Some years ago, a CFO brought us in to rationalize two tax functions.

His company had grown by acquisition and ended up with two of nearly everything, including two separate tax departments serving two divisions. They reported differently, staffed differently, and — this was the part that mattered — thought differently about what a modern tax function should be. One wanted to consolidate onto a single platform and centralize. The other saw itself as already effective and had no interest in being absorbed. Left to themselves, the two VPs would never agree.

So, I did what any consultant does when organizational peers don’t agree: I looked at the org chart and engaged the CFO to become the project sponsor, not either VP. That felt like solid ground. When the sponsor sits above the disagreement, you tell yourself the politics can be handled.

They were not handled. What I had not priced was the CFO’s own appetite for the disruption he had just paid us to design. One of those two functions was quietly delivering real value — cash tax savings and a defensible position that supported enterprise value — and when our recommendations were perceived by one of the VPs as potentially threatening how he produces that value, the CFO’s enthusiasm cooled. He had signed the engagement. He had never signed up for the fight. Our plan was, in the language we use privately, dead on arrival.

I want to be precise about what went wrong, because it is not the obvious thing. The analysis was sound. The design was right. What I misread was a person — and specifically, I mistook his authority for his will. The signature that made me confident was the very thing that killed the project.

Subscribe now

Reading You Is the Real Work

I have been on the selling and delivering side of tax and finance transformation for a long time, and I have come to believe something that never appeared in any proposal I wrote: when you hire a firm, the solution described in your RFP is the smaller half of what changes hands. The larger half is an outsider’s assessment, formed in those first meetings, of whether your organization will actually let the solution happen. Reading your politics is the real work. Your scope is mostly just the part we write down.

That read is happening whether we admit it or not. While you are walking us through your requirements, we are watching something else. We are watching whether the people around the table actually agree on what this project is for or are quietly describing four different projects. We are watching whether the executive who called the meeting can spend the political capital the work will require or is hoping the work will spare them the fight.

We are watching whether there is one organization here or, as in that CFO’s company, two that will never reconcile. And we are watching for the hardest thing to voice: whether you bought transformation but are only prepared for improvement.

Who Pays for the Politics

None of this is in the RFP, and almost none of it makes it onto an invoice. This is the part companies misunderstand about the cost of their own politics. You assume that if the read is bad, it shows up as a higher fee. Occasionally it does. Far more often it shows up somewhere you cannot see it — in a plan that quietly narrows from “transform” to “install,” in a delivery team absorbing friction that was never scoped, in a result that lands at a fraction of what was promised and gets filed under “well, these things are hard.”

That is the real Political Premium, and here is how I would put it to a client if the incentives ever allowed me to be that blunt: you think you are paying in dollars, but in most instances, you’re paying in “suboptimal” outcomes.

Share

When the Best Signal Fools You

If reading politics is the job, you would expect someone who has done it for decades to be reliably good at it. Here is the more honest confession: the signals you rely on to make that read are themselves unreliable, and the best-looking signal is often the one that fools you.

A large bank once engaged us to build a customized tax technology roadmap — the kind of document that, fully implemented, would remake how the function operated. Their RFP was excellent. It was so well constructed that it mirrored our own workplan almost point for point, and that gave us real confidence. A buyer who can write that RFP, we assumed, knows exactly what they are buying and what to expect from us.

The RFP, it turned out, had almost certainly been written by procurement or someone in IT. The person who would actually decide what happened to our roadmap was the VP of Tax, a career tax attorney who had no framework for what to do with it. We saw this partway through and did the responsible thing — we slowed down and taught, walking him through each recommendation, its cost, and its value.

In the end, it didn’t matter. Faced with a roadmap full of things he didn’t fully understand, he did the rational thing a careful professional does with unfamiliar risk: he chose the one project he understood. He approved the tax accounting automation and left the rest. By my estimate, he captured about twenty percent of the value the roadmap represented.

I’m not telling this story to fault him. Caution in the face of the unfamiliar is a reasonable instinct for someone personally accountable for a function that cannot afford to be wrong. The failure was ours. We read the quality of a document and inferred the readiness of a person, and those were two different people entirely. We never made the roadmap safe enough for the one who would actually have to say yes.

Both of those stories share a principle. In each, the thing that made us confident was an illusion dressed as a green light. A CFO’s sponsorship that was authority without will. A flawless RFP that was a document without a buyer behind it.

And this is the part worth considering, because it is not really a story about consultants: the signals your organization uses to reassure itself that a transformation is safe — an executive sponsor’s name, a rigorous procurement process, a detailed roadmap everyone admired — are frequently the same signals that hide the political reality underneath.

They manufacture confidence on both sides of the table at once. You feel covered. We feel confident. The project proceeds toward a wall that either of us might have seen, if the signals had not been so reassuring.

Readiness Is the Fault Line

Underneath all of it is a single fault line. You were sold transformation, and transformation is not a technical event. It is a demand that an organization change what it does, who decides, and what its people are accountable for — and change is political all the way down.

No roadmap, no platform selection, no beautifully scored RFP tells you whether the organization is actually willing to be changed. Both of my failures were readiness failures wearing the costume of a well-run project. The tell was there in the first meeting each time. I simply trusted the green light over the tell.

In the transformation market, the buyer holds the private information about their own consensus, their own sponsor’s spine, their own readiness to change — and, understandably, keeps it hidden. You would never open an RFP by admitting your two divisions will never agree, or that your sponsor prefers the status quo. So, the consultant prices and plans against a picture they know is incomplete, and everyone proceeds on borrowed confidence.

Name Your Politics First

Which points to the one thing I wish more companies understood, because it would move outcomes far more than any tool selection ever will. The instinct is to hide your politics from the firms you are hiring — to present a united front, a decisive sponsor, a clean mandate — because you assume disclosure is a weakness a vendor will price against or exploit. The reverse is true. The politics you hide do not disappear; you simply pay for them later, invisibly, in a diminished result.

The buyer who says the quiet part out loud before the RFP — our two functions have never agreed on this; our sponsor is enthusiastic but will not overrule the head of the division; our real decision-maker has never owned technology and will need to be brought along — has not weakened their position. They have handed the consultant the one piece of information that lets the work be priced honestly, staffed for the real fight, and sequenced around the resistance instead of driven straight into it.

And distrust your own green lights. The sponsor’s name on the engagement letter is not the same thing as the sponsor’s willingness to spend capital when the resistance arrives. The pristine RFP is not the same thing as a buyer who knows what to do with what they asked for. When a signal makes you feel most covered, that is precisely the moment to ask what it might be hiding.

I spent years on the other side of that table, reading organizations for a living and quietly repricing engagements on what I saw. The most useful thing I can tell you, now that I am not selling you anything, is that the outcome of your next transformation was largely decided before the first system was ever configured — in a meeting where the politics were already plain to the person you were hiring and already hidden by the person signing the check.

Your roadmap didn’t decide it. Your politics did.

Every significant technology shift of my career reached the Tax function the same way -- late.

By the time Tax was asked about the ERP, the chart of accounts was frozen, and the data model had no room for the detail we needed. By the time Tax got onto the automation program, IT had already picked the platform and assigned priorities. By the time “AI strategy” became a standing agenda item, finance had defined the use cases and Tax was left arguing for scraps of the budget. Each time, Tax spent years retrofitting, reconciling, and explaining why it should have been asked sooner.

I’ve watched this repeat often enough to stop treating it as bad luck. Even though Tax is arguably the largest consumer of enterprise financial data, it’s also a downstream consumer, so we tend to inherit decisions rather than help make them.

Quantum computing could be a rare exception. It’s a technology far enough from commercial reality that being early is still possible. One narrow corner of it is already being sold today, and I’ll come to that, but the capability that will actually change how a Tax function works is still well over the horizon.

That’s why I’m writing about it now, well before it can do anything useful for a corporation. Not because Tax needs a quantum strategy this year. Because the learning curve is steep, the vocabulary is genuinely difficult, and the only way to be present when the enterprise starts is to have done the reading first.

So let me try to make the reading easier.

Subscribe now

What is Quantum Computing, very briefly

Let me state the obvious – I’m no expert. And so, my novice interpretation may be exactly what you need for now. That idea is the source of my courage to write this so here goes.

A normal computer stores information as bits, each one a definite 0 or 1. A quantum computer uses qubits, which can hold a blend of 0 and 1 at the same time (superposition), and which can be linked so that their states depend on one another (entanglement).

A well-designed quantum algorithm arranges these blended possibilities so that wrong answers cancel out and useful ones reinforce the way water waves can cancel or amplify each other (interference). At the end you measure, and the machine reverts to ordinary 0s and 1s. The craft is in the arrangement, because measurement gives you only one classical answer, not the whole universe of possibilities.

To be clear, a quantum computer does not try every answer at once and hand you the winner. It manipulates a vast space of possibilities so the odds tilt toward the right one. That’s as nuance that feels relevant and potentially important although I can’t honestly describe why or how for now.

Two things follow, and both matter for how you think about this. First, quantum will not replace classical computing. Your close won’t run on it, your ERP won’t run on it, your spreadsheets certainly won’t.

The realistic model is hybrid: classical systems do nearly all the work, and a quantum processor acts as a specialized accelerator for a narrow class of problems, closer to how a GPU handles graphics than to a general-purpose CPU. Second, and I’ll say this plainly because the hype runs the other way, quantum is not “bigger than AI.” AI touches almost every information task in the company. Quantum is a specialized instrument that may produce extraordinary results in a few domains and remain irrelevant to everything else.

The clearest of those domains is chemistry. Nature runs on quantum mechanics, and simulating molecules on classical computers gets brutally hard as they grow. A quantum computer is itself a quantum system, so it may eventually simulate molecules far better than we can today, which is why pharmaceutical, materials, and energy companies are the ones paying real money to watch this space. That’s a genuine use case. It’s also nothing like a Tax problem.

One more distinction, because the timeline depends on it. Everything I’ve described so far is the universal, general-purpose quantum computer, the kind that might someday simulate chemistry or break encryption, and the kind that is a decade or more from doing either reliably.

There is a second, narrower kind: the special-purpose optimizer, built to do one job, which is to search enormous combinatorial spaces for good enough answers. That kind you can rent today and is the qualifier to any “quantum is years away” argument.

What’s happening right now

Today’s universal quantum computers are rudimentary and error prone. Qubits are so fragile that heat, vibration, or stray radiation destroys the calculation, and the field’s central challenge, error correction reliable enough for large problems, is unsolved. Real machines exist and you can rent time on them through the cloud, but no one can tell you whether the commercially decisive moment arrives in three years, seven, twelve, or more.

The special-purpose optimizer is a different story, and it has a name: D-Wave, a public company that has sold quantum annealers for years and markets them as ready for real business today. The customer stories are specific: a food distributor cutting the time to build complex staff schedules by roughly 80 percent, an automaker producing factory schedules 83 percent faster, telecoms tuning their networks. This is the strongest case that quantum is already useful.

Those are impressive results, yet it remains difficult to ascertain what came from the quantum processor itself, from a quantum-enabled hybrid that leans heavily on classical computing, or from a very powerful classical computer that would have produced much the same answer on its own.

Where it may touch Tax first

Quantum’s first real contact with Tax will probably come through data security. A powerful quantum computer could eventually break the public-key encryption that protects data in transit today, which is why standards bodies have already published post-quantum cryptography, classical encryption, running on ordinary computers, built to survive a future quantum attack, and why enterprises are beginning to migrate to it.

The reason this matters before the machines exist is a strategy called “harvest now, decrypt later”: an adversary steals encrypted data today and simply waits to read it once the hardware arrives.

