all together.

Google (Drive, Docs, Gmail, Calendar). This is the foundation. Almost everything I produce starts or ends in a Google file, so connecting it means I stop pasting context in by hand.

Asana. This is where the assistant moves from informing me to doing the work with me. After a client call, I can turn the notes into properly structured tasks in the right project instead of typing them out one by one.

HubSpot. If you run any kind of go-to-market motion, connecting your CRM is the one that pays for itself fastest.

Granola. This is the one most people haven't thought to connect, and it's become one of my favorites. My meeting notes get captured automatically, and once they're connected I can do a variety of things with them.

• “Create drafts for my networking follow-up emails this morning, each with a key talking point pulled from that contact’s record in HubSpot.” (HubSpot plus Gmail)

• “Draft a proposal in the [client] folder in Google Drive based on my Granola notes from the calls on this project.” (Granola plus Google Drive)

• “Before each meeting on my calendar today, pull the contact’s history from HubSpot and write me a short prep note.” (Calendar plus HubSpot)

• “Take my Granola notes from this morning’s client call, log a summary to the contact in HubSpot, and turn the action items into tasks in the right Asana project.” (Granola plus HubSpot plus Asana)

• “Look at who I’m meeting tomorrow, find our last call in Granola, and remind me what we left open.” (Calendar plus Granola)

• “Pull my open deals from HubSpot that have gone quiet for two weeks and create a follow-up task for each one in Asana.” (HubSpot plus Asana)

• “Read the proposal in Google Drive and update the deal in HubSpot so the scope and amount match what we actually landed on.” (Google Drive plus HubSpot)

These are all things that I say throughout the course of the week. Once you have your system built and your connectors in place, the sky is the limit on how you combine the connections to act on your behalf.

And save you a ton of time!

AI Tech I’m Currently Using

I get quite a few questions on Substack, LinkedIn, and through my website on what AI tools I use for what use cases. I’ll try to share a few in each newsletter so it will give you some ideas of tools you can try for specific purposes. Here are a few for this week:

Everyday LLM - ChatGPT and Claude: I cover a lot of this above so won’t dive in too far here.

Call and Meeting Transcriptions - Granola: I like Granola for a couple of reasons. It takes the transcription and rolls it up into very good notes and action items but doesn’t need to be added to a meeting as an attendee. Also, it syncs great with my Hubspot CRM which makes it very easy to send notes to the contact record.

Presentations / Branded Documents - Gamma.app: I use this 3-4 times a week for anything I do with client presentations or professional document creation. I cover why I’m a heavy Gamma user in this post:

Video Recording / Editing - Descript: I’m trying to work in time to my weekly routine for video creation. I feel it’s important to give people a more personal connection. Descript is a huge timesaver. Imagine being able to erase or add text from a script and have it be reflected in the actual video in seconds. Unbelievable.

Sales Prospecting - Apollo.io and Clay: I don’t do a ton of outbound prospecting for my business but I’ve been using the free versions of Apollo.io and Clay to find companies that I could get to know through LinkedIn that may be interested in my services. I’m only touching on a sliver of the capabilities of these tools.

What I’m covering in this newsletter:

• What is Exa?

• How to start building this with Claude Cowork

• What data can be enriched and automated using Clay

• How to provide live campaign data results using a Claude dashboard

One of the biggest use cases I’m seeing for AI right now in go-to-market (GTM) operations is automating warm prospect list builds. Basically creating warm outbound opportunities for sales teams while you sleep.

Seems too good to be true, right?

I would have thought the same thing a couple years ago, but I’ve been working on some AI automations for customers that have completely changed my mind about what the opportunity and potential is here.

I’m going to work through a real-life scenario below to demonstrate where I see the power here. For this demonstration, I’m using one of my new favorite AI tools, Exa.

Exa is a search engine built for AI Agents you can hand a description instead of keywords, and it builds a verified list that stays alive. Point it at a signal like "Midwest SaaS that just changed CEOs," and it keeps finding new matches and feeding them downstream on its own.

Let’s dive into a real workflow using this tech stack:

• Claude Cowork (ChatGPT could also be used here)

• Exa

• Clay

• Google Sheets

• Slack

Email newsletters have the tendency to make everything look easy, but it took me a few tries to get this right. Hooking everything up wasn’t the issue. It was getting the output that I wanted, which took a couple of tries.

Glad you’re here and thank you for reading Approachable AI by Pat Schaber! In case you missed some recent articles, here’s a recap:

Kicking the Campaign Off

I opened up Claude Cowork, selected my operating system workspace for the folder, and put in this simple prompt (in the image below).

For this example, let’s assume that one of my ICPs is a president or CEO of a $20 to $200 million company that needs to figure out how to use AI better in his or her organization.

My telling trigger that says this person could be open to speaking with me about my AI consulting services is if they’re new in their position or the company has recently taken private equity money. They’ll want to show growth without adding a ton of headcount and may be open to using AI to help.

I want to start building my list up, but I also want it to be well-timed based on any announcements made by the company.

Exa MCP connection is set so Claude has a direct line to my Exa account and knows what it’s capabilities are there.

What Claude Cowork is going to do here is create what is called a “Webset” in Exa.

A Webset is a living list Exa builds from a plain-language description: it searches the web, verifies each company or person against your criteria, and fills in the fields you asked for. Unlike a one-time search, it stays alive, so you can put it on a schedule and have new matches show up on their own.

Here’s what Claude built in the Webset in Exa. It’s too long to show in one screenshot so I’ll show portions of the table. Here are 10 results of the actual search portion.

The top eight matched my four main criteria:

• B2B SaaS Company

• Annual revenue between $20-200m

• Headquartered in the Midwest

• Company announced a new CEO or round of PE funding

You can see the top 7 results matched with the number of references it found to ensure the match. You can click on each to see the references.

The bottom three missed on at least one criteria so information was not gathered.

For each of the seven results that matched all four criteria, Claude had it pull from it’s research:

• Company Name

• Website

• Headquarter location

• Trigger event

• Announcement URL

• The name of the current CEO

Now that we have a live, functioning dataset and we can start to use that information.

Automating the Clay Connection for Data Enrichment

Now I want Claude to connect my Clay table I’ve built for this to Exa so the information flows from Exa to clay on its own. I’m going to do that with what’s called a “webhook”.

A webhook is a URL that one app gives another so it can automatically push new data over the moment something happens, instead of you having to go get it.

Here’s what this looks like on the Clay side. Claude has configured my Exa Webset to push to this webhook in Clay.

(Apologies for the redactions here, but I don’t want information from these companies getting out through my newsletter.)

Clay is gathering the company information here and doing some analysis for me on revenue as well as more information behind the triggering event. This can be used in any outreach I may do.

All of this is done automatically. I don’t touch a thing.

Once a company hits that first Clay table, it is set up to automatically find the right contacts for me and send that to a second table.

There’s quite a few more columns in this second table than the one I’m showing below, but it’s tough to fit in a screenshot!

When I ask Clay to enrich the contact, I’m gathering as much information as I can to give me the competitive intelligence I need to do a warm, educated outreach. Below are a few of the columns.

Automating the Gathering of the Data

Next, I don’t want the information and new data gathered sitting in a clay table untouched. I want to get the information into the right hands. For this portion of the demonstration, let’s assume that I had a full team of sales and marketing professionals ready to act on this data.

First, I want to get the information into a central location. For this example, I’m just using a Google Sheet, but it could very easily have been HubSpot, Salesforce, or any CRM where you’re gathering the information for a campaign.

Second, I’m scoring the information towards a potential opportunity, and if there is urgency in reaching out, a message is sent to a Slack channel giving our salespeople a heads up that they may want to reach out soon.

Again, I want to stress that I’m not touching a thing here. As soon as Exa finds the information and sends it to Clay, this whole process is automated. The messages are sent and the databases are populated without my involvement.

The End Result

What I have now is a fully running prospect research mechanism that builds a prospect list based off certain triggers and will alert me if there is one it’s prioritizing as needing outreach sooner.

But let’s not stop there! That’s where campaigns have the potential to stall out.

Let’s assume, for this example, that we are sending the information into our CRM, where the contacts are tagged appropriately and the salespeople are actively reaching out to the contacts as they’re surfaced and enriched.

We’ll have Claude put together a live dashboard that I can share with the team that keeps everyone up to speed on the results.

Claude’s dashboard build:

Now we have a live, running reminder of the campaign and what’s working and what’s not

See what I mean by the potential here?

This is hours of work streamlined by AI that, in the end, gives us better, clearer information that we can use to improve the results of our outreach.