Tax holds exactly the kind of data that is still damaging a decade from now. Tax returns, transfer-pricing documentation, cross-border filings, and the financial and idiosyncratic detail behind all of it. When the enterprise runs its post-quantum migration, Tax should be asking which of its data flows are covered and which are exposed. That is not a project for this year. But it’s the first question worth knowing you’ll eventually need to ask.

Why bother now

Because the alternative is the example I opened with. If Tax waits until quantum is real enough to be obvious, the enterprise will already have decided where it fits, who owns it, and whose data gets protected first and Tax will once again be retrofitting a decision it could have helped make.

The technology is a decade or more from mattering to us. The habit of showing up late is here today. This is one wave worth getting ahead of.

In a Transformation, AI’s Payoff Is Time

Last week I drew a line between two ways of thinking about AI in tax: transforming your function to use AI and using AI to run the transformation itself. This piece is about the second, specifically about execution, the day-to-day operation of a program once it is underway.

Start with the number that should bother anyone who runs these programs. McKinsey’s* long-running research finds that fewer than a third of transformations succeed at both improving performance and holding the gains, a figure that has barely moved in years, despite better tools and methods. Transformations rarely fail for lack of analysis. They fail on adoption, alignment, and judgment, the human work.

The connection most people miss is that programs have always used their people’s time poorly — and the human work is what pays the price.

So where does the time go?

On most programs, a large share of skilled people’s time is spent producing the view of the program rather than acting on it. That’s a critical distinction. An unnecessarily disproportionate effort is spent on project management activities like:

• Assembling status across workstreams

• Chasing updates and rebuilding the steering deck each cycle

• Reconstructing the benefits case after assumptions move

• Preparing for meetings and tracking the actions afterward

Look, I firmly believe in the value project management professionals bring to transformation programs. The point I’m building up to is simply – they can be and should be using their time very differently. In many instances PMO activities represent the operational load that sits on top of transformation workstreams — and it consumes hours that could be devoted to looking forward, anticipating future challenges in lieu of reporting on historical progress.

Subscribe now

What to automate

This is where AI does real, unglamorous work. Pointed at the operational layer of a program, it can handle:

• Status collection: drafting workstream status from source systems, trackers, and meeting notes instead of by hand.

• Risk and dependency detection: surfacing issues from the program’s own data rather than manual log review.

• Reporting: generating steering-committee materials from live data instead of rebuilding them each cycle.

• The benefits model: keeping the financial case continuously current instead of stale between gates.

• Meeting mechanics: preparation, action capture, and follow-up.

• First drafts: current-state analysis from process documentation, and stakeholder communications tailored by audience.

BCG reaches the same conclusion from the top down. In its “agentic transformation office,” program tracking moves from periodic validation to continuous transparency. But BCG’s version is a corporate build, meaning a new, AI-enabled Transformation Office, staffed and mandated by the CEO, on a horizon that runs to 2030. That’s not what I am describing. Even a single-person transformation team running a single project can start using AI this way now, with tools already available, and without waiting for the organization to stand up anything.

What the reclaimed time buys

The automation is the means. The value is what the freed hours go to, the work that has always been underfunded because time has been a tightly constrained asset:

• Judgment: deciding what the information means and what to do about it, rather than assembling it.

• Alternatives: the options analysis that gets cut under deadline, so programs settle for the first workable answer instead of the best one.

• Change management, sooner: affecting behavior while there is still time to change it, not as an afterthought at go-live.

• Communication: more frequent, better-tailored, more candid contact with stakeholders.

These are the activities that separate transformations that hold from ones that stall. Return the time to them and the program gets stronger precisely where it usually fails.

The discipline that makes this real: Reclaimed time only counts if it is deliberately reinvested in the human work. Bank it as a headcount cut and you get a cheaper program, not a better one.

The trap

There are three easy ways to get this wrong, and each one wastes the opportunity.

The first is using AI to run the old process faster without changing what you do with the time it saves. You generate the same monthly status deck in a day instead of a week but if those saved days don’t go into judgment, analyzing options, change management, or communication, you have simply made a clerical process quicker. The program is no better; it’s just cheaper to administer.

The second is letting constant visibility become constant noise. When AI can surface every risk, update, and variance in real time, a team can drown in signals and stop acting on any of them. More information is not the same as more insight, and a stream of alerts that no one has time to interpret is worse than a monthly report someone actually reads.

The third, and most serious, is treating the model’s output as the decision. AI can draft the analysis, flag the risk, and recommend the option but a person still has to weigh it, apply context the model does not have, and own the call. AI lowers the cost of producing the picture. Deciding what to do about it remains human work.

What’s Next

That’s the case for why execution changes and where the value comes from. Next week, the how, stage by stage across a program, where to point AI and where a person still has to make the call.

*Source: McKinsey & Company research on transformation success rates (fewer than one-third of transformations succeed at both improving and sustaining performance).

Almost every conversation about AI in Tax right now is about the same thing:

1. How to get AI into the Tax function.

2. Which processes to automate.

3. Which tool to buy.

4. How to govern it.

5. How to redesign the operating model around it.

The assumption underneath all of it is that AI is the thing you are moving toward, the destination of the transformation.

There is a second question, that’s equally important. Not how to transform your Tax function so it can use AI, but rather, how to use AI to run the transformation itself.

Those are two different projects. The first treats AI as the outcome: at the end of the program, the function does things with AI it couldn’t do before. The second treats AI as the tool you use to get there, in the diagnosis, the current-state assessment, the roadmap, the stakeholder analysis, the business case, the monitoring after go-live. One is about where you are going. The other is about how you get there.

Subscribe now

The odd part is that the people who run transformation programs have adopted the first question and largely ignored the second. Tax leaders hear a lot about how AI will change how their people work, and then we plan that change the same way we planned it ten years ago: weeks of interviews, a current-state deck, a future-state deck, a spreadsheet roadmap, a business case built once and never touched again.

Using AI as Transformation Enabler

I’ll be concrete about what the second question looks like. A tax operating model redesign usually opens with a current-state assessment — how work moves across regions, where the manual effort sits, which controls are real and which are theater.

That is weeks of interviews and document review before anyone can even see the picture clearly. A LLM that can read process documentation, prior audit findings, and system configurations, and produce a first draft of the current state in days, changes what the assessment phase is.

The judgment about what the picture means still belongs to a person. The labor of assembling it doesn’t have to.

A Perpetual and Real Time Business Case

Same with the business case. Most transformation business cases are a single moment built to get funding, presented once, then quietly they go wrong within a quarter as assumptions move.

A business case you can keep current, because the cost of rebuilding it is low, is a different instrument. It stops being a slide you defend and becomes something you actually steer with. This is the point that matters, and it’s a management point, not a technology one: the same tool that changes how a Tax function operates can change how you run the program that gets it there.

Speed, Cost & Judgement - The AI Value Prop

A leader who only sees AI as the destination has chosen the slower, more expensive version of the job for no reason. Not because they lack the tool but because they never thought to direct it at the transformation program itself.

Using AI to run a transformation does not remove the hard parts. The judgment about what to change, the political work of getting people to agree, the discipline to say no to the low-value automation everyone wants.

Those are untouched. If anything they become the whole job, because they are what’s left when the assembly work gets cheaper. That’s a real shift in where a leader’s time should go, and it deserves its own treatment. So that’s where this goes next. If you can genuinely use AI to run a transformation, the transformation itself works differently. The phases don’t take the time they used to, and the constraint moves from doing the analysis to judging it.

Coming in Part II

What actually changes, and where a leader should put their attention as a result, is next week’s piece. For now, the distinction is enough, because most people in this field have only been asked one of the two questions. You are being told, constantly, how to transform your function to use AI.

You are almost never asked whether you’re using AI to transform.

Every tax executive greenlighting a multi-year transformation program is making a bet that their program will beat the odds and actually succeed.

Do Transformation Programs Deliver?

Here’s the data:

• BCG’s 2020 study included 825 executives across 70 engagements, with defined success criteria and concludes that only 30% of transformations delivered full, sustained value.

• Prosci’s research on 2,600 change practitioners shows that success tracks change-management quality almost exactly – of the one-third who report success, 88% of them attribute it to excellent change management. Yet most organizations rate their change management effectiveness as “fair” at best.

Tax’s own data tells the same story. KPMG’s Global Tax Function Benchmarking Survey asked leaders why their transformations failed. The most frequently cited answers: resistance to change (29%), lack of skills (31%), thin buy-in (28%), and budget (28%).

Technology wasn’t the leading cause. Only 23% of tax functions have anyone dedicated full-time to running transformation. Can your team build organizational capability they’ve never had to build before, on top of their day jobs? Will you be surprised when it doesn’t stick?

The data says Tax Functions aren’t good at delivering transformation programs and worse at making them stick.

Consider these questions during the earliest possible phase:

1. Do you believe you can succeed where most fail? Why?

2. Are you OK investing multiple millions of dollars in a transformation program that statistically is unlikely to deliver its budgeted ROI? Why?

3. If successful transformation is largely dependent on your organizational change management effectiveness, why do you believe that competency is well developed among your team?

Subscribe now

That’s the case for outsourcing. The idea is to transfer the transformation risk to an external service provider. Preferably one who can:

• Spread its technology investment across a broad portfolio of its clients in a way your company can’t, and

• Become highly proficient at transforming itself through repetition and practice. Doing it so often that transformation itself has evolved to become a core competency to a degree your company could never achieve on its own.

But this isn’t “outsourcing beats transformation.” It’s a real trade-off, and it deserves scrutiny beyond a simple cost comparison.

Best in Class or Best in Cost

Go into this with a philosophy. Categorize your Tax Function deliverables into two buckets - (1) where you believe you should be Best in Class, and (2) those were cost matters more. Return compliance, provision mechanics, statutory filings — production work that requires minimal judgment are good candidates for outsourcing or, preferably, a hybrid approach.

Controversy strategy, planning positions, anything that requires interpreting ambiguous law or managing a relationship with a tax authority — that doesn’t survive the handoff. This gets decided deliverable by deliverable.

Hybrid Approach (i.e., Co-Sourcing)

This is where it gets tricky. A hybrid approach requires hard work to decide what to outsource and what to retain. In many instances, that doesn’t mean a clean break along deliverables but requires dissecting the process to retain judgment, policy, and risk assessment and outsource what’s left. That’s a tough tightrope to walk but well worth the effort.

Balancing the Tradeoffs

Outsourcing requires giving up things that should be weighed against the cost savings. Here are a few things you may be giving up:

1. Training ground disappears — compliance work is where junior tax professionals build the pattern recognition that becomes judgment later; move the work out and you move that development out with it. You need a plan to counter this effect.

2. Institutional knowledge may erode — outsourcing providers aren’t obligated to hand back documented workflows when a relationship ends, so switching or reversing course usually means rebuilding from scratch, not reclaiming what you had.

3. Cost savings may evaporate — coordination overhead and contract change-orders routinely erase what looked like savings on the pitch deck. We call it scope creep, and it’s real.

None of this means don’t do it. It means treating the decision with the weight it deserves. Here are five questions worth answering before outsourcing a process, deliverable, or outcome:

1. Judgment intensity. Is this production or judgment? Everything else hinges on the answer.

2. Talent pipeline. Can we still develop future leaders without the experience that builds judgment?

3. Institutional knowledge. What do you lose access to, and how much of it actually comes back if you reverse course?

4. Reversibility. Financial terms of an exit, data portability, IT infrastructure carryover (e.g., AI agents that execute processes specific to your data and systems) — can you actually take the work back?

5. Real economics. Push for transparency — it matters. Here’s what you need to know:

• Is the provider earning a profit or did they “buy” the work? Either way, what are the implications for you?

• What approach is the outsourcing firm taking to ensure their margins grow over the full term of the contract (leveraging their proprietary tech stack, off shoring, sell more consulting services, raise fees, price incremental growth in deliverables at higher margin, etc.)?

• What are expectations for Tax advisory work? Are there required minimums, or loose expectations?

• Consider M&A activity – what happens to the overall fixed fee if you acquire or divest of operations?

• Is the implied rate the same globally, or are member firms outside the US free to change fees if the economics don’t pan out? What if the concept of “rate” disappears? How will you benchmark cost?

AI introduces another dimension – the firms will expect to pocket the margin improvement resulting from their AI strategy. Are you comfortable watching their margins expand while your costs remain static or increase with inflation?

Conclusion

The odds of a Tax Function leading a successful transformation program simply aren’t very good, which is exactly why outsourcing enters the conversation. When it does, don’t stop at the deliverable level. Get granular to the level of process components so you retain the critical judgment that shapes your tax policy.

A hybrid approach can transfer the transformation risk, manage cost, and preserve your hold on judgment, risk management, tax policy, and governance.

It won’t be easy. It should be worth it.

I have always been drawn to history, not as nostalgia, but as a learning mechanism. So, what can we learn by looking back at the modern history of Tax transformation? How can that insight instruct our priorities today?

To facilitate this discussion, I observed three distinct waves occurring in the Tax transformation field beginning shortly after the Sarbanes-Oxley legislation became law, let’s say 2004, two years after SOX was enacted. The graphic below defines each wave.

Wave One

In 2004, I helped start PwC’s Tax Function Effectiveness practice. I’d been leading a tax compliance managed services team when Sarbanes-Oxley changed everything for the VP of Tax and their CFOs. For the first time, Tax was subject to controls audits, and in the early years, Tax was consistently among the top three causes of a material weakness. CFOs who had never spent a dollar on tax process suddenly had a reason to. Tax function management consulting, at scale, was born.

Wave One was all about enabling the Tax function to operate better. We focused on process improvements for tax accounting, tax provision system implementation, upgrading spreadsheets to handle data more efficiently, SOX controls remediation, and standing up shared service centers and outsourcing arrangements.

It was a period dominated by incremental progress that felt good enough to satisfy the regulatory environment. The leaders of the global Tax function were largely attorneys who lacked adequate exposure to technology, data collection methods, and the broader science of process improvement. In most cases, this led to doing the bare minimum because Tax attorney leadership couldn’t effectively present the business case and ROI for something like ETL technology (extract, transform, and load).

Wave Two

Ten years later, a different challenge arrived, spawning the second wave. BEPS asked multinational companies to document transfer pricing consistently across every jurisdiction, defensible on demand rather than reconstructed once a year in a spreadsheet. The ERP upgrade cycle sweeping through finance in that same window supplied the technical means to actually do it.

Operational Transfer Pricing, running transfer pricing through live intercompany transactions instead of after the fact, was where BEPS and the ERP upgrade met. The work built in Wave One didn’t fail. It was just designed for a different set of requirements.

Keep in mind, this was an era of uninformed Tax leadership.

A senior PwC Tax Partner once asked me during this wave, “What happens to your practice once you’ve transformed all the Tax functions?” A famously naïve question I won’t ever forget because it provides honest insight into the prevalent thinking during that decade.

Wave Two focused on ERP configuration to capture tax-sensitive data at the transaction level. The Tax Data Warehouse received plenty of attention as companies sought ways to leverage their investment in enterprise financial reporting, consolidation, and ETL systems.

ETL was a real novelty and largely unknown to the Tax function that had been spending more than 50% of its time collecting, standardizing, and validating its data – the precise automation provided by ETL applications their company had licensed for years.

Another hallmark of Wave Two teaches us an important lesson about the early days of today’s Wave Three: the realization that in too many instances, the Tax function’s technical requirements had not been considered when ERP and other financial reporting systems were implemented or upgraded.

This bias was rooted in two concurrent circumstances:

1. Tax leaders being unfamiliar with the possibilities a tax-sensitive ERP can deliver, and

2. Finance transformation leaders who naturally avoided making a large, expensive project even bigger and more expensive by including the requirements from Tax.

The default had been to leave Tax out and catch up later. It proved to be a costly precedent with implications for today’s AI build-out.

Wave Three

Today, we’re in a third wave created by the promise of AI and global tax policy changes like e-invoicing and Pillar Two. This time the question is explicit: how does Tax operate inside an AI-enabled organization, not just whether Tax adopts AI tools of its own.

That’s an important distinction. I’m not speaking about the need for Tax to adopt AI — that’s a given. I’m talking about how the Tax function integrates and participates in the projects IT and Finance are already running to build that capability.

There is real risk that finance will AI-enable its core accounting, close, consolidation, forecasting, and reconciliation systems faster than Tax can independently validate that those changes still produce accurate tax reporting, at the same time finance is reducing the junior-level staff that used to absorb exactly that kind of manual workaround.

If history repeats itself the way it did in the ERP era, when finance implemented or upgraded the ERP with little or no participation from Tax leadership, how will Tax respond when asked to certify outputs from an increasingly AI-automated finance function without a real way to audit how those outputs were produced? What implications does that have for the quarterly and year-end close cycles?

What does all this mean?

We now have three waves, on a clean decade cadence, each harder and more expensive than the last. I don’t think that’s a coincidence, and I don’t think it suggests transformation programs failed.

The people who built SOX-era controls and BEPS-era data models made sound decisions for the conditions they were given. What none of us identified correctly, in any of the three waves, was what the pattern was telling us: this was never a project with an end date.

Rather, Tax transformation is better characterized as a prevailing and constant operating environment for the Tax function ever since SOX created a need for Tax transformation in the first place. We just kept calling it a project because that’s the only language budget cycles recognize.

Mapped side by side, the three waves look like this:

Two Takeaways

First

A function that treats transformation as a temporary phase puts a temporary owner in charge of it. A function on its third wave in twenty years needs a permanent, full-time Director of Change Management, with the same standing as the executives running International Tax, Provision, or Controversy.

I’ve made this case before, and the pattern is the strongest evidence yet that the role was never just for one program’s duration.

Second

Wave Two’s lesson is repeating itself with AI, only faster: Tax can’t keep finding out about finance’s technology decisions after they’re locked.

The twenty-year history makes one thing clear: missing a seat at finance’s AI strategic planning and AI governance table doesn’t have to happen. I’m not referring to implementation or design teams. Those are important too but I’m calling out a role for Tax in AI strategic planning and governance specifically.

I know this sounds incredibly obvious -- get involved early and meaningfully. But not doing so has happened before. We can see it on the transformation timeline.

So, this is a friendly heads-up. This time will be different. Right?

Leading the effort for enterprise AI cost management

Most Tax executives believe their AI spending is contained. They’re tracking a vendor invoice, a license, maybe a usage tier and the number looks manageable, comfortably within their cost center budget. The actual cost, however, is more extensive and more troubling: at many companies it’s strewn across the budgets of multiple cost and profit centers, accumulating without much transparency or control.

EY’s research breaks the true total cost of an AI agent into seven layers, and most tax functions are only budgeting for the first three¹:

1. Tokens and API calls — the per-use charge for every prompt, response, and retry; the one component that actually shows up on a vendor invoice.

2. Subscriptions and licenses — the fixed contracts signed before any agent runs: model provider agreements, orchestration platform fees, committed-use deals with cloud vendors.

3. Platform infrastructure — the compute, storage, and orchestration runtime that keep an agent operating, usually billed as a cloud expense rather than an AI expense.

4. Governance burden — the ongoing investment in guardrails, audit trails, and human-in-the-loop review needed to keep an agent’s output safe to rely on.

5. Organizational change — the cost of redesigning roles, retraining staff, and rebuilding workflows around what the agent now does, a cost that resets with every model upgrade.

6. Expected failure and recovery — the cost of correcting a wrong output, the rare but expensive incident, and the remediation budget most functions haven’t built yet.

7. Emerging AI-specific regulatory obligations — the reporting or compliance requirements regulators are assessing but haven’t finalized, a cost with no number attached yet.

The first three are visible and budgeted. The last four compound quietly until the financial consolidation presents a much different picture.

An example cited by EY: a customer service AI workflow that went from four cents per interaction in 2023 to a dollar twenty today, about thirty times higher, as a simple chatbot became an orchestrated agent with planning loops, sub-agents, and tool use.¹

Separately, analysts from Gartner project that more than 40% of agentic AI projects will be canceled by the end of 2027 because of escalating costs, unclear value, or inadequate risk controls.²

And a widely cited 2025 academic study by the MIT Media Lab found that 95% of generative AI pilots are producing no measurable return.³

None of these are stories about bad technology. They are stories about cost structures and value streams that don’t readily align with Industrial Age financial reporting. Those methods roll up the numbers by business unit, cost center, and geography, using push-down entries, cost allocations, and other long-standing techniques — leaving AI costs dangerously opaque.

The Materiality Moment

Call this the Materiality Moment: the point at which fragmented AI costs, individually tolerable and scattered across IT, Risk, HR, and Tax Technology budgets, consolidate into a number large enough to require explanation.

Senior Tax executives already understand this pattern. It’s the same logic behind the aggregation of passed adjustments in audit doctrine, where items immaterial in isolation become material in combination.

Subscribe now

Why Tax Is More Exposed Than Most Operating Functions

Tax is more exposed to this moment than most operating functions, for reasons specific to what it does. Its governance burden is disproportionate: for example, an AI agent that touches a provision number or a transfer pricing position carries regulatory and audit exposure a customer service chatbot does not.

Call the Meeting Before the CFO Calls You

The instinct so far has been to wait and watch the invoice, but the better move is to raise this now, before the cost aggregation arrives. Tax doesn’t need to solve it alone to do that.

Tax has lived through a version of this before. In the early days of ERP implementations, Tax was routinely left out of design discussions until late in the project, and systems went live configured around business requirements that excluded tax’s needs. The result was years of manual workarounds and rework — costs that would have been far cheaper to avoid by building tax’s requirements into the design from the start. AI is moving faster than ERP ever did.

The opportunity is to get in front of it this time, showing up as an active participant alongside Finance peers rather than a function correcting course after the fact.

The frameworks already emerging around AI risk management and AI-specific financial governance converge on a familiar handful of components:

• Named ownership

• A complete inventory of what’s deployed

• Risk tiering

• Hard spending controls

• Audit-grade monitoring

• Documented policy

• A review cadence

None of that needs to be invented. It needs a lead willing to assemble it before the bill forces the conversation.

What Tax Can Bring to the Table

Tax can start with its own AI agents, before asking anyone else to act.

1. Ownership — Name a Tax AI Cost Lead accountable for the full cost picture across every active initiative, the same way a control owner is accountable for a SOX position.

2. Register — Build a Tax AI Cost Register that catalogues every agent’s purpose, data sources, and full cost exposure.

3. Tiering — Tier each agent by materiality rather than generic risk labels.

4. Traceability — Hold every AI-touched position to audit-grade traceability.

5. Charter — Put it all in a one-page Tax AI Governance Charter.

That’s not internal hygiene for its own sake. It’s a working draft Tax can use as a straw-man to refine with Controllership, Risk, IT, and HR, instead of arriving with just a warning.

Getting a Seat at the Table

The realistic next step isn’t a Tax-built standard the rest of Finance adopts. It’s Tax convening the functions already absorbing pieces of this cost — Controllership, Risk, IT, and HR, and turning five fragmented budget conversations into one proposal.

The pitch to the CFO is: here’s an aggregated cost build-up nobody has been tracking, here’s a starting draft to improve AI spend governance, and here are the five of us asking for time to align before this becomes a number you’re explaining to the Board.