I encourage you to check out Exa, as I believe it has the power to truly impact AI agents with an elevated level of search functionality.

If you have any questions about how this was all put together, be sure to reach out!

Ways to Work with Me

Quick Win Session
A focused, high-impact working session designed to identify one immediate AI opportunity and give you a practical roadmap you can implement right away.

AI Readiness Assessment
A structured evaluation of your people, processes, data, and tools to determine where AI will create the most value and what needs to happen before you scale.

Workshops & Projects
Hands-on, team-based engagements that turn AI from theory into execution through tailored workshops, use-case design, and guided implementation.

Fractional AI & Marketing Leadership
Ongoing executive-level support that embeds AI strategy, operational rigor, and revenue-focused marketing leadership directly into your business.

“AI just makes our content sound and look like everyone else.”

No!!! It doesn’t have to!! (my actual response)

In fact, AI can make content sound just like you, look different than everyone else, and lift your brand up in the eyes of your customers.

Here’s how.

What I’m covering in this newsletter:

• Why so much content is sounding and looking the same

• How to build a ContentOS that keeps your unique voice and helps you stand out to your customers (in new formats).

• An example of how this works in my own content operating system

• Bonus for Paid Subscribers: Download my Claude Cowork Content Operation System(OS) that takes content production to the next level.

I have a special spot on my heart for content marketing, and I always will. I believe it’s a way to differentiate yourself in whatever market you’re in and build infinite trust with your target market.

The problem is, you can quickly lose that trust and crush that strategy.

Open your feed and read five posts about any topic of interest. They’re blurring together. I feel like so much content has the same structure, the same confidence, and the same three takeaways you saw in the thing you just read before. Ugh.

It’s not your imagination. It’s what happens when companies use AI to speed up content production and don’t use AI as a strategic partner to differentiate their content.

If you are a content marketer or manage a content marketing function, this is the threat hiding inside “AI makes content easier.”

Easier for you is easier for everyone.

Speed stops being an advantage the moment it produces the same thing your competitor just shipped. And I think that’s where a lot of companies are stuck in 2026.

I hate to tell you this but your content may be starting to stink.

If you have a content marketing function in your Marketing team or are responsible for producing content, this newsletter is for you.

This is how to rethink content marketing and treat AI as a partner to “level up” content and not the crutch to enter the sea of sameness.

Glad you’re here and thank you for reading Approachable AI by Pat Schaber! In case you missed some recent articles, here’s a recap:

Remember when production was the whole job?

For most of the last decade, a content marketer’s day was production. Researching the post, writing the draft, editing it, formatting it, shipping it, then doing it again. The skill was the output, and the output took real time. Volume was the constraint. You hired more people or you published less.

And then that constraint sort of went away overnight.

A draft that used to take a day takes an hour. The blog post, the email, the social cuts, the first version of the landing page copy all got faster, and it will keep getting faster.

But watch out, this is where the trap is.

When production gets cheap for you, it gets cheap for everyone, and the floor of average content rises while the ceiling stays flat. That is the sameness problem.

So the question for anyone managing the role is no longer how to produce more. It is what the role does once producing is the easy part.

This is how I would evolve the content function

Increasing content production and moving farther into the sea of sameness is not the answer.

What if that same person, freed from the production grind, moves to the work that actually separates you from the field and that AI can’t do on its own?

Stop with the vague prompts. Put together a system of skills and agents that work together (this is doable for anyone).

Here’s how to get there with an AI content operating system.

Give the system the right knowledge (also called context)

First, give the system the context it needs to get your first drafts of any type of content 80% of the way there.

I’m not talking about letting AI create all of your content. I’m talking about building a system that knows you, your voice, your customers and your positioning so well that it can produce quality content that gets you 80% of the way to a finished draft.

At a minimum you need context files for your content marketing that include:

• Audience Overview

• Brand voice

• Positioning

• Product

• Style Guide

These are simple text files.

Then, give the system examples of what great looks like.

(For paid subscribers, I have a download below that will help you create these files and put them in your content system automatically.)

Give the system your strategy

No, you’re not throwing your content and company strategy out into the ether for everyone to copy. You’re simply telling your small, localized OS what your strategy is to reach your target market with content.

Include information such as:

• Audience map: Who your write for, what content resonates, and when it resonates

• Calendar: Let it know what is coming up and why. Literally, work on this with your system.

• Format picker: Give the system the recipe for how you pick what format, for what type of content, for what type of customer, etc.

Teach the system how content gets created

These are your Skills and Agents.

This is where you teach your system to create content that’s not like anyone else. That’s a step above what your competitors are doing.

You’re simply teaching your operating system how the content gets created so that it can do it for you.

For instance, a Repurpose skill might let the system know that every time you put together a Blog post, you want it broken up into content for Facebook, LinkedIn, X, and Threads. Along with that, you’re telling it exactly how you want your content for each of those platforms to be created.

See what’s happening here? Your content function is spending more time on the strategy of what gets created and for whom, rather than cutting and pasting content and rewriting it for different platforms.

Give the system some starting points

Oftentimes, we have specific formats and layouts we like for our content. I’ve found that when you leave it up to AI to match that, it typically falls short.

So I like to give it a template to start with.

For instance, if I want to create a landing page, I want my system to know how I typically structure that: how I want the flow of information to be, what layouts I want to be consistent through all the landing pages I create. I give it a template to begin with.

Example Time

As I was creating this newsletter content, I decided to give my content operating system a workout. I asked it to create me an interactive content workflow calculator for my website.

This is the exact prompt I used. It’s one sentence.

The reason it can be one sentence is because my Content Operating System knows everything about why I would want to create a content workflow calculator for my website and what I’d want it to show.

It knows who it’s for, and it knows what message and positioning I’m trying to put into the calculator. It also knows my strategy is to create more interactive lead magnets rather than static PDFs.

Voila. For this exercise, I’ve made zero edits to this so you can see what it came up with on the first try.

In reality, I would definitely spend probably a half hour working with it to get it exactly the way I want it.

Note: I said a half hour. A couple years ago, this would have taken three months to get a developer to help put this together and build it into HTML for the website. It would have taken meetings to figure out what we’d want in there, and then, of course, we would have had to figure out how to style it correctly. Good riddance to those days!

One final component: The Brain

For this content operating system exercise, I used Claude Cowork as my engine. And to make that engine run, it needs a brain, commonly known as a CLAUDE.md file.

Whenever you start a session utilizing this folder, the first stop is the brain, which gives it the instructions on how the whole thing operates.

For my paid subscribers, I have this as a complete download below, where there is an onboarding skill called “START-HERE.md” which walks them through how to set this up in Claude Cowork.

No, your content doesn’t have to look and sound like everyone else when you use AI.

You can maintain your unique voice and blow your customers away with new ways of presenting content.

You can elevate your content marketing function beyond just an execution engine that rewrites and reformats content.

You just need to take that next step with AI.

Download your content operating system starter kit

If you’re looking to take what I just described above and put it into action today, I’ve put together this complete system that includes an onboarding that will walk you through all the steps to create your full content OS.

Read more

I’ve always loved crunching larger amounts of data.

Pulling spreadsheets from different sources, combining them, creating pivot tables, and finding unique ways to pull information from all the data stories.

I’ve always felt that decision making without this type of data analysis was hollow.

With AI technology at our fingertips, there’s really no excuse for anyone not to be able to do this data analysis in minutes rather than hours and create the backing that makes decision making solid.

Gone are the days of toiling away in Excel for hours or days to find ways to put different sources of data in a common format so that you can make use of it.

With the right setup using AI technology (Claude Cowork or Claude Code for these examples), this can be done in minutes and on a routine schedule that you don’t even have to think about.

I’ve had the opportunity recently to work on some of these bigger data projects with clients, and I’ve come across some great use cases.

Here are 5 you can start to plan out today!

(Next week we’ll build one of these out so you can learn how to create your own.)

Glad you’re here and thank you for reading Approachable AI by Pat Schaber! In case you missed some recent articles, here’s a recap:

1. Voice of Customer → “One source of truth for what customers actually want”

The great internal debate on what the customer actually wants can become much more factual.

Consider these sources:

• Sales call recordings

• Customer support tickets

• Current customer review calls

• Public reviews

Let AI do the gathering and sorting. Use Claude Cowork or Claude Code to gather information from the your desired sources and have it create an ongoing database (I’ve been using Supabase) with a unified taxonomy (pain points, feature requests, competitor mentions, sentiment by stage).

Outcome: You stop arguing about which feedback to prioritize. The report says “this issue shows up in 40% of sales calls, 60% of support tickets, and 30% of reviews.” Anything hitting all three goes to the top of the product roadmap. Anything only in support is usually a documentation problem, not a product problem.