Nobody has built the enterprise version of this yet. That’s the opening. Whoever names the problem and shows up with a credible draft, shapes how the conversation goes from there.

The Materiality Moment is coming regardless of who’s ready for it. The only real choice is whether Tax calls the meeting or waits to be invited to someone else’s.

References

1. Diasio, Dan, Shawn Smith, Steve Wanner, and Whitt Butler. “Unlocking Agentic Value: A New Investment Discipline for the Agentic Era.” EY Total Cost of Agents series, EY, June 1, 2026. https://www.ey.com/en_us/insights/ai/agentic-ai-token-costs

2. Gartner. “Gartner Predicts Over 40% of Agentic AI Projects Will Be Canceled by End of 2027.” Gartner Newsroom, June 25, 2025. https://www.gartner.com/en/newsroom/press-releases/2025-06-25-gartner-predicts-over-40-percent-of-agentic-ai-projects-will-be-canceled-by-end-of-2027

3. Challapally, Aditya, Chris Pease, Ramesh Raskar, and Pradyumna Chari. “The GenAI Divide: State of AI in Business 2025.” Project NANDA, MIT Media Lab, July 2025. https://www.media.mit.edu/groups/nanda/overview/

Tax transformation programs rarely fail outright. They stall. They achieve partial adoption. They go live on schedule then quietly underdeliver for years. This article makes the financial case for the one investment most Tax functions have not made and why, properly understood, it pays for itself.

The Real Risk: Partial or low adoption leading to ROI dilution

Here’s the all too frequent scenario:

• The tax engine goes live on schedule.

• Training hits 95% completion.

• The steering committee signs off, then six months later,

• Managers have rebuilt parallel Excel models or dusted off the trusty old and disconnected versions.

• The workpapers look essentially the same as they did before implementation, and

• The promised efficiencies simply don’t show up.

This is Adoption Theater, the outcome of programs that contract for implementation and leave adoption unfunded, understaffed, and unmeasured. As I’ve argued previously, adoption gets a slide in the presentation deck when it really needs a line in the SOW.¹

Prosci’s Best Practices in Change Management research, drawing on more than 8,000 data points from global organizations documents this pattern precisely.

• Programs with excellent change management achieve an 88% success rate.

• Programs with poor change management achieve just 39%.

• Organizations with excellent change management are 7x more likely to meet or exceed their objectives. ²

What are the financial consequences of landing in the lower half of that distribution for a Fortune 500 company?

You likely end up with one or more of the following, some being more visible than others:

• Shelfware: A $500K annual license operating at 60% utilization wastes $200K per year every year the adoption gap persists

• Manual workaround cost: A team of ten senior Tax professionals each spending 25–30% of their time on processes a functioning system should have eliminated represents $750K to $1.25M in annual waste

• Re-implementation: When partial adoption is eventually recognized, a second implementation is scoped. A $3M program requiring a second attempt becomes a $4M–$6M program

• Slower ROI on AI initiatives: AI tools deployed into a function without change management infrastructure follow the same adoption curve as every prior technology investment. Only 29% of organizations currently see significant ROI from generative AI deployments³

• Audit and regulatory exposure: Manual workarounds create undocumented, unauditable decision trails. Pillar Two requires defensible data lineage. A function relying on spreadsheet workarounds alongside a platform not fully adopted has a documentation problem that doesn’t surface until an examiner asks for it

• CFO credibility: The next time the Head of Tax approaches the CFO for platform investment, the memory of the last program’s subpar delivery shapes the conversation. Credibility is a finite resource. A stalled transformation spends some of it, making the next ask harder to approve

So What’s Actually at Stake?

Most Tax Functions have never aggregated their cumulative transformation investment. When they do, the number is often larger than expected.

A typical Fortune 500 global Tax Function investing in modernization over a three-to-five year period commits:

• Tax provision platform (Corptax, ONESOURCE, Longview): $500K–$3M implementation plus $200K–$1M annually in licensing

• Indirect tax automation (Vertex, Avalara): $300K–$1.5M implementation plus $150K–$500K annually

• ERP tax workstream within a broader SAP or Oracle migration: $2M–$10M

• External advisory and implementation support: $500K–$5M per major program

• Internal staff time diverted from production work: often untracked, but measurable in the hundreds of thousands annually

Subscribe now

A reasonable estimate for a serious, multi-program transformation effort: $5M to $20M over five years. The questions worth asking are:

1. How much of that investment is being actively protected?

2. What is the cost of only partially adopting a transformation program?

It stands to reason that the cost can rise to the multi-million-dollar level. Of course, that cost is sunk. It was already spent on the transformation program yet few measure adoption efficiency.

So perhaps the cost of low adoption is more appropriately described as financial waste.

The Strategic Cost Almost No One Models

A Tax Function operating in chronic adoption deficit loses something beyond efficiency. It loses strategic capacity precisely when the business needs it most.

• M&A transactions: Rapid tax due diligence and deal modeling require clean, integrated data. A function without it takes three to four times longer to produce the analysis a competitive deal timeline demands. That is not a tax problem, it’s a deal risk

• New product and business model changes: Tax modeling for new revenue streams or new geographies requires data a partially transformed function cannot access quickly. Business units eventually stop asking

• Supply chain restructuring: Transfer pricing, permanent establishment risk, and indirect tax implications must be modeled simultaneously. Without integrated data infrastructure, that work becomes a multi-month manual exercise rather than a rapid advisory response

The answer isn’t a better change management workstream inside the next implementation. It’s a permanent executive who owns the problem before the next implementation begins.

Remind the CFO Organization What They’ve Already Learned

The CFO organization encountered identical failure patterns during shared services transformations and ERP consolidations throughout the prior decade. In many instances, the response was to build permanent change leadership capacity into those functions, not project resources, not borrowed consultants, but a dedicated executive accountable for the ongoing capability to absorb change.

Tax is at the same inflection point. The function that builds permanent change capacity now will absorb the next wave at a fraction of the cost of the function that doesn’t.

If you and your CFO accept the value an executive level Change Manager can deliver in terms of transformation adoption, you will see the role pays for itself. What it represents is insurance on the transformation investment already committed, against the most predictable and expensive outcome in the business: a program that goes live but never fully arrives.

The Case for Permanence

Framing this role as insurance against a discrete risk understates its value. Tax transformation doesn’t end but it’s not linear either. It moves in waves. Major implementations create peaks of intense activity. Regulatory changes, platform upgrades, and AI integration create troughs that are quieter but never fully inactive. A permanent change executive earns their value across both.

During peaks, they protect the investment. During troughs, they build something no external resource can: the relationships, institutional trust, and deep operational knowledge of how this specific function works that make the next wave of change adopt faster, more completely, and at lower cost. By the time transformation intensity rises again, and it will, that foundation is already in place.

With borrowed resources, you start building it from zero. Every time.

¹ Schutzman, Mark. “Adoption Theater.” The Tax Transformation and Technology, 2026.

² Prosci, Best Practices in Change Management, 12th Edition, 2023. prosci.com

³ Enterprise AI ROI Research, Terminal-X, 2026. terminal-x.ai

Share

What if the next major transformation at your organization actually stuck? Not because the technology was better configured, not because the implementation team was more experienced but because someone inside your Tax function owned the change from day one and was still there a year after go-live?

What if that person wasn’t borrowed from HR, wasn’t on a time-limited engagement, wasn’t an external consultant and hadn’t spent her career facilitating adoption workshops for industries she’d never worked in? What if she knew your team, your politics, your legacy processes, and your unspoken history with the last three implementations that didn’t quite land?

That person doesn’t exist yet in most Tax functions. She should.

What Isn’t Working

For the past decade, the dominant model for managing change in Tax transformation has been to import it.

When a major ERP upgrade or tax engine implementation gets underway, the program scope typically includes an Organizational Change Management (OCM) workstream staffed by consultants skilled in change methodology but new to Tax. They build stakeholder maps, run readiness assessments, and facilitate adoption workshops. The work is professional. The adoption is often thin.

This isn’t a critique of the consultants, I’ve been one. It’s a candid observation about why adoption often misses expectations: you can’t outsource trust.

Change earns adoption by building credibility with the people being asked to change. In a specialized professional function like Tax, where the team carries deep institutional knowledge, a long memory for programs that overpromised, and legitimate skepticism of outsiders explaining how they should work. Credibility takes years to build. A time-boxed consulting engagement doesn’t have years.

The change management workstream can produce compliance but it rarely produces the required degree of conviction to make change stick.

Subscribe now

The precedent is already in your org chart

Twenty years ago, the Tax Technology Director didn’t exist. If your global Tax function needed technology support, it went to IT, or it relied on whoever in the group happened to be comfortable with Excel. Technology was treated as a project resource, something you called on when needed, then sent back.

That model broke down as technology became a permanent condition of the Tax function’s work. ERP deployments, tax engine implementations, data architecture decisions — these weren’t episodic anymore. They were continuous. And so the function did something sensible: it built standing internal capability.

The Tax Technology Director role emerged, formalized, and then became standard. Today, a global Tax function without dedicated internal technology leadership is visibly behind the curve.

The arc followed a recognizable pattern:

• Non-existent, to

• Informal hybrid role, to

• Recognized specialty, to

• Leadership team requirement

It took about fifteen years. The forcing condition wasn’t any single large implementation, it was the acceptance that technology was no longer a project but a permanent operating condition.

Transformation is the same condition

To be clear, I’m proposing a permanent senior level role on the Tax function organization chart for a Tax Change Director on par with the VP roles for International, SALT, Controversy, Planning and Tax Accounting.

If transformation is not a project with a finish line but a permanent state of the Tax function at varying levels of intensity, and I believe it is, then the logical conclusion is uncomfortable: the function needs standing change leadership the same way it needs standing technology leadership.

The Tax Change Director, or whatever title fits your organization, is not an OCM consultant with a permanent badge. The distinction matters. This person has domain fluency in tax, continuity across multiple transformation cycles, and relationships earned over years.

Her accountability doesn’t expire when the contract does. She maintains adoption of the last implementation while preparing the function for the next operating model shift. She is present for the quiet failures that happen six months after go-live, when the consultants have left and the old habits have returned.

The objection worth taking seriously

The honest pushback is budget. Tax functions are cost centers under pressure, and proposing a new leadership role is a hard conversation. But the framing matters: this is not adding overhead.

It is internalizing a cost the function already bears in the form of failed adoption, expensive re-work, and re-implementations of things it already paid to implement once. The external OCM workstream has a line item too. It just doesn’t show up as headcount.

How to start before the role exists

Most organizations aren’t ready to create this role today. But they can begin building the capability by:

• Identifying the senior tax professional who has the most credibility with the team

• The one having the sharpest instinct for the people-side of change

• Provide visibility and voice in the next transformation program — not as a stakeholder, as a co-leader of the change effort

• Watch what she does. Document what she knows. You may be writing the job description from the inside out.

The Tax Technology Director didn’t begin with a complete job description. She was recognized before being named, then expanded the role as conditions required.

My thought - history is repeating itself, this time in the change management specialization.

AI arrived but no one asked what happens to your spreadsheets.

Let’s start with something everyone in Tax knows but almost nobody speaks aloud: the most widely deployed tax accounting platform in the world isn’t from the leading tax software vendors like Corptax, or ONESOURCE, who both offer world-class applications.

It’s Microsoft Excel.

It has been for decades. Global Tax functions at Fortune 500 companies run provisions, workpapers, transfer pricing analyses, state apportionment and intercompany calculations on linked spreadsheets. Even those using one of the leading tax provision applications rely on a complex spider web of supporting spreadsheets to get the job done. That’s simply an accurate description of how the world’s tax work gets done.

The industry conversation has long treated this as a problem to be solved, as the transitional state between where Tax functions are and where they should be. The vendors are right that enterprise platforms offer real capabilities Excel cannot replicate.

• AI is making Excel meaningfully smarter making it a potent data management platform in its own right

• The broader Microsoft stack most organizations already license remains largely untapped by the Tax Function

This combination represents a legitimate strategy for Tax Functions that cannot, or choose not to, make a major, AI-centric platform investment right now.

The Excel Doctrine is a set of deliberate decisions about how to govern, extend, and intelligently evolve what you already have.

What you already have is Ungoverned Excel characterized by:

• Linked workbooks prone to breaking links

• Key-person dependency – critical spreadsheet knowledge not institutionalized

• Hardcoded cells without anyone able to explanation why

• Legacy worksheets that are no longer relevant yet add unnecessary clutter and complexity

• Version uncertainty (the final version vs the final final version)

• No audit trail

This is a real operational and compliance risk. Governing it well is a legitimate starting point, whether or not AI investments that will touch the Tax Function at your company are happening now or not.

Subscribe now

What’s Changed

Microsoft has made three additions to Excel that most Tax functions have not yet absorbed.

1. Excel Copilot writes and audits formulas, identifies inconsistencies across data ranges, and surfaces anomalies that previously required manual review. But it’s underused.

2. Python integration brings statistical analysis and data transformation capabilities that previously required a specialist working outside the workbook entirely. But Python isn’t used at scale in most tax departments.

3. Microsoft 365 Copilot, available across the suite, can accelerate the documentation work that consumes more time than most admit.

None of these eliminate the underlying discipline required to manage a spreadsheet-based environment well. They reduce the skill premium required to do it safely and raise the ceiling on what a well-managed Excel environment can produce.

The Stack You Already Own

Most companies in the Fortune 500 category operate under Microsoft 365 licenses that include Power Automate, Power BI, and SharePoint, tools the IT organization manages, and the Tax function usually does not optimize. This is the immediate opportunity.

Power Automate deserves specific attention. Consider the quarterly tax provision process. In my experience working with global Tax functions, the data collection and reconciliation phase alone consumes as much as 50% - 70% of the total effort: emails go out to subsidiaries and business units requesting financial data, spreadsheets come back in different formats, someone reconciles the versions, another person consolidates them into the master workbook, and the process repeats when numbers change. Every manual handoff introduces error risk. Every email thread is an undocumented decision unlikely to ever migrate from human to institution.

Power Automate replaces that workflow. A structured data collection form routes to the right contributors on a defined schedule. Responses populate a centralized data source automatically. Exceptions trigger alerts. The consolidation that previously took days of manual effort runs in minutes. The audit trail exists by design rather than by reconstruction.

This is a workflow built inside tools the organization already licenses, often deployable in weeks by someone with moderate technical fluency — no steering committee required.

That’s CFO nirvana – incremental value from a sunk cost enhancing ROI.

Power BI adds a governance and visibility layer — dashboards showing provision status, data completeness, and variance analysis in real time rather than at the point when someone finishes the manual consolidation. SharePoint adds version control and document management to workpapers that currently exist in shared drives with ambiguous naming conventions and uncertain ownership.

Where This Fits in the Broader Market

The Big Four are now offering a more comprehensive path. A data readiness engagement followed by a purpose-built platform implementation can address foundational problems at scale with depth that internal resources rarely have.

For some organizations that is exactly the right investment. It is also expensive and most effective for functions that have already addressed data acquisition, data governance and the process discipline this article describes.

The Microsoft path builds that readiness without the price tag, on a timeline the function controls. The organization that has governed its Excel environment, automated its manual workflows in Power Automate, and built visibility through Power BI is a meaningfully better candidate for a major platform investment and a far more informed buyer when that conversation begins.

What the Excel Doctrine Requires

It requires deciding, explicitly, that the spreadsheet is a core operational tool to be managed with the same rigor as any other enterprise system, not a transitional embarrassment to be explained away to auditors and vendors.

It requires someone with ongoing accountability for the health of the Excel environment, not a project that closes and moves on. And it requires a willingness to invest modest time and internal resources into tools the organization is already paying for.

The Tax functions who are watching the AI-centric transformation announcements from the sidelines, waiting for budget, consensus, or the right moment, have more options than the market conversation suggests.

The infrastructure for meaningful progress is already in the license. The only thing missing is the decision to use it.

It was twenty years ago today, around the time Sgt. Pepper taught the band to play, that I planted some seeds I didn’t expect to think about again this morning.

I’ve been reading Rewired, McKinsey’s guide to competing in the age of digital and AI, and somewhere in the early chapters I put the book down and began to consider.

I’ve heard all of this before.

Not as criticism, it’s an excellent book, and has my highest recommendation. But as someone who spent the better part of the last twenty years inside the transformation of the Tax function, the concepts felt less like revelation and more like reunion. Operating models, data strategy, the tax function as a business within a business, process before technology and change management as the variable that determines whether any of it actually works.

I was making these arguments in 2004 and 2005, when I helped launch what we called the Tax Function Effectiveness practice at PwC. And the firm was not ready.

History Matters

The questions I fielded were earnest but revealing. Senior partners, all of whom were accomplished, intelligent, and deeply experienced asked me why I believed I could run a corporate Tax function better than the people already doing it. My answer was a question in return: are you really prepared to indict the entire management consulting industry? The conversation usually moved on quickly. Not because anyone was persuaded but because the discomfort was easier to exit than to resolve.

As the practice matured, I proposed a return to a software strategy at a moment when the memory of direct software ownership was still raw. The instinct to move toward platform thinking felt to me like the obvious next step. To others, it felt like reopening a wound.

I spoke about the platform of the future to leaders who had built their careers on having answers, not on placing bets under uncertainty. Many were tax attorneys by training, technically brilliant, professionally authoritative, and genuinely unequipped, through no fault of their own, for the particular challenge of allocating capital toward something whose value wouldn’t be visible for years.

The back-handed compliment I heard most often, sometimes spoken as if I weren’t in the room: ahead of his time.

It was meant kindly but landed as a polite way of saying: You’ve lost your mind.

What I learned – Twenty Years Later

Here’s what that experience taught me about trusting your instincts and I use that phrase deliberately, knowing how easily it slides into cliché.

Trusting your instincts is a leadership quality, a habit and it’s hardest to sustain precisely when the environment is giving you reasons to stop.

The organizational immune system is sophisticated and prone to rejecting unfamiliar ideas. It produces social friction, polite dismissal, and a particular type of exhaustion that emanates from making the same argument to different audiences over months and years without visible progress. That’s hard work.

The people in those conversations weren’t obstructionists. They were cautious stewards of institutions that move slowly for reasons that are sometimes driven by hierarchy, sometimes cultural, and sometimes simply human. Twenty years of hindsight makes the path look obvious. I can say with certainty and clarity, it wasn’t obvious then.

Subscribe now

What perseverance looks like in that environment is less heroic than the word implies. It’s about remaining engaged after the idea has been dismissed. Refining the argument rather than abandoning it. Finding the one person willing to explore rather than the many who prefer to move on or the always valued - finding validation through commercial success at a small scale.

The Real Return on my Investment

Years after leaving PwC, I found myself in a client meeting that included personnel from my former firm. A consultant pulled me aside, someone who had come to PwC from another Big4 and was now working in the next iteration of the very practice I had launched. He looked at me and said, simply: thank you for what you have done for my career.

I didn’t see that coming. And I’ve thought about it many times since.

Because that moment clarified something I hadn’t fully understood while I was in it. The work produced something quiet and durable, a foundation that someone else built on, a direction that outlasted the resistance it first encountered, a career that wouldn’t have taken the shape it did without groundwork that had been largely forgotten.

You don’t always get to see what your ideas become. Occasionally, if you’re lucky, someone tells you.

The transformation leaders reading this today are operating in a genuinely different environment. The case for technology investment in Tax is broadly accepted. AI has arrived with enough force that even the most skeptical voices are paying attention.

But there is always the next frontier. There is always an idea that is correct before it’s accepted, uncomfortable in real time, obvious only in retrospect. The specific details of that idea, the one today’s leaders are sitting with in conference rooms that aren’t ready, I won’t pretend to know. The signals are ambiguous and that’s precisely the point.

What I do know is that patterns repeat and the people who move industries forward are rarely the ones who wait for consensus.

So, if you are the person who sees something others aren’t prepared to act on, trust your instincts. Refine the argument. Stay patient with the system without becoming a prisoner to it. Accept the back-handed compliment gracefully and keep going anyway.

The seeds you plant may grow in ways you can’t predict, tended by people you haven’t met yet, producing results you may only learn about years later in an unexpected hallway conversation.

That’s not a consolation prize but a real-life example, one plucked directly from my legacy book, demonstrating how real change moves.

In May 2026, KPMG and Anthropic announced KPMG Digital Gateway Powered by Claude, with tax clients named as the initial focus. By then KPMG was the fourth Big Four firm to commit at scale. Deloitte deployed Claude to its 470,000 employees in October 2025. PwC expanded its Anthropic alliance in May 2026 and separately partnered with OpenAI to build agentic systems around the office of the CFO, including tax and the close. EY has 150 AI agents in production for its 80,000 tax professionals and has publicly committed to scale to 100,000 agents by 2028.

Collectively, the four firms have committed more than $9 billion in AI investment. Entry-level Big Four hiring is down 29% over two years. The traditional consulting pyramid is being explicitly restructured. This is an industry restructuring built on the premise that AI agents will perform work tax professionals have done for decades. The shift is on.

For tax practitioners, the question is what this actually displaces and what it makes more valuable.

The Bifurcation

The answer is a clean bifurcation. Three categories of tax work are likely to migrate substantially toward platform-delivered, agentic AI service over the next 24 to 36 months. Three other categories will remain human-led, and may grow more valuable in relative terms as the first three shrink.

The boundary between them sits at a specific line: between work that can be done with clean data and rule-based logic, and work that requires judgment in genuinely ambiguous fact patterns.

The first side of the bifurcation will feel like loss for practitioners whose careers were built there. The second will feel like opportunity for those who can position themselves on it. Both are happening simultaneously, and both deserve careful attention.

Figure 1. The Tax Work Spectrum, with the bifurcation between work that migrates to platform delivery and work that remains human-led.

What Migrates to Platform Delivery

The clearest cases for platform migration share three characteristics: the work is rule-bound, the underlying data is or can be structured, and the deliverable is discrete and measurable.

Recurring compliance in well-defined jurisdictions is the first category. When a multinational files the same form quarterly with substantively the same facts, the work is exactly what agentic AI is designed to do. Extract the relevant data from source systems. Apply the rules. Generate the return. Flag anomalies for human review.

The work doesn’t go away but the labor that used to be billed by the hour against it does. KPMG’s tax-regulation-adjustment agent is an early production example; I expect every Big Four firm is building similar capability.

A second example is the computation of most book-tax Schedule M-3 adjustments can be automated, particularly at companies with the data infrastructure to support automation. Human review, of course, will be required.

But the data at large multinationals is increasingly structured, the calculation logic is well-defined, and the output is a discrete deliverable that flows into the Federal tax returns. The pieces are aligned for agentic delivery. The core tax compliance is not going to be calculated by hand by a tax senior in five years at companies with the data infrastructure to support automation.

Indirect tax determinations in clean-data environments is the third. Sales/use tax and VAT engines have been getting more capable for years; what changes with agentic AI is the ability to handle exceptions, edge cases, and ambiguous transactions without escalating to a human. The first wave of indirect tax automation handled the easy 80%. The current wave is going after the next 15%. The remaining 5% — where judgment is genuinely hard — stays human.