2. Sales & Revenue Intel → “Reverse-engineer your actual sales playbook”

Let’s make your sales process smarter with factual information about what actually works. Too many times we work off too much theory here.

Consider these sources:

• 12 months of won/lost deal notes from your CRM

• Email reply threads from outreach campaigns

• Proposal templates tagged with win/loss outcome

Feed all three sources with deal metadata such as ACV, segment, rep, close date, etc.) Then ask Claude to identify language, sequencing, objections, and price points that show up in winners vs. losers.

Now you have a reality-check on what your team thinks works vs. what actually does. The new-rep onboarding doc basically writes itself, and your VP Sales finally has data behind the coaching they're doing.

3. Competitive & Market Intel → “See your competitor’s next move 60-90 days early”

All of us have wanted better insights into what our competitors were about to release to the market. Well, now you can get pretty damn close.

Consider these sources:

• Competitor job postings scraped monthly

• Their press releases and blog posts

• For public companies: earnings call transcripts and 10-Ks

Here’s an example scenario: Quarterly, Claude looks for directional signals across all three: "5 ML engineer hires + 3 PR mentions of 'agentic workflows' + earnings call reference to an automation roadmap = they're building X."

You spot product launches and strategic pivots before the official announcement. Your sales team gets battlecards before the competitor's launch hits, and your marketing gets a head start on positioning. Wins all around!

4. Operations & Internal Knowledge → “Turn a year of internal chatter into an SOP library”

How many times does your client services team answer the same questions over and over again, but yet no one has the time to create the database of answers? It happens a lot.

There’s a missed opportunity happening with an internal FAQ database and standard operating procedure (SOP) documents being created from known knowledge.

Consider these sources:

• 12 months of Slack channel archives

• Granola/Fireflies meeting transcripts

• Project retrospective notes

Claude clusters recurring questions (the ones that get asked over and over again), repeated decisions, and failure modes from retros. Then cross-references against your existing documentation to find gaps.

You can have a prioritized list of the 10 SOPs you actually need to write with Claude drafting the first version of each. Onboarding time for new hires drops because the tribal knowledge is finally documented.

5. External Signals & Trigger Mining → “A trigger-event engine for outbound”

From experience, I can tell you that marketing and sales teams spend a lot of time trying to figure out the perfect time to reach out to their ideal customer profile.

Now you can spend more time on the strategy of the outreach versus the drudgery of the information gathering.

Consider these sources:

• LinkedIn job change feeds for target accounts (champion moves, exec hires)

• Funding/M&A news for your ICP target companies

• Industry podcast transcripts where execs reveal priorities.

With a good-sized target company list and scheduled Claude information retrievals, you could be sitting on 3-5 outbound-ready emails per day in your drafts folder with tied to a real, timely signal.

Reply rates lift because every email is specific and actually relevant to something happening in the prospect's world right now.

Conclusion

In an upcoming newsletter, I’ll dive further into how you can set these up, but I wanted to get you thinking about how to use AI to organize the vast amounts of data at your fingertips.

Everyone is doing more these days with fewer resources, but the beauty of AI is that some of these important data mining tasks for better decision making can happen without utilizing your time.

AI Tech I’m Currently Using

I get quite a few questions on Substack, LinkedIn, and through my website on what AI tools I use for what use cases. I’ll try to share a few in each newsletter so it will give you some ideas of tools you can try for specific purposes. Here are a few for this week:

Everyday LLM - ChatGPT and Claude: I cover a lot of this above so won’t dive in too far here.

Call and Meeting Transcriptions - Granola: I like Granola for a couple of reasons. It takes the transcription and rolls it up into very good notes and action items but doesn’t need to be added to a meeting as an attendee. Also, it syncs great with my Hubspot CRM which makes it very easy to send notes to the contact record.

Presentations / Branded Documents - Gamma.app: I use this 3-4 times a week for anything I do with client presentations or professional document creation. I cover why I’m a heavy Gamma user in this post:

Video Recording / Editing - Descript: I’m trying to work in time to my weekly routine for video creation. I feel it’s important to give people a more personal connection. Descript is a huge timesaver. Imagine being able to erase or add text from a script and have it be reflected in the actual video in seconds. Unbelievable.

Sales Prospecting - Apollo.io and Clay: I don’t do a ton of outbound prospecting for my business but I’ve been using the free versions of Apollo.io and Clay to find companies that I could get to know through LinkedIn that may be interested in my services. I’m only touching on a sliver of the capabilities of these tools.

What I’m covering in this newsletter:

• An example of a live, interactive Claude dashboard that helps me manage my work

• How to securely build a Claude Live Artifact dashboard

• How context in a personal Claude operating system can bring this to life

• Bonus for Paid Subscribers: A Claude OS Starter Kit built for Claude Cowork

It was time I got more sophisticated with my daily and weekly project management. And boy, did I ever do an overhaul.

(I’m going to put this in the bucket of “it’s a good problem to have.”)

My business is growing as I’ve signed more customers, and with each new customer comes more action items and things I need to keep track of.

The previous method of keeping projects and tasks in spreadsheets wasn’t scalable and it was getting confusing. Time to grow up!

I centralized everything in Claude Cowork and created a Cowork live artifact that helps me completely manage my day and syncs up to all my external sources.

I’m going to walk through everything below and show you exactly what I did to set this up. Whether you’re a solopreneur or the president of a company, what I’m going to show below can honestly change how you approach your week of work.

Note: I’m blurring out customer and financial data in this newsletter

Glad you’re here and thank you for reading Approachable AI by Pat Schaber! In case you missed some recent articles, here’s a recap:

Let’s Start with the Finished Product

I think we’ll start with the end result this week and then walk through how it was created.

In my Claude Cowork (desktop app), I go to Live Artifacts where I have two dashboards I look at daily:

• My Daily Work Briefing

• Morning Financial Summary (future newsletter)

For this newsletter we’re going to focus on the Daily Task Briefing.

My Daily Work Briefing

My Daily Work Briefing - What’s Included

• Overview of customer project status and what needs to be accomplished

• Anything overdue, due today or labeled as high priority that needs my immediate attention

• Items that Claude thinks it can do solo while I work on other things (love this)

• What is coming up as due in my weekly content plan for my business

• Any follow-up tasks I have in relation to any of my open pipeline opportunities

• Any outstanding invoices and if I need to do follow-ups

Customer and Business Action Items

Pulling directly from my Asana projects with ability to update the project in Asana directly from this interactive dashboard. Each client has an expandable section with it’s own items.

Centralized Content Plan for the Week

Pulling from my content calendar in Asana paired with my LinkedIn content skill in Claude which is my strategic partner on topics and angles of interest. I still do most of the writing myself (I enjoy that too much to hand that over to AI).

Open Opportunities Follow-Ups

I’m only showing two here but it lists all my prospect opportunities from Hubspot with an open action item so I never miss a follow-up. I can open up the opportunity directly in Hubspot if I need to.

Invoice Follow-Up

One of my least favorite parts of my business, but a necessity, of course. I like the daily reminder of what needs to be followed up on in the next few days. Plus, I can click on that Follow up button and an email will be drafted in my drafts folder.

How to Build a Claude Live Artifact Dashboard

Below is a step-by-step tutorial of how to get started building out your own live dashboard.

First things first

A key element for me that makes this a strategic partner is my personal operating system I built within Claude. I won’t dive into this for this newsletter because I’ve already covered it HERE but I can’t stress enough how important this is to truly taking advantage of everything Claude offers.

This is not required to build the live artifact. But this is the part most people will skip, and it's the difference between a briefing that feels generic and one that feels like it gets you.

The Claude desktop app

If you’ve signed up for a Claude Pro or Max account and you’re only using the browser version, you’re utilizing about 10% of what Claude has to offer. Download the desktop app and start learning the other 90%. Trust me, it’s worth it.

Below is the main screen of the desktop app. You’ll want to click on the Cowork tab and then head to Customize.

Connect up your data sources

The key to making this live dashboard truly functional is to connect up your data sources that you want to populate it with.

One note here: If you have an AI Policy in place that has guidelines for making data connections, now is a great time to review that.

Find the Connectors section and hit the + button.

Next, you want to search for the applications that you want to connect to. For me, I recommend sticking with Anthropic and Partners because they’re the most trusted connections within Claude, but know there are other options out there.

My Connections for this dashboard include:

• Asana for all project and customer related work

• Hubspot for my open prospect opportunities

• Google Drive for my invoicing information (eventually will be moving this to Quickbooks which is also a Connection)

A quick note on Connector security

All of these connectors have settings you can set for permissions. One of the key areas I’d always look for is the Write/Delete tools. I’ve learned the hard way that you definitely want to have Claude ask you for approval before making a physical change to your data on a connected source.

Once you have these connections made, I’d suggest going back to Cowork and just testing them, asking Cowork to go out and find certain information, just to make sure that everything is hooked up correctly.

Building the actual live artifact dashboard

With Cowork selected, click Live Artifacts in the upper left and then click New Artifact. Claude will give you the option to create with Claude. You’ll press that to continue.

Be ruthlessly specific in your build prompt

Claude kicks off the chat with a generic prompt which is just a starter. It will list out your MCP connections and make suggestions on what could be hooked up.

You’ll notice in my screenshot below that my PatOS is pre-loaded and pumping in context to the discussion.

Now you create a very specific next prompt.

"Build me a daily briefing" produces a mush. The prompt you give Claude when creating the artifact should spell out: what sections you want, in what order, what time windows each section covers, what counts as worth surfacing versus hiding, and what the top of the page should always show. Treat it like you're writing a spec, not a wish.

What happens next

Plan for at least three iterations: The first version will surface the wrong stuff, miss things you care about, and over-index on stuff you don't. That's normal.

Build it, look at it for a couple of days, then go back to Claude and say "this section is too noisy, tighten it" or "add a row that shows X." This is where the artifact actually gets good.

I’ve edited mine probably 20 times because I keep thinking of things I want to add or functionality I want to add. Don’t accept it right away. Keep working with it!

A Starter Claude OS Kit

I mentioned above that having all the context related to your business, team or individual work can really bring this dashboard to life.

I’ve put together a starter Operating System kit built for Claude Cowork that has an onboarding skill that walks you through setting up your key context files and getting started with your own personal or professional ClaudeOS.

Below are instructions on how to download, set up and install.

Read more

What I’m covering in this newsletter:

• The AI burnout is real and shouldn’t be ingored

• Why AI leadership is a team design discussion and not an individual initiative

• What a group operating structure can look like

• Actions leaders can take today to get started

I had two speaking engagements this week that were an interesting dichotomy of perspectives. I spoke to a classroom of junior and college seniors who are navigating an entry into the workforce like no other generation before them. Then the next day I spoke to a group of highly experienced senior marketing leaders who are navigating management of a new type of work like no generation before them has had to do.

Then there are friends of mine that are just worn out reading about it.

I know the AI burnout is real. It’s heavy.

While I can’t control that, I can control putting out AI content, frameworks, and systems that help people navigate these uncharted waters. I maintain that great AI adoption can mean up-skilling for employee careers and true measurable results towards a company’s objectives.

I want to focus this week on leaders navigating a new era of work.

If you manage a team right now, you’re getting it from both sides.

From above, the pressure is to drive AI adoption, show productivity gains, show how your team is using it and demonstrate ROI.

From below, the anxiety is real. Your people are watching layoffs at other companies get blamed on AI. They’re reading about which jobs disappear next. Some of them are quietly burning out trying to figure out which AI tool will make them indispensable. Others are quietly hoping it’ll all blow over.

You’re stuck in the middle and no one trained you for this. There isn’t a playbook yet.

So most managers do what’s reasonable in the absence of a system. They look for people to try things, to figure it out and bring back what works.

I think that’s the move that’s quietly burning everyone out.

Let’s dive into a system to fix that.

Glad you’re here and thank you for reading Approachable AI by Pat Schaber! In case you missed some recent articles, here’s a recap:

Why individual AI effort isn’t working

When you tell people to figure out AI on their own, three things happen.

• They build private workflows nobody else sees, so the team gets zero compounding benefit.

• They hide their experiments because they don’t want to look stupid in front of peers.

• They get exhausted, because learning a new technology while still doing your day job is a lot to carry.

Plus, this does not all happen in equal measure across every employee. Does this look familiar?

The business world is asking leaders and managers to reshape how their group thinks and works while still managing how their group currently thinks and works. With a wide disparity of skill sets on top of it.

That’s a heavy load!

Reframe AI adoption as a systems problem

Here’s where I’d push you to flip how you’re thinking about this.

Treat AI adoption as a systems design discussion, because that’s what it is.

Your job as the leader is to build a team operating system. A shared way of working with AI that lets your group learn, calibrate, and improve together. The companies and teams winning with AI right now share a way of working with it. Individual AI skills matters far less than the collective architecture around it.

This is how I challenge leaders to think about this. Stop optimizing for individual capability and start designing the team OS.

Let’s examine what an AI system architecture can look like.

Four main components of a team AI OS

A working team OS for AI has four pieces. Each one needs to exist deliberately.

Context

I don’t care what LLM you’re using at your company. Context is still the key component to get the best outcome.

Context is the shared knowledge layer your AI uses to do anything useful.

• Your ICP

• Your customer language

• Your brand voice

• The internal terminology your team uses every day

• The way you describe your product

• The way your buyers describe their problems.

Without a shared context layer, every AI output is generic. With one, every output starts from your team’s actual reality and your team needs to own this.

This lives in shared files the whole team can access and update. Owned together, maintained together, treated as the team’s source of truth.

Skills

Skills are the prompts, workflows, and processes your team uses to actually get work done with AI. Built around the real work your team does.

A skill might be:

• Qualify an inbound lead from a webform

• Draft a follow-up email after a discovery call using the meeting notes

• Rewrite a sales asset for a specific industry vertical

These are documented, versioned and improved over time. New team members get the skills on day one. Veterans contribute new ones as they figure out what works.

Remember, many of these skills will cross-employee function, so it’s important the engagement across the team is high on the development of them.

Outcomes

This is the “what great looks like” library. Gold standard examples of finished work, so the AI has something to calibrate against and so the team has a shared definition of quality.

• What does an excellent qualification email look like?

• An excellent case study?

• An excellent pricing summary?

Most teams don’t have these examples written down anywhere. They live in someone’s head, or in a folder of stuff that worked once.

Pull them out, put them in the OS, point your AI at them. The output gets dramatically better. So does the team’s shared understanding of what they’re aiming for.

Team Roles

None of this happens on its own and they’re not set-it-and-forget-it pieces. Someone has to own each piece. This is also a great opportunity for employee upskilling.

Five roles to define explicitly:

• Context owner: Keeps the shared files current. Makes sure new ICP changes, brand updates, and customer language get reflected.

• Skill developer: Builds new workflows. Tests them. Documents what works. Pushes them out to the team.

• Outcome owner: Maintains the gold standard library. Decides what makes the cut.

• Team trainer: The early adopter who teaches others. Usually emerges naturally, but you have to give them the time and authority to do it.

• Human touchpoint owner: Defines where humans review, decide, and own the relationship. More on this in a minute.

On a small team, one person can wear multiple hats. The point is that the roles exist on paper, with names next to them, so things actually get owned.

This is also part of the ongoing challenge that managers will face as far as rethinking work and roles on their teams.

The human touchpoints piece

This is the part I’d encourage you to spend the most time on, because it’s where employee anxiety actually gets resolved.

Most of the fear about AI in your team comes from one specific thing: nobody is telling them where their job still lives.

When you define touchpoints explicitly

• This is where the human reviews the AI output

• This is where the human decides which path to take

• This is where the human builds the customer relationship

• This is where the human owns the judgment call

the fear drops fast.

Your team can see where they’re amplified by AI, with their judgment still at the center of the work.

Treat touchpoints as a specific deliverable. List the moments in each workflow where a human stays in the loop and why. Document them. Make them visible to the team. That’s the fastest path to reducing anxiety while still moving forward.

Where to start this week

You don’t need a 90-day rollout plan. You need to start designing.

Step one. Sit with the four components yourself first.

Block 60 to 90 minutes of focused time. Just you. Walk through each component for your team:

• What should be in our shared context layer? What does our AI need to know about us, our customers, our voice, our work? (also marketing 101)

• Which workflows would benefit most from AI skills? Where is the actual friction in our week?

• What does great look like for each of those? Do I have examples I can point to?

• Who on my team should own which roles?

You don’t need perfect answers but I’d recommend a working draft. This is design work, and it’s yours to do as the manager.

Step two. Run a workshop with the team

Bring your draft to the team. Pick two or three processes that need real improvement, and redesign them together with AI infused. Use the workshop to surface:

• what context is missing

• what skills you need to build first

• what great actually looks like

• who’s going to own what

• how do we measure ROI

The system architecture builds from that workshop. The team owns it because they helped design it. Then you keep iterating, weekly or biweekly, as the team gets sharper.