Subscribe now

What Remains Human-Led

The other side of the bifurcation is harder to automate and may grow more valuable as a result.

ERP-tax architecture and data governance is the first category. AI applied to bad data produces bad answers at scale. The first wave of agentic tax automation will fail loudest at companies whose tax data infrastructure isn’t ready for it, which is most companies.

The work of getting the underlying data architecture right, of designing tax requirements into finance systems from the start rather than reconstructing tax data after the fact, is the upstream remediation everything downstream depends on. Far from being displaced by agentic AI, this work becomes the precondition for AI to function at all. It will grow.

Tax integration and carve-out execution in M&A and private equity situations is the second example. Post-acquisition tax integration combines non-recurring complexity, severe time pressure, judgment under ambiguity, and data quality that is by definition messy.

Two different tax functions, two ERP stacks, two methodology histories, none designed to integrate cleanly. The platforms that will handle steady-state tax compliance of the combined entity cannot handle the transitional period. The senior tax practitioner who can navigate a post-deal integration with confidence will be in demand for as long as M&A activity exists.

Complex restructuring and transfer pricing controversy is the third. These are situations where the cost of error is too high to delegate to agents. A restructuring position that costs the wrong thing, a transfer pricing position that loses on appeal, a controversy outcome that creates years of follow-on exposure. Senior judgment is the substantive deliverable. The platform serves the senior advisor; it does not replace them.

Share

The Boundary and Why it Sits Where it Does

The sentence that captures the bifurcation: AI applied to bad data, or to genuinely ambiguous fact patterns, produces bad answers and the cost of a bad answer in tax can be enormous. That determines where the boundary sits.

For the work that migrates, the precondition is structured data and rule-based logic. The Big Four are building the platforms; clients with the most mature data infrastructure will be first to adopt. The transition may take 24 to 36 months to play out in substance across the Fortune 500 and longer to reach the broader market.

The interesting middle case is upstream remediation, getting clients’ tax data architecture ready for agentic AI. That work is structured enough to attract significant investment, complex enough to require senior judgment, and growing precisely because clients increasingly want to deploy AI but cannot do so without first fixing their data foundation.

Practitioners who can deliver this work sit at the most attractive point in the bifurcation: building the precondition for the AI transition and benefiting from the demand the transition itself creates.

Implications for Tax Practitioners

For individual practitioners thinking about career direction, the implication is clarity rather than alarm. The work that goes away has been visible as labor-intensive recurring work for years. The work that grows in value has been visible as senior judgment work for years.

What changes with the AI alliances is the velocity. Practitioners investing in their judgment capabilities, in their ability to handle ambiguity, in substantive depth in tax accounting, M&A, state nexus determinations, transfer pricing or international tax consulting, are positioned well. Practitioners whose careers have been built on the recurring compliance side face a transition they can either resist or get ahead of.

For tax practice leaders thinking about where to invest, the implication is also clarity. The categories migrating to platform delivery are the ones where the Big Four are committing nine-figure investments.

Competing head-on requires comparable investment most firms cannot make. Competing on the categories that remain human-led requires investment in senior talent, codified methodology, and the upstream data work that makes everything else possible. Firms that align their investment to the side of the bifurcation they can credibly serve will do well. Firms that try to be everything spread themselves thin.

For the profession as a whole, this is the kind of change diagnostic imaging brought to medicine and electronic discovery brought to law. Some work disappears. Other work gets harder, more specialized, more valuable. The practitioners who adapted to those earlier shifts did well. The ones who didn’t, didn’t.

What to Do Now

The shift is not yet uniformly visible. Most tax functions are still operating in the model of the past five years. The transition will be felt unevenly, first at the largest multinationals with the cleanest data, then at the F1000, then more broadly.

But the direction is clear. Practitioners with three to seven years of experience should be deliberately deepening their substantive expertise in the work that remains human-led. Practice leaders should be auditing their current revenue mix against the bifurcation: how much of what we do is on the platform-migration side, and how much is on the human-led side? The answer shapes where the next round of hiring, methodology investment, and client positioning should go.

The Big Four AI alliances are the moment that forces every tax practitioner and every firm to take a position on which side of the bifurcation they intend to compete on. That decision is becoming harder to defer.

Mark D. Schutzman is a former Big4 Tax Partner who has spent more than two decades in tax transformation work and practice leadership. He co-authored Tax Function Effectiveness (CCH, 2008).

Walk through any large finance organization and you can identify, within minutes, which function owns which data.

Controllership owns the general ledger. The trial balance reconciles to it. The auditor accepts it as the source of record for financial accounting. The vocabulary is settled, chart of accounts, sub-ledger reconciliation, period close, intercompany eliminations, and the politics are institutionalized. Nobody inside finance argues with controllership about who owns financial reporting data. The position is so secure that the question is no longer asked.

Treasury owns the cash, the bank connectivity, and the FX rate environment. Their data conversation is real-time, multi-currency, multi-entity, and they solved it years ahead of the rest of finance because they had to. A cash position cannot wait for a monthly close. They built API-based integration with banks while the rest of the finance function was still pulling extracts. They acquired a system, the TMS, that encoded their domain claim, and the rest of the organization speaks their dialect back to them.

FP&A owns the plan and the forecast. The vocabulary is forward-looking: variance, scenario, driver, segment view. The platform, the EPM, sits in every enterprise IT budget conversation. The function produces what the executive team consumes weekly. The position is recognized because the output is recognized.

And Tax owns…?

Most tax leaders, asked the question, will pause. Then they will describe outputs like the return, the provision, the rate, the workpapers, rather than data. Outputs are not a data domain. They are work products that consume someone else’s data and produce something the function is then held accountable for. The gap between describing outputs and naming a domain is the entire diagnosis.

Tax is the only major finance silo that has not claimed a data domain, developed a recognized vocabulary for it, or institutionalized a system category around it. Every “clean your data” conversation in tax begins here, and every one of them stalls for the same reason: there is no defended territory to clean, centralize, or govern against.

Subscribe now

I’ve watched this play out

I have sat in finance leadership meetings where treasury, controllership, and FP&A each walked through their data ambitions in turn. Each presented a system, a roadmap, and a defended position. When the conversation reached Tax, the shape of the discussion changed to projects, to programs, to point solutions. Never to a system. Never to a domain.

The change was visible to me because I was in the room. It was invisible to the participants because each function was, in good faith, describing what it had. The other three had systems. Tax had work. Spreadsheet work.

That asymmetry is the structural condition the Tax function has accepted, in most organizations, for thirty years. And in 2026 it is the binding constraint on every AI initiative the function will try to run.

The missing system category

The other functions did not earn their data positions by writing better requirements documents. They earned them by acquiring tooling that encoded their domain claim. The GL lives inside the ERP. Treasury lives inside the TMS. Planning lives inside the EPM. Each system is, in addition to whatever else it is, a positioning artifact, the existence of the category forces the conversation about who owns it.

Tax has tax engines and provision tools. These are calculation engines, not systems of record. They take data as input, perform a deterministic transformation, and produce a result. The architectural claim they make is “we compute outputs,” not “we hold the data the function depends on.” That is a meaningfully different claim, and it produces a meaningfully different conversation in every IT roadmap, every data-platform decision, every enterprise-architecture review.

Subscribe now

What’s missing, and what the function now needs, is a recognized system category of its own. Call it the Tax System of Record.

A Tax System of Record is not a tax engine, not a provision platform, and not a data warehouse. It’s the system that holds — with provenance, reproducibility, and audit defensibility — the data the Tax function depends on, structured around the dimensions that make tax data different from every other kind of finance data.

There are three of those dimensions and each one is primary.

Why JET is the right structure

Each of the three dimensions is something no other finance function has to carry or carries in the same way.

Jurisdiction is the dimension only Tax has to handle as a primary structure. Tax obligations arise from jurisdictions; everything else is conditioned by them. A single transaction can have a U.S. federal characterization, a state characterization, and a foreign characterization that all differ and all coexist.

Pillar Two introduces a fourth, the GloBE characterization, which is neither the federal nor the state nor the foreign view but a regime-specific structure layered over all three. A Tax System of Record holds these views simultaneously without flattening any of them. Most current architectures cannot do this because they were not designed to.

Entity is the legal entity hierarchy the tax function uses, not as the consolidation engine uses it. Legal entities, branches, disregarded entities, partnerships, tax groups, fiscal unities, controlled foreign corporations, hybrid arrangements.

The accounting consolidation tree and the tax entity hierarchy share a name and almost nothing else. I have lost count of how many tax data programs began by inheriting the consolidation tree from finance and assuming the work was done.

Time is the dimension most often underestimated. Tax data operates across at least four axes simultaneously: the compliance calendar (when filings are due), the transaction date (when the economic event occurred), the statute-of-limitations look-back (can be a decade or more, with selective reopening through amendments), and the forward period (provisions, forecasts, uncertain tax positions). No other finance function carries this complexity. Treasury is real-time. Controllership is period-close. FP&A is forward-looking. Tax is all of them, indefinitely.

Drop any one of the three, or treat any one of them as metadata applied after the fact, and the system is no longer a Tax System of Record. It is a data warehouse with tax labels.

What it takes to build one

Once the function has named the category, the work has principles, in the order a practitioner actually encounters them.

1. Source-of-truth lineage discipline

Every input that lands in a filed return must be traceable, in one or two steps, to a system the business already trusts: the GL, a sub-ledger, the master data system, or a relevant statutory record. Tax data built on extracts of extracts does not survive the question “where did this number come from?”

In a project I worked on, the data lineage trace ran from a spreadsheet, to a second spreadsheet, back to a tax data extract, back to a finance extract, back to a planning system that had been retired the year before. The numbers reconciled. The lineage did not exist. This is the sort of audit risk that seems all too prevalent and buried in the murky depths of Tax data.

2. JET as primary structure, not metadata

Most tax data architectures treat jurisdiction, entity, and time as tags applied to data that was structured for someone else’s purpose. I recommend them to be designed in a first-class dimension of their own. Retrofitting fails, not occasionally, but predictably.

The reason is that the data is shaped, at extraction, around the consuming process. A flat fact table with jurisdiction codes appended is not a jurisdiction-primary structure. A consolidation tree with tax entity flags is not an entity-primary structure. The difference shows up the first time the data needs to answer a question the original consumer did not anticipate.

3. Tax-aware standardization

Treasury did not accept generic enterprise data standardization for FX rates; they built their own because the requirements were unique. Tax has to do the same for withholding flags, treaty residency, transfer-pricing dimensions, exemption codes, state apportionment, place-of-supply determinations and indirect-tax taxonomies that differ by jurisdiction.

Generic data hygiene produces clean data that is still wrong for tax. Standardization that is not designed by people who know both the data domain and what tax does with it produces a clean lake the function cannot drink from.

4. Storage with reproducibility as a design input

The audit-defensibility requirement is uniquely strong for tax because the external stakeholders, the tax authorities, statutory auditors, courts and controversy counsel, apply uniquely strong evidentiary standards.

The system has to be able to reconstruct any output at any prior point in time, against the data and the rules that were in force then. “We kept the data” is not the same as “we can defend it.” Most current implementations satisfy the first standard. Very few satisfy the second. That gap is its own form of debt, accumulating quietly under every AI overlay being deployed.

5. Governance as a named role, not a steering committee

Controllership, treasury, and FP&A institutionalized their data positions by giving someone a job, not by scheduling a meeting. The Tax System of Record needs a named owner, defined decision rights, and an accountability path that runs into IT, finance, and tax simultaneously. Without that, the four principles above erode within two reporting cycles of go-live. The discipline that builds the system is not the discipline that maintains it; the second one is harder, and it is the discipline most programs never staff.