That’s the playbook. Manager designs first. Team shapes it together. The OS gets better every week.

This shift

The change I’m asking you to make is small, but it changes everything.

Stop trying to make your people AI-savvy individually. Build the conditions where your team gets better together and grows together. That’s what leading a team through AI actually looks like.

The technology is moving fast. Your team is allowed to be uncertain about it. So are you. The leaders coming out ahead are the ones who built the systems that let their teams figure it out together.

That’s the PLAYBOOK!

Ways to Work with Me

Quick Win Session
A focused, high-impact working session designed to identify one immediate AI opportunity and give you a practical roadmap you can implement right away.

AI Readiness Assessment
A structured evaluation of your people, processes, data, and tools to determine where AI will create the most value and what needs to happen before you scale.

Workshops & Projects
Hands-on, team-based engagements that turn AI from theory into execution through tailored workshops, use-case design, and guided implementation.

Fractional AI & Marketing Leadership
Ongoing executive-level support that embeds AI strategy, operational rigor, and revenue-focused marketing leadership directly into your business.

AI Tech I’m Currently Using

I get quite a few questions on Substack, LinkedIn, and through my website on what AI tools I use for what use cases. I’ll try to share a few in each newsletter so it will give you some ideas of tools you can try for specific purposes. Here are a few for this week:

Everyday LLM - ChatGPT and Claude: I cover a lot of this above so won’t dive in too far here.

Call and Meeting Transcriptions - Granola: I like Granola for a couple of reasons. It takes the transcription and rolls it up into very good notes and action items but doesn’t need to be added to a meeting as an attendee. Also, it syncs great with my Hubspot CRM which makes it very easy to send notes to the contact record.

Presentations / Branded Documents - Gamma.app: I use this 3-4 times a week for anything I do with client presentations or professional document creation. I cover why I’m a heavy Gamma user in this post:

Video Recording / Editing - Descript: I’m trying to work in time to my weekly routine for video creation. I feel it’s important to give people a more personal connection. Descript is a huge timesaver. Imagine being able to erase or add text from a script and have it be reflected in the actual video in seconds. Unbelievable.

Sales Prospecting - Apollo.io and Clay: I don’t do a ton of outbound prospecting for my business but I’ve been using the free versions of Apollo.io and Clay to find companies that I could get to know through LinkedIn that may be interested in my services. I’m only touching on a sliver of the capabilities of these tools.

What I’m covering in this newsletter:

• Why AI gives you a new way to think about trade show operations

• AI Workflows for what could happen before the trade show, at the event, and after the show

• A structure for outlining and automating the ROI analysis

• [Bonus for Paid Subscribers] A starter TradeshowOS for Claude Cowork that includes a completely integrated ROI spreadsheet in Excel.

Having spent 25 years in marketing, the trade show conundrum has always been there. You need to be at the show so your brand stays top-of-mind, but yet showing a consistent positive return on the trade show has always been an elusive initiative.

But I feel we have a real opportunity to rethink trade shows with an operational layer built with AI.

So, I created a fake company and trade show to bring to life how I’d build an AI operating system to help plug any leaks in the revenue bucket from the event.

Time to rethink how this could work with AI.

Northpoint Software (the fake company) just dropped $87,000 on a booth at MfgConnect 2026 (the fake show). Their four-person team scanned 184 badges across three days. Two weeks later, the VP of Marketing pulled the post-event report. Three meetings booked. One opportunity in HubSpot. No clear ROI story to give the CEO.

Sound familiar

The booth was fine but missed opportunities stacked up pre-show, at the event, and post-show.

The numbers back it up.

• 80% of trade show leads never get a single follow-up

• Of the teams that do follow up, 40% wait three to five days, by which point qualification odds drop another 80%

• And 64% of exhibitors call ROI their biggest challenge.

But the average trade show still returns 4:1. Which means the teams that DO follow up are running away with the value while everyone else just funds their own travel budget.

AI can change the math here. Used as a system, it turns a 184-scan event into a manageable pipeline build for the company.

Let’s walk through what a 4-block TradeshowOS could look like.

Hi, I’m Pat! Glad you’re here and thank you for reading Approachable AI by Pat Schaber! I help businesses transform go-to-market operations with AI. I also write a ton of content about that. Here are some recent articles:

The TradeshowOS Setup

For this example I’ll be using Claude Cowork but this concept it could be replicated on other platforms.

Let’s get the Northpoint Software’s event coordinator set up with a well-organized context and skills operating system in Claude. Here’s what the folder structure could look like. For Paid Subscribers this setup will be available with Skills and templates at the end of the newsletter.

if you’ve been following this newsletter for a while, you know I’m a fan of this OS concept, so I won’t dig in too deep here. If you want more of an overview, you can check out a recent post that talks more in depth about what all of this means

Basically, we’re packaging up all Context files related to company, show and strategy. We’re also bundling Skills and Agents that I’ll dig into a little bit more below, as well as Templates.

And , most importantly, an ROI spreadsheet that everything rolls into.

Pre-Show Enrichment and Outreach

Okay, let’s get Northpoint going early on show operations.

Northpoint has the registrant list 30 days out. They run the list through these steps using the Skills in the OS:

1. Enrich the company and contact information. Put the raw spreadsheet of information into Claude and say “enrich the attendees”. Northpoint has a CLI or MCP connection to Clay (or something like Apollo) which automates the process of giving them a consistent contact and company information on show attendees.

2. Score with Claude. A new Skill takes the enriched list and ranks every contact Tier A, B, or C against Northpoint’s ICP, with a one-line “why they matter” for each Tier A. Remember that has all the context is baked into the operating system.

3. Outreach to Tier A. Next, a personalized email sequence (based off the enriched information) is automatically created for the top 30 accounts. “We’re going to be at MfgConnect. Here’s the specific reason 15 minutes is worth it.” Booked through Calendly with a two-hour booth window blocked off each day.

The booth team walks in with eight pre-booked meetings on day one instead of zero.

At-Show Capture and Same-Night Triage

I’ve always been amazed at how much context is lost after a scan. There could be a great conversation that is lost after the scan if a process isn’t in place to move that information along.

The standard pattern is: scan the badge, hope the rep wrote a useful note, deal with it Monday. Monday could be too late.

Here’s how we can rethink this for the Northpoint team.

1. Voice memos after every conversation at the booth. Options for this could be Granola on a phone, Apple Notes, or directly into a predefined Slack channel. 1-2 minute note. Who, role, pain mentioned, fit signal, next step.

2. End-of-day Claude processing. The team drops the day’s voice notes into a Slack channel (if not recorded directly there) before leaving the booth for the day. Claude is scheduled to run that night to process the notes. They get back a triaged list with suggested tier and follow-up angle for every contact.

3. Hot lead alerts in Slack. Anything Tier A gets flagged to the rep responsible before they go to bed. A follow-up could happen while the conversation is fresh with the prospect

This sounds like a small change. It’s actually the biggest lever in the whole system. By the time the team flies home Sunday night, the follow-up plan is already written.

The 72-Hour Follow-up Sprint

The 72 hours after the show is where teams either run away with the value or hand it back.

Contact and company re-enrichment. Claude processes a completely updated list of original contacts along with everything new from the show and Clay re-enriches with your preset Clay template using new information.

Tier A gets a personalized email from the rep within 24 hours. Claude drafts it using the voice note context. Specific, short, references the actual conversation, includes one clear next step. The rep reviews and sends.

Tier B gets enrolled in a HubSpot sequence with a soft CTA. A case study, a webinar invite, or a 15-minute discovery slot. AI handles the personalization tokens. The rep handles the replies.

Tier C goes into long-tail nurture. Newsletter, LinkedIn retargeting, a follow-up campaign two weeks out. No rep time required.

The main thing here is that everyone gets a touch and no one falls through the crack. The rep’s calendar stays sane because AI handled the volume.

ROI Tracking that Compounds

The fourth block is the one that turns a trade show line item into a strategic asset.

A simple spreadsheet with three views:

1. Single event view. Inputs like booth, travel, swag, sponsorship, and team time. Outputs like contacts, meetings, opportunities, and closed-won. Calculated metrics include cost per qualified lead, cost per opportunity, and payback period.

2. Multi-event comparison. A rolling view of the last three to five shows. This is the killer feature. Most B2B companies have no idea which trade shows are actually paying off because nobody tracks it longitudinally.

3. Lead triage tracker. Tier A/B/C status, last touch date, current stage.

After two events, Northpoint’s VP of Marketing knows MfgConnect is a 6:1 show and the regional one in Q3 is barely break-even. They cut the regional show, doubles down on the bigger one, and re-allocates $40K of budget. That’s the conversation she has with the CFO. That’s how event budgets stop getting cut.