Sequencing matters

These principles are not parallel work. They are sequential in this order.

1. Master data - the entity hierarchy, the jurisdictional cube, the calendar structure.

2. Transactional lineage - the discipline that ties every downstream output back to a source the business already trusts.

3. Standardization - the semantic layer that makes the data tax-usable rather than just clean.

4. Storage architecture - the design decisions that determine whether the system can defend its outputs years later.

5. Governance - the institutional commitment that determines whether any of the prior four survive the next reporting cycle.

Most programs invert this sequence because the visible AI use-case lives at the bottom of it. The pull is real. A reasonable executive looks at a roadmap that starts with master data and asks, correctly, when will value appear? The honest answer is that value appears when the function can do something it could not do before and that capability is built from the foundation up, not from the use case down. The successful programs resist the pull. The unsuccessful ones spend three years building dashboards on top of data that cannot reconcile to itself.

Why this conversation is happening now

Pillar Two is forcing the global MNC Tax function to adopt a platform it should have built ten years ago. The data input requirements published by global tax authorities for Pillar Two compliance assume a level of jurisdictional granularity, entity coverage, and traceability that most Tax functions cannot produce from any existing system. The published catalog of Pillar Two data inputs alone runs to dozens of distinct line items, each tied to specific legal entities, jurisdictions, and reporting periods, and in many cases to a defensible methodology trail.

The Tax function is being asked to produce data it does not have, structured along dimensions it has not built, on a regulatory clock it does not control.

E-invoicing mandates rolling out across Europe, Latin America, and parts of Asia are imposing similar requirements on indirect tax, in real time, with no implementation grace. The shift, in many jurisdictions, is from periodic filing to continuous reporting which means the data has to be right, complete, and defensible at the moment of transaction, not at the end of a month-end cleanup cycle.

And the AI conversation, the one that is dominating every consulting deck and vendor demo, depends entirely on having data that an AI system can be trusted to operate on. The major tax operating model surveys from the last two years all report the same pattern — leaders see AI as transformational; almost none of them describe their data foundations as ready. That gap is not new.

This is the AI conversation, not a precondition to it

The current conversation about AI in Tax is loud about outputs and quiet about foundations. “Clean your data” has become the universal first line of every AI roadmap deck, delivered with enough confidence to pass for a plan. It’s not a plan. It’s more accurately described as a placeholder for work the function has never named and powered through a labyrinth of spreadsheets with human heroics to get it right.

Until the function has a Tax System of Record: a recognized category, owned by Tax, structured around the JET dimensions, governed by a named owner, every AI initiative layered on top of it is layering on top of nothing. The investment will produce a demo. It will not produce a defensible deliverable. And the same conversation about “data readiness” will repeat in 2028, in 2030, and again in 2032, exactly as it has repeated since the first ERP went in.

A true Tax System of Record is an enterprise technology category that does not yet exist.

The tax engine goes live on schedule. The dashboards look right. Training is 95% complete, the comms plan runs its course, the steering committee signs off, and someone with authority declares the program complete.

Six months later, the managers have rebuilt parallel Excel models because the tax engine couldn’t accommodate a few complex calculations. It handles routine compliance and standard provision items. But the work that drives risk and value, the complex jurisdictional logic, the M&A-driven anomalies, the workpapers that get pulled into audit, still happens in spreadsheets the tax engine never touched. The workpapers look essentially the same as they did three years ago. The promised efficiencies show up in slides, not in calendars.

The customary explanations get repeated by everyone involved: change management was under-resourced, training didn’t stick, leadership stopped paying attention. But the pattern repeats across organizations, vendors, and engagement models. That doesn’t look like a string of execution failures. It looks like an outcome the system is producing. But why?

The Schneier frame

In 2003, the security technologist Bruce Schneier coined a phrase that has aged well: security theater. He used it to describe visible measures like shoe removal at airports, color-coded threat levels, ID checks that verify nothing, each signals vigilance without measurably reducing risk. His point wasn’t that the people running these systems were acting in bad faith. It was that institutions tend to produce the appearance of doing something when they cannot, or will not, commit to doing it in practice. Theater isn’t a moral failing; it’s a systemic outcome of misaligned incentives.

Tax transformation has its own version of this, and it is the most consequential and least-discussed example. Call it adoption theater.

Every consulting firm has an adoption playbook. They will show you the deck. Every program has a change management workstream, a sustainment plan, a behavioral adoption methodology. The discussion is loud. But what gets contracted, scoped, budgeted, and staffed is implementation. The adoption content lives in the pitch slides and the methodology deck; the adoption commitment doesn’t survive into the SOW.

The gap between what companies say about adoption and what companies actually pay to deliver it is where the failure pattern lives. Adoption gets a slide. It doesn’t get a line in the SOW.

Subscribe now

How adoption gets engineered out

I have sold, led and delivered enough of these programs to recognize how this happens. The components are not hidden. They are visible in every engagement, every SOW negotiation, every project plan. They just don’t get named. They divide into two sets: the agreed upon workplan that defines the engagement before kickoff, and the forces that close it down afterward. Let’s look at this more closely.

The direction is set before kickoff. Start with the SOW. Implementation deliverables such as; configure, test, train, go-live and stabilize are sized in detail. Adoption typically appears as “hypercare,” a thin 60-90 day line item that funds residual issue resolution, not behavioral change.

No one drafts the SOW that funds a 24-month adoption arc, and no one signs it if they do. Transformation leaders are looking for ways to reduce the cost of their programs – the adoption effort is often the first casualty.

The program budget compounds this. Implementation costs are knowable, capitalizable, and defensible to a finance committee — fixed-price or T&M against a defined scope. Adoption costs are not knowable in the same way, which puts them in the soft middle of the proposal. When negotiation pressure arrives, the unknowable line moves first. Adoption gets compressed before implementation does, every time.

Share

The economics of closure

Once a program goes live, every force in the system points toward closure. Consultants roll to the next engagement, billable elsewhere by the next utilization cycle. Sponsors want the milestone reported, finance wants the capex closed, the exhausted tax team wants relief from two years of dual-track work. And the success metric was always go-live, not adoption. No one defined adoption metrics past day 90, so no governance body reviews them six months later.

Adoption theater isn’t the failure of any one party. It’s what the system produces when implementation is contracted, funded, and staffed and adoption is not.

Why managing it better doesn’t fix it

The instinct, once people see this, is to fix it by managing adoption better; add a change manager, build a stronger comms plan, run more training, stand up an adoption workstream inside the implementation program.

It doesn’t work. Adoption isn’t a workstream you can run inside an implementation; it’s a byproduct of commitments made before kickoff, in the contract, the budget, the staffing plan, and the metrics. Treating it as a deliverable replicates the implementation mindset that produced the gap in the first place.

You cannot project-manage an outcome the engagement design has already engineered against. It is bailing water out of a boat with a hole below the waterline, at some point you have to stop bailing and address the hole.

The better adoption playbooks don’t fail because they’re poorly written. They fail because they’re applied inside engagements whose budget, and resourcing make adoption unreachable no matter how good the playbook is.

The question before kickoff

The implication for the executive sponsor is uncomfortable but actionable. Before the implementation begins, before the SOW is countersigned and the project plan locked, one question has to be answered honestly:

If the honest answer is ‘the consultants’, as part of the SOW, verify the SOW actually says that. Verify that it’s funded past 90 days. Verify there are named individuals with adoption-specific deliverables and accountability metrics, not just a hypercare clause. If those things aren’t in the document, the answer is functionally ‘no one’, regardless of what the methodology slides claim.

If the honest answer is ‘we’ll own adoption internally’, show me the team. Show me the budget reserved for the 18-24 months past go-live. Show me the operating cadence, who runs the adoption reviews, with what metrics, reporting to whom, with what authority to redirect resources when adoption stalls. If those things don’t exist before kickoff, they will not materialize after go-live when the implementation team has rolled off and executive attention has moved on.

There is a third option worth naming, which is restructuring the engagement model itself. Rather than expanding scope, move from a project model to a multi-year partnership in which the consultant supports an internal adoption owner. Same total spend, potentially; very different outcome. A handful of firms are starting to offer this. Almost no clients ask for it.

A subtlety that gets glossed over and shouldn’t be: the internal-ownership path is harder than most organizations admit. The reason they hired consultants in the first place is often that they didn’t have the internal capacity. So “we’ll own adoption internally” is itself a significant commitment that requires building or borrowing a function that didn’t previously exist. That commitment has to be made before the implementation, not discovered six months after go-live when the engine is being bypassed.

The reason adoption fails in tax transformation isn’t motivational, methodological, or technological. It’s structural. The engagements take the form governed by workplans, SOW conventions, budget cycles, utilization economics, and the mutual interest in clean closure, to deliver implementation and to leave adoption unfunded, understaffed, and unmeasured. Then, when adoption predictably doesn’t land, we blame the change management and start the cycle again.

The fix has to come before the engagement, not during it. The honest question is the test: who owns adoption, with what authority, with what budget, for how long. Anything less than a clear answer, funded and resourced before kickoff, is adoption theater and the next program will look like the last one.

In nearly every transformation business case I have reviewed, the math is accurate but the assumptions are not.

The cost of change is modeled in extraordinary detail — license fees, implementation hours, consulting partners, internal time, risk reserves, contingencies, contingencies on contingencies.

The cost of not changing is included, but only in one narrow form: the current-state cost model is used as a static baseline against which projected efficiency gains are translated into dollars saved. That is the math behind the ROI number on the cover slide.

The assumption beneath that math — that today’s run-rate continues, flat, forever — fails a simple empirical test. In the organizations I have worked with, the cost of running the Tax function does not grow in step with the business.

It grows faster. Volume rises. Regulation accumulates. Manual workarounds harden into permanent fixtures. Key people leave, taking institutional knowledge with them. The gap between what the function is structured to do and what the business now needs widens, quietly, every year.

Tax function cost rises faster than revenue, faster than transaction volume, faster than headcount in the rest of the organization. In many cases, the Tax function comes to rely on human heroics to deliver against expectations or minimum requirements.

This accumulated cost has no line item in the business case. It does not appear on the balance sheet. It’s rarely discussed at the steering committee. But it is real, and it is rising — and the failure to count it distorts almost every transformation ROI calculation in finance and tax today.

This hidden cost deserves a name: Transformation Debt.

Transformation Debt is the off-balance-sheet accumulation of all the tax transformation work a company needs to do in the future.

The Anatomy of Transformation Debt

Transformation Debt is not a single number. It is the cumulative weight of structural choices — the way work is divided, the systems it runs on, how data is collected, the people who do it — that were appropriate when they were made, that have persisted past their useful life, and that now exact a quiet price on the function every day.

In my experience, it accumulates along five recognizable dimensions.

Volume Drift. The business grows. Transaction counts rise. New entities are added. New jurisdictions are absorbed. The function’s structure was sized for a smaller, simpler version of the business. The gap between volume and structure is filled by extending hours, hiring contractors, and escalating spreadsheets. None of this shows up in a strategic plan. It shows up in fatigue, errors, quietly rising costs, and a creeping risk profile of the Tax function.

Regulatory Accumulation. Each new rule — Pillar Two, CAMT, BEAT, evolving state nexus, the next thing — does not arrive in a vacuum. It arrives on top of every prior rule the function is still adapting to. Most organizations absorb regulation through manual extension: a new spreadsheet, a new reconciliation, a new exception process and human heroics. Each addition is small. The accumulation is not.

Workaround Hardening. What begins as a temporary fix becomes a permanent fixture. A spreadsheet built by an analyst on a Tuesday afternoon becomes load-bearing for a quarterly close. A workaround designed to bridge two systems becomes the de facto integration. These structures are invisible in the org chart. They are not invisible in the risk profile.