The beauty of this is that with the right connections set up to your LLM of choice, much of this work can be automated and dashboards created. No digging through endless CRM records and exported spreadsheets.

Conclusion

Is all of this easy? Nope!

This takes work and ownership (by a human) to maintain and scale but the backend benefit is less leaky opportunities from show operations and more strategic insight into how events are impacting company results.

The Complete TradeshowOS for Claude Cowork

The system includes a starter Claude workspace with a built-in setup guide that walks you through configuration before you run anything. The zip file includes:

• 8 Skills that are set up to work within this system and are ready for your company customization

• Pre-existing Context files that are ready for your company information

• Onboarding system that walks you through system setup (expect at least a half hour)

• A trade show ROI tracker in Excel that is integrated into the system

Drop the folder into your Claude workspace, customize and set up to your needs, and let the OS take care of the rest.

Read more

What I’m covering in this newsletter:

• Why it’s time to rethink the marketing and sales relationship and how it operates

• Use Case #1 - The Lead Brief Generator

• Use Case #2 - The Discover Call Insights Assistant

• Use Case #3 - Closed Won/Lost Analysis Generator

• [Bonus for Paid Subscribers] A starter marketing/sales handoff system set up for Claude

There is revenue being leaked out of the pipeline between sales and marketing, but the fix is different than you think.

The outdated leadership model that highlights a culture problem between sales and marketing and proposes more “touch-base” meetings or a cross-department team bonding experience needs to stop. Those aren’t the issues.

This is indicative of leaders who don’t know how to build the infrastructure and architecture around cross-department communications and collaboration. They don’t understand go-to-market operations in 2026.

And now with the introduction of AI, this lack of leadership will be quickly exposed.

I could go on for hours about the infrastructure that I would put in place immediately if I were leading a marketing and sales organization, but no one has the time to read that in a newsletter.

Instead, I want to focus on some actionable and approachable ways that you can start to put pieces in place today with AI that build a stronger bridge between sales and marketing.

So for this newsletter, we’re going to focus on a few marketing/sales handoff use cases and ways that I would solve those and enhance those communications easily with AI.

It’s time to think differently about how you approach this type of work.

Let’s dive in!

Hi, I’m Pat! Glad you’re here and thank you for reading Approachable AI by Pat Schaber! I help businesses transform go-to-market operations with AI. I also write a ton of content about that. Here are some recent articles:

The MQL to SQL Handoff

Probably the lowest hanging fruit right now of any AI application I work on.

There are two types of leads that typically land in a sales rep’s queue. Both have the same problem.

The first is the inbound form-fill. Someone fills out a contact form or requests a demo. The rep gets a notification: name, email, company, maybe a “how did you hear about us” field. What’s missing is any sense of why this person showed up, what they were looking at before they hit submit, or what they’re actually trying to solve.

The second is the scored MQL. Marketing has been watching this prospect accumulate engagement over time. At some point the score hits a threshold and the lead gets handed off. The rep sees a number. They don’t see the behavioral story behind it: which pages were visited, which content was downloaded, which emails were opened, how long this person has been in the funnel, and what that pattern actually suggests about where they are in their thinking.

Both types of leads deserve a more information before heading over to sales.

And no, the marketing person doesn’t have to spend a half hour putting this together for every inbound lead or newly scored MQL.

All of this can be automated with AI, saving the marketing team a ton of time and enabling the sales team to make a more informed outbound reach.

A base Claude skill for this is bundled in the downloadable zip for paid subscribers at the end of this newsletter

The Post-Discovery Call Package

The discovery call is where the most useful information about a buyer gets generated. It’s also where most of it disappears.

Sales quickly updates the CRM with their memories, including objections, language, competitors, and conversation complexities. Most of this data isn’t structured or shared with marketing.

Marketing aims to improve with each deal but relies on campaign data and assumptions. The buyer’s voice, captured during sales conversations, remains locked in CRM notes.

A call transcript that routes two ways simultaneously solves this.

Sales receives a structured output: pain points, buying signals, stakeholders, risks, and CRM field updates. No more relying on reps to fill in fields they skip.

Marketing receives a parallel output: the prospect’s language, competitive mentions, and patterns indicating messaging gaps or content opportunities. Not the full transcript, but relevant parts.

This approach provides both teams with the same ground truth, eliminating silos and improving market understanding.

Note: This skill assumes a transcription source (Gong, Fireflies, Otter, or manual paste) feeding into it. It handles the synthesis from there.

Why spend time doing all this when AI can handle it easily (and better) for both teams?

A base Claude skill for this is bundled in the downloadable zip for paid subscribers at the end of this newsletter

The Closed Won/Lost Analysis

Most sales and marketing teams have monthly deal reviews.

These meetings often involve three hours of leadership posturing, defensive explanations, and occasional insights that everyone agrees on but no one acts on.

Think differently about these meetings.

Build the analysis before the meeting. The conversation shifts from reviewing past events to discussing future actions. This shorter meeting takes 30 minutes instead of three hours.

Automate the analysis of closed deals to identify objection patterns, competitive mentions, messaging gaps, and commonalities between winning and losing deals. This highlights trends across multiple deals, avoiding the noise of reviewing individual losses.

Include MQL rejection analysis too. Marketing often focuses on closed deals and misses a pattern in leads that sales keeps bouncing back. This feedback loop rarely gets closed due to time constraints.

Automate this analysis to provide marketing with the full picture, including rejected leads and reasons for rejection.

The insights this uncovers: “Deals stall when implementation complexity increases. Marketing over-emphasizes vision and under-emphasizes execution clarity.” This signal is in your CRM but takes too long to manually pull for consistent action.

A base Claude skill for this is bundled in the downloadable zip for paid subscribers at the end of this newsletter

None of these require a dedicated RevOps team or an enterprise budget. They require decent data hygiene and a willingness to define what a good handoff actually looks like for your team. AI handles the synthesis. The humans still have to agree on what goes in and act on what comes out.

If you’ve been running on misaligned leads, lost call context, and monthly reviews that don’t move the needle, this is where the practical work starts.

A Starter Marketing / Sales Handoff Claude System

The skill pack includes a starter Claude workspace with a built-in setup guide that walks you through configuration before you run anything. Three skills are included:

• Lead Brief Generator — turns inbound leads and scored MQLs into a one-page rep brief before first outreach

• Discovery Call Package — processes a call transcript and routes structured output to sales and marketing simultaneously

• Closed Won/Lost Analysis — generates your monthly deal review automatically, including MQL rejection patterns, before anyone walks into the meeting

Drop the folder into your Claude workspace, fill in one config file, and the system does the rest.

Read more

What I’m covering in this newsletter:

• The concept around my new service offering

• The setup I used to start building it out

• Creating the landing page with Claude Code (my first time)

• Developing other resources with Claude Cowork

• [Bonus for Paid Subscribers] How to build out your own ClaudeOS

I’ve had a new service offering bouncing around in my head for a little while, and I finally decided to put some time into getting it launched this week.

I also thought this would be a great way to demonstrate the power of the Claude OS system I’m using and how it can give marketers a real leg up.

First, some background.

With the nature of my business and the niche that I’m in, I often get people reaching out with questions and curiosity regarding how to use AI in their work and in their companies. I had written recently about owners and senior leadership and the disconnect between what they expect from AI and what their employees are going through implementing it.

That drove a business owner to reach out to me through LinkedIn and ask for some of my time. He has been disappointed by the progress his company is making with AI and wanted some insight as to why that might be.

We jumped on a call. I shared my screen and showed him the complexities of what his employees are trying to accomplish with AI. We actually built things on the call. I showed him what’s challenging about AI agents and automations, and how complex it can be to actually implement AI into new and existing workflows. I also showed him where the opportunities are.

At the end, he said to me, “Why are you not charging me for this?”

So here we are!

Let’s walk through my process for launching this and see how heavily I relied on my Claude Operating System (patOS).

Hi, I’m Pat! Glad you’re here and thank you for reading Approachable AI by Pat Schaber! I help businesses transform go-to-market operations with AI. I also write a ton of content about that. Here are some recent articles:

My Base of Business Operation: PatOS

I won’t go into great depth here because I’ve already done a post about my OS operating system concept.

In short, this is the operating system that helps me run my business, and it’s based in Claude Cowork / Claude Code. It knows everything about my business, my point of view, my brand, and allows me to skip right to the point without a lot of prompt engineering.

Here is my current setup.

I jumped in Claude Cowork and off we went.

Most readers of this newsletter have done research with AI, and if you’re in business, you’ve probably done some messaging and positioning work. I’m going to largely skip over that step for this newsletter.

We nailed research, messaging, and positioning in about a half hour!

The Landing Page with Claude Code

I’m going to focus more on the landing page creation with Claude Code because I think that is an unlock for getting website assets created quickly.

My website is built with WordPress and Elementor as my page builder. Every page on my site to date has been built with those two resources.

I’ve typically in the past used Lovable to give me a design concept, and then I recreate it with Elementor. But I’ll be honest, I get annoyed quickly with spending too much time in Elementor building out pages. It’s not where I want to spend my time.

So I gave Claude Code a shot. Here’s how we got started:

Here are some key points on this:

• When I’m using Claude Code, I typically am doing it within VS Code which sounds scary, but it’s really not. We’ll cover that in a future newsletter. For creative work in Claude code, I don’t mind using just the desktop app interface. (remember I’m not a coder)

• You’ll notice I have my patOS workspace selected again. It’s important for Claude Code to build within this framework too, so it has all the context. I don’t have to spend a lot of time providing information on what we’re doing because it already knows my business so well.

• When I was done with the research positioning and messaging in Claude Cowork, I had it prep me a prompt that I could cut and paste into Claude Code. I typically don’t do a lot of chatting back and forth with Claude Code.

• In my PatOS workspace, there is a web page creator Skill. I worked with Cowork to create a skill around the strategy that I want to deploy for creating web pages, so Claude Code was able to reference that.

The prompt gave me a great start on this. The first draft fits my brand very well and follows the web page creator Skill almost perfectly.

Claude Code went ahead and created a web pages folder in my outputs folder of PatOS and placed the HTML file that we’re working on in there.

One thing that might drive marketing people a little bit crazy is that you can’t just go into the text and make edits. Claude did release Claude Design this week. I’ll be demoing that in an upcoming newsletter, but you can make in-line changes in that tool.

But it’s pretty easy here.

You simply click on the pointer in the upper right. Select the section or block of text you want to edit. That section is then referenced in the chat, and you explain what you want done.

You can put a block of rewritten text in there, or you can simply ask it to rewrite it in a certain way.

I spent a little time tweaking text, but really not too long. It knows my brand voice, tone, and the style. It also knows my ideal customer profile and what pain points to touch on, so there wasn’t much to edit.

I had created a booking page in Google Calendar, and I simply shared the embed code for that with Claude Code, and it updated the page easily.

Once finished, I copied the HTML from the page, went into Elementor, and pasted it into an HTML block on a page. Below is my new 1:1 AI Coaching Session page on my website.

Marketing Assets with Claude Cowork

I don’t need as many marketing assets for this as I would need if I was in a corporate marketing team, but I did want a slide for my company PowerPoint deck as well as some potential verbiage for outreach.

Here was exactly what Claude Cowork pushed out with the first draft of this. I haven’t touched a thing on the the slide yet. Not sure I’ll need to touch much!

Email Outreach Drafts

Eventually I’ll be reaching out to some people to let them know about the offering and see if they know anyone that could be interested. Let’s see how Claude Cowork does with drafting up some emails and putting them in my Gmail drafts.

In a couple minutes, I have five emails drafted. You can see an example of one of them. It’s pretty darn close. I’ll definitely be editing portions of it before I hit send, but I’m pretty happy with the start.

Conclusion

When this is all set up properly, I’m so impressed at how efficient and high quality the output of my work is with Claude. Yes, it takes a little bit to build out the operating system that I have, but once it’s in place, it seems you’ve built a really good base from which to launch a product.

How to Build Your Own ClaudeOS

The included guide for paid subscribers this week covers a step-by-step process for building out your own Claude Operating System starting with the important context files.

Read more

What I’m covering in this newsletter:

• What is Clay and why does it pair so well with Claude Cowork

• A daily outbound prospecting set up using these tools

• An example of creating a prioritized, lead-scored prospect list enriched with company and contact data.

I’ve found that one of the more powerful AI connections right now for sales and marketing teams is the integration between Claude Cowork and Clay.

You should be familiar with Claude by now, but let me do just a brief overview of Clay.

Clay is a data enrichment and prospecting platform that pulls from dozens of sources simultaneously to build hyper-targeted lead lists with verified contact info, company signals, tech stack data and more.

For sales and marketing teams, it cuts out the manual research grind and lets you build automated workflows that find, enrich, and prioritize prospects before they ever hit your CRM. Think of it as the research assistant your SDRs wish they had, running in the background 24/7.

Combined with Claude, this is sort of like the dream team. The integration of Clay into the Claude interface is spectacular.

I’m going to walk through two use cases of this powerful integration. I’m using the paid version of Claude, and I do have a subscription to Clay. Although a free account with Clay will get you some initial credits to experiment with.

Let’s dive in!

Hi, I’m Pat! Glad you’re here and thank you for reading Approachable AI by Pat Schaber! I help businesses transform go-to-market operations with AI. I also write a ton of content about that. Here are some recent articles:

Daily Outbound Prospecting

I want to take a very simple example to start, which is daily routine outbound prospecting. This could be for salespeople hitting target accounts or for marketing people preparing targeted lists and customized outreach.

Let’s, for a second, just take a step back and think about what outbound research and prospecting looks like for most small to mid-sized companies.

• Multiple tabs and sources of data for company research and contact/company data enrichment

• LinkedIn searches and profile reviews

• Drafting emails where you try to incorporate your competitive differentiation and positioning, tailored towards their pain points

• Update CRM with new data

• et cetera, et cetera

The manual task list here could go on and on. I’ve seen it first hand. It’s painful at times.

Let’s rethink how this workflow could work.

If you haven’t read my recent article about building an operating system for yourself in Claude Cowork, check that out. It takes what I’m going to show you below to the next level. I promise!

Anthropic wouldn’t be a target for my business, but for the sake of this example, I figured we’ll use them. Throughout the example, I’ve made sure to block out any contact information so we don’t contribute to the people at Anthropic getting hit with a bunch of emails.

I’m literally starting with this simple prompt because Claude doesn’t need any more context; it’s already built into my operating system.

Because Claude knows my business so well, it has some targeted follow-up questions before it gets to work.

What I love about this connection between Clay and Claude so much is that the Clay information is presented so cleanly within the Claude interface. You can see here we’ve already started with a list of contacts, and we can do further filtering or provide further data points in the table, all enriched by Clay.

If I want to zero in on a certain contact, each contact has its own card that has:

• Contact information

• LinkedIn URL

• Summary of work history

• And any other kind of enrichment we have per contact

I’m literally scratching the surface on what kind of information we can have here!

We can then switch over to the company tab and get further data points on Anthropic as a company. You can see a few suggested data points, but there are hundreds more.

Jumping back over to Claude, you can see Claude has prioritized the contacts in Tier One and Tier Two for me and suggested ones to filter out because they doesn’t fit my ICP.

Remember, it knows my ICP and who I’m targeting because that is context provided in my pre-configured operating system in Claude Cowork. I have similar information in clay as well.

Let’s assume I’ve chosen a couple of contacts that I want, and I ask Claude to go ahead and draft up my emails and put them in the drafts folder in Gmail.

I have to give very little instruction here because Claude already understands me and my business.

And voila! We have two nicely drafted emails that are tailored towards the contact and the company and are able to touch on my messaging and positioning for my company.

Now, I would never simply send these. I would go through and edit, verify any key data points, and then send.

That process took me, only a couple minutes!!!

Prioritized Prospect List with Multiple AI Agents

The connection between Claude and Clay allows us to create enriched contact and company information that goes way beyond a standard company name, contact name, and email address that we may have used historically for prospecting.

I don’t do any outbound prospecting for my business yet, but for the sake of this example, I decided to create a small campaign for myself to show the power of these tools.

I started in clay and used their Sculptor tool which allows me just to speak into it and let it know what I’d like to find. Here is what I asked:

Find any Vice President of Sales or Chief Revenue Officers at small to mid-size B2B manufacturers in Minneapolis and St. Paul.

After some tweaking of the query, I wound up with about 30 target contacts for my campaign.

There are literally thousands of types of enhancements and enrichments you can do to the data in Clay, but for this quick example, I chose to have Clay find the tech stack used by each company so I could narrow down people that use HubSpot and Salesforce.

The reason for this is that this shows me that they have an active go-to-market operation between sales and marketing. Then, with the data table in Clay, I can create a column that lets me know which companies are utilizing those technologies in their tech stack.

I didn’t grab contact information quite yet. I’m choosing to export the table, and I’m going to work with Claude Cowork in my operating system to score and prioritize this list of leads. Claude spins up multiple agents at one time to tackle this efficiently and gets to work.

It’s tough to get a great screenshot of the spreadsheet output that Clay gave me, but you can kind of see what it looks like below. I’ve hidden the names of the contacts so that information isn’t public here.

I also asked Claude to go look at the contacts’ LinkedIn profiles to see if they have any posts on AI. This will give me a little more insight into their knowledge of utilizing and implementing AI.

Claude Cowork has given me three tiers for prioritization and has scored each contact within those tiers.

From here, I’ll simply let Claude know which contacts I’d like to get the actual contact information for, and then it will go out to Clay and get that information for me.

I’ll then have Claude draft email outreach and LinkedIn connection notes for me based on my company positioning and messaging and any research it does on the target company.

Hours and maybe days of research and information compiled in minutes, and a campaign underway in just an hour or so.

I honestly believe that combinations like Claude Cowork and Clay really even the playing field for go-to-market operations between sales and marketing. What used to take multiple meetings, a lot of research time, and a lot more money can now be done in hours with minimal budget spend.

Connect Clay into Claude Cowork and give it a shot!

Ways to Work with Me

Quick Win Session
A focused, high-impact working session designed to identify one immediate AI opportunity and give you a practical roadmap you can implement right away.

AI Readiness Assessment
A structured evaluation of your people, processes, data, and tools to determine where AI will create the most value and what needs to happen before you scale.

Workshops & Projects
Hands-on, team-based engagements that turn AI from theory into execution through tailored workshops, use-case design, and guided implementation.

Fractional AI & Marketing Leadership
Ongoing executive-level support that embeds AI strategy, operational rigor, and revenue-focused marketing leadership directly into your business.

AI Tech I’m Currently Using

I get quite a few questions on Substack, LinkedIn, and through my website on what AI tools I use for what use cases. I’ll try to share a few in each newsletter so it will give you some ideas of tools you can try for specific purposes. Here are a few for this week:

Everyday LLM - ChatGPT and Claude: I cover a lot of this above so won’t dive in too far here.

Call and Meeting Transcriptions - Granola: I like Granola for a couple of reasons. It takes the transcription and rolls it up into very good notes and action items but doesn’t need to be added to a meeting as an attendee. Also, it syncs great with my Hubspot CRM which makes it very easy to send notes to the contact record.

Presentations / Branded Documents - Gamma.app: I use this 3-4 times a week for anything I do with client presentations or professional document creation. I cover why I’m a heavy Gamma user in this post:

Video Recording / Editing - Descript: I’m trying to work in time to my weekly routine for video creation. I feel it’s important to give people a more personal connection. Descript is a huge timesaver. Imagine being able to erase or add text from a script and have it be reflected in the actual video in seconds. Unbelievable.

Sales Prospecting - Apollo.io and Clay: I don’t do a ton of outbound prospecting for my business but I’ve been using the free versions of Apollo.io and Clay to find companies that I could get to know through LinkedIn that may be interested in my services. I’m only touching on a sliver of the capabilities of these tools.

What I’m covering in this newsletter:

• Why business leaders don’t care about AI agents and what they actually do care about.

• A real-life example of an outcome that gets business leaders excited.

• Tips and tricks for finding the right processes and workflows to impact with AI.

• [Bonus for Paid Subscribers] My full guide on uncovering the best process candidates for AI and how to fully prepare for the implementation.

I’ve talked with enough customers and prospects that I can honestly tell you businesses don’t care about AI agents. They aren’t paying for AI agents.

Hear me out here.

Customer #1 sees my newsletter and says, “Oh wow, that automation you built with Claude is really cool. Do you mind showing that to our marketing VP? I think we could use that.”

Customer #2 sees the same newsletter and says, “We’ve really been struggling with ABC. I think that Claude Automation you built could fix it. Do you mind showing us how that might work?”

Customer #1 may like what they saw, but it’s unlikely that they’ll actually purchase anything.

Customer #2 envisions an outcome. They see the potential to fix a friction point. They get it.

Customer #2 will likely buy THE OUTCOME.

Real-Life Example

Let’s take a look at what an actual real-life situation looks like where you find a process with friction points. A common one I hear about often is sales proposal development.

I’m going to outline why this could be a friction point and what an AI agent and automation output would look like. (this is something I built to solve a similar problem)

Common Proposal Development Process:

• Salesperson wraps up the final discovery call with the prospect.

• Compiles and combines all their notes on this opportunity from various different sources.

• Updates the CRM record for the opportunity. (Hopefully)

• Meets with the sales manager to start the discussion on proposal strategy, which likely turns into a multi-team meeting.

• Salesperson opens up a Word document and begins to write the proposal. (May or may not use brand templates and messaging)

• Salesperson sends the proposal around for review and receives feedback, then goes back into editing mode.

• Salesperson works on a presentation deck to present the proposal to the customer.

• Salesperson redoes the deck after realizing they weren’t using the most updated brand template.

This may seem like the norm to some sales teams and companies because this is just the way it’s always been done. They’ve come to accept the time needed and questionable output of the proposals.

Just think for a minute what the opportunity cost is of having the salesperson spend all that time on all of those steps when they could be already working on the next opportunity with another prospect?

Now let’s say that there was a business leader that understands that AI can help them do things more efficiently and with higher quality outputs. They’re educated on AI and they realize that this is a prime example of a process with many friction points and inefficiencies.

Hi, I’m Pat! Glad you’re here and thank you for reading Approachable AI by Pat Schaber! I help businesses transform go-to-market operations with AI. I also write a ton of content about that. Here are some recent articles:

Now let’s envision what that same task looks like with the process rethought and AI integrated.

• Salesperson gets off the last discovery call before proposal development.

• The call has been recorded and automatically updates the deal in the CRM which includes all the other notes that were automatically put in for this opportunity.

• Salesperson opens up Claude Cowork, selects the AI Agent folder (that Pat built), turns on the proper connectors, and puts in a simple prompt.

• An AI Proposal Automator Agent is activated that contains:

  • Full Agent instructions (CLAUDE.md)

  • Everything the agent needs to know about the company, the brand, and the product messaging and positioning (context files)

  • Examples of what a fantastic proposal looks like

  • The skills that were built to do the proposal generation

  • Exact templates that are on brand and in the right layout and format that the agents should use

• Agent gets to work (maybe 3 minutes of work):

  • Finds the opportunity information from the CRM and gets updated on all the history

  • Reads and understands all the context provided

  • Gets to know what a great proposal looks like and the templates it should be using

  • Builds proposal and proposal presentation

  • Puts documents in the assigned location

  • Sends updated material to a Slack channel so salesperson and sales manager can begin review.

• Salesperson and sales manager review and either edit directly in the new documents or utilize Claude to do the edits

• Salesperson finalizes the document and sends it to the customer

In this example, the AI agent’s work took maybe three minutes. the output had the correct messaging and positioning and was on brand with company requirements.

Now, scroll up to original proposal process and just check off how many things were replaced by the Agent.

Note: No human was replaced by an AI agent here. Instead, the human was able to work on other opportunities and prospecting while the agent did the work. The human also reviewed and edited before sending to the customer.

All we did was cut out the manual grunt work associated with proposal creation and likely increased the quality of the output as well.

A visionary business leader educated on the power of AI sees a process like we outlined above and understands how AI can eliminate the friction and inefficiencies in it.

How to determine the right processes to work with

The #1 mistake I see with AI implementation: starting with the tool.

Most teams pick a shiny new AI product, layer it on top of how they already work, and wonder why nothing changes. The reality is AI doesn’t fix broken processes. It speeds them up. That means if your workflow is a mess, you’re just going to have a faster mess.

The people getting real results from AI share one habit: they start with the workflow, not the tool.

That means finding the process that’s actually painful (not the one that looks impressive to automate), mapping how it really works versus how the SOP says it works, and asking hard questions about data quality, security, and ownership before a single tool gets selected.

Only once that work is done does the tool decision make sense because now you know what you actually need.

Paid Subscriber Bonus!

I put together a full breakdown of how to do this right, from finding the right workflow all the way through implementation and building in a feedback loop so it doesn’t drift.

The guide covers:

• Phase 1 Find the Right Workflow

• Phase 2 Audit the Workflow Before You Touch It

• Phase 3 Run an AI Readiness Check

• Phase 4 Fix the Process Before You Automate It

• Phase 5 Choose the Tool Last

• Phase 6 Run a Narrow Pilot First

• Phase 7 Build in the Feedback Loop

• What to Avoid Across All Phases

Read more