Key-Person Dependency. Every operating model that has been allowed to drift accumulates institutional knowledge in specific people. The Tax director who knows why the transfer pricing model was built the way it was. The senior analyst who maintains the rate reconciliation largely in her head. This is not a problem until it is. It is also not a cost until it is.

Strategic Drag. This is the most expensive dimension and the least measured. As the gap between what the function is structured to do and what the business now needs widens, the function increasingly absorbs the time and attention of senior leaders.

Tax VPs spend less time on planning and more time on operations. Controllers explain reconciliation gaps instead of analyzing margin. CFOs allocate bandwidth to fixing the function instead of using it. The opportunity cost of senior leadership attention is the largest hidden line item in any transformation business case.

These five dimensions compound. Volume drift creates more workarounds. Workarounds increase key-person dependency. Key-person dependency reduces the function’s ability to absorb regulation. Regulatory accumulation deepens strategic drag. The cost is not additive. It is multiplicative.

Why This Doesn’t Make It Into the Model

There are three reasons Transformation Debt is underweighted in business cases.

First, it is genuinely hard to measure. Efficiency savings from a new platform are relatively easy to quantify — full-time equivalent hours reduced, cycle time compressed, error rates lowered. The cost of compounding operating model decay does not have a meter. It’s felt before it’s measured. Finance organizations are trained to model what they can count, and Transformation Debt resists counting. That doesn’t make it less real.

Second, the do-nothing scenario is rarely extrapolated. Tax VPs are accountable for this year’s budget. The current operating model has owners, and it gets actively managed. But the multi-year projection of where that operating model is headed — with realistic decay assumptions, year over year — is not a standard step in the business case template. The question is not asked, so the answer is never built.

Third — and this is the one worth sitting with — the cost of Transformation Debt is paid in many small currencies that do not sum cleanly. A few extra hours during close. A reconciliation that takes two days instead of half a day. A senior analyst who leaves for a competitor and takes three years of context with her. A regulatory change that requires three months of manual adjustment instead of three weeks. None of these is a board-level number on its own. The aggregate probably is.

A Necessary Acknowledgment

This is not an argument that all transformation pays off.

The cost of poorly scoped, poorly governed, or poorly executed transformation is real and well-documented. Implementation failures are expensive. Selecting the wrong platform is expensive. Underestimating change management is expensive. The history of finance and tax transformation is dense with cases of organizations that paid the cost of change and never realized the benefit.

The argument here is narrower. It’s that the “do nothing” scenario is itself an investment with a rising cost. That cost is rarely modeled. The decision to remain structurally the same is not a decision to avoid risk. It is a decision to accept a different kind of risk, one that compounds quietly rather than visibly.

A serious business case has to be honest about both sides. Transformation has cost and risk. Inaction has cost and risk. Comparing the first against a frozen baseline of the second is not analysis. It is a thumb on the scale.

What Better Looks Like

The fix is not complicated. It requires three practical changes to how transformation decisions are framed, evaluated and presented.

Build two curves, not one. Most business cases produce a single cost curve — the transformation investment, with benefits accruing over time against a flat status quo baseline.

Build a second curve. Project the cost of the current operating model forward with a realistic decay assumption: Tax function cost growing faster than business volume, regulation absorbed through manual methods, senior leadership attention drawn down at a measurable rate. The assumption does not need to be aggressive. It needs to be honest.

Put the two curves side by side. This is the move that changes the conversation. A single ROI number on a slide is an abstraction. Two curves on the same chart — one rising because of transformation cost, one rising because of compounding inefficiency — is something an executive can read in three seconds. Most will recognize the second curve immediately. They have been watching it for years. They have not seen it modeled and drawn before.

Make this a recurring conversation, not a one-time business case exercise. Transformation Debt is not a number you calculate once. It is a quantity that grows. The right cadence is quarterly, alongside operational metrics — a brief, honest assessment of where the function sits on the second curve. Concepts that have a name and a place on the agenda get managed. Concepts that do not are absorbed silently into next year’s run-rate.

Most transformation conversations are framed around whether change is worth what it costs.

When the second curve enters the picture — when the cost of staying the same is modeled honestly, alongside the cost of changing — the conversation shifts. The question is no longer “should we transform?” It becomes “which alternative carries the lower long-run cost?” Sometimes the answer is to transform. Sometimes it is to transform differently — narrower scope, faster cadence, different sequencing. Sometimes it is to accept the debt for another year while a better option matures.

The point is not that transformation always wins. The point is that the decision becomes a real decision — informed by both costs, not just one.

Maybe the cost of doing very little or nothing is one your management is prepared to accept, even after seeing a true picture of the Tax function’s cost escalation for the foreseeable future.

Or maybe the post-transformation cost model is equally striking, but for the opposite reason — the transformation savings are too large to leave on the table.

The decision is yours. I would simply encourage you to study the full picture as you weigh your appetite for substantive transformation or incremental adjustment.

That’s Transformation Debt. And like all debt, the interest compounds whether or not it appears on the balance sheet

Are you familiar with Jevons Paradox?

It may become one of the most important concepts executives need to understand in the AI era.

In the 1860s, economist William Stanley Jevons observed something unexpected:

As steam engines became more efficient, Britain did not use less coal.
It used dramatically more. Why?

Because efficiency lowered cost. Lower cost expanded usage. Expanded usage increased total demand.

The efficiency gains themselves were not the real prize.

The real differentiator was the ability to reinvest those gains at lower cost and scale economic activity even further.

Subscribe now

This counterintuitive idea became known as Jevons Paradox.

It may explain why many assumptions about AI and workforce reduction could prove incomplete.

If AI makes knowledge work dramatically cheaper, companies may not consume less analysis, forecasting, software development, tax modeling, research, or reporting.

They may consume exponentially more of it.

History suggests that when a capability becomes cheaper and more accessible, organizations tend to expand usage rather than shrink it.

That raises a much bigger strategic question for leadership teams:

Will AI primarily become a cost-cutting tool? Or will it become a capability expansion engine?

Efficiency gains do not always reduce demand.

Sometimes they unleash it in ways we cannot fully see in the moment.

Walk through almost any tax or finance transformation program right now and you will hear two messages at the same time, usually from the same executives.

The first is the formal message: we are transforming. The roadmap is approved, the platform is selected, the AI pilots are funded, the consulting partner is engaged. There are steering committees and KPIs and slide decks.

The second message is operational, and almost no one says it out loud: do not disrupt anything. The compliance calendar still needs to be hit. The audit is still due. The provision still has to be filed on the same timeline, in the same format, with the same level of accuracy as last year.

Both messages are sincere. Both are repeated by the same leadership team in the same week. And both, held side by side, describe a contradiction that most organizations underestimate.

Subscribe now

The contradiction is structural

It’s easy to read this and reach for the familiar explanation: people resist change. Middle managers protect their turf. Staff are anxious about AI. That story is comfortable because it locates the problem within people, which makes it solvable through training, town halls, or a stronger change management workstream.

The contradiction is more durable than that, and it lives at the level of the operating model, not the org chart.

Most tax and finance organizations have been deliberately engineered, over decades, to optimize for a narrow and demanding set of outcomes: precision, predictability, continuity, control, audit defensibility, and deadline adherence. Every process, review layer, workpaper convention, staffing assumption, and escalation protocol exists in service of those outcomes.

Transformation introduces the opposite. It introduces temporary inefficiency, ambiguity, learning curves, productivity dips, role confusion, governance friction, and experimentation risk. Those are not side effects of transformation, they are the conditions under which actual transformation occurs. Skip them and you have not transformed; you have re-decorated.

The structural conflict, then, is this: the function is asked to produce the conditions of disruption while still being measured against the standards of continuity. Employees respond rationally to that. They protect the close, they protect the filing, they protect the audit and the transformation work is what slips.

Industry research keeps surfacing this same pattern from different angles. Year after year, tax and finance leaders report ambitious operating-model ambitions alongside the same complaint that capacity is the binding constraint. Programs are not failing because the technology does not work. They are failing because the function was never given the operating space to absorb the disruption.

I have seen this movie at companies of every size, in every industry, with every consulting partner, on every platform. It is not a failure of will. It is what happens when a function with a non-negotiable production calendar tries to run a transformation program out of its own residual capacity. Transformation is treated as the variable that flexes around the work, when in fact the work is what should have flexed around the transformation.

Share

The leaders who actually move the function forward are the ones who recognize the trade and make it explicitly. They accept a temporary, visible degradation in some part of the operating rhythm, a slower close one quarter, a deferred analytic, a reduced scope on a non-statutory report, in exchange for the bandwidth to do the work. I’m not suggesting lowering quality standards is the solution, but some well-thought-out trade-offs need to happen. But most organizations refuse the trade. And then they wonder why the transformation did not happen.

Why this trade is so rarely made

The reason the trade is so rarely made is that the cost of continuity is loud and the cost of forgone transformation is silent.

When a transformation slips, no auditor asks about it. No regulator issues a fine. No analyst downgrades the stock. The cost compounds quietly, over years, in the form of a function that ages out of strategic relevance, less able to support the business, more dependent on offshore arbitrage and human heroics, more exposed to the next regulatory shift it cannot absorb.

When the close slips by two days, every one of those parties’ notices. The CFO is on the phone. The audit committee has questions. The story writes itself.

A rational function, given those incentives, will always protect continuity at the expense of transformation. That is not a failure of leadership, it’s the leadership operating exactly as the incentive structure designed them to. The contradiction is not that people are behaving irrationally; it is that the organization is asking for two rational outputs that cannot be optimized simultaneously.

What this means for leaders thinking about transformation

If you are running a transformation program inside a tax or finance organization, the question worth asking is not do we have the right roadmap or did we pick the right platform. The honest question is what operational disruption are we willing to tolerate, in which areas, for how long, and visible to whom.

If the answer is none, the program will not transform anything. It will produce a new system that runs the old process slightly faster, and the function will tell itself a story about progress then declare victory.

If the answer is some, in these specific areas, for this specific window, with this specific air cover from the CFO and the audit committee, then the program has a chance because the contradiction has been named and the trade has been made.

Transformation is much broader than decisions about technology and process. It’s also about which forms of disruption an organization is structurally willing to absorb. Almost everything else follows from that.

The unspoken truth

The contradiction at the heart of most tax and finance transformations is not that the work is too hard, the technology too immature, or the talent too scarce. It’s that organizations want the outcomes of transformation without tolerating the operational disruption required to achieve them.

Until that contradiction is named and budgeted for, calendared for, and protected by the leadership team, the function will keep running the same play, year after year. New roadmap. Same result.

Share Tax Transformation & Technology

Most tax transformation discussions focus heavily on technology.

AI. Automation. Data platforms. Workflow tools. ERP modernization. Dashboards. Copilots.

All of those things matter and they matter big.

But after spending decades inside large tax functions and leading transformation efforts, I’ve come to believe the larger challenge is rarely technology.

It’s human.

The uncomfortable reality is that many tax transformation efforts struggle not because the software is inadequate, but because organizations consistently underestimate the human infrastructure required to support meaningful change.

What do I mean by ‘human infrastructure’?

It includes:

• Leadership alignment

• Communication plans and execution

• Governance effectiveness

• Organizational structure

• Training

• Incentives, and

• Emotional realities that accompany disruption inside complex organizations

Technology implementation is often the easy part.

Sustained organizational adoption is much harder.

And increasingly, research in organizational transformation is pointing toward the same conclusion. McKinsey, Prosci, and decades of organizational change literature have consistently emphasized that successful transformation depends as much on leadership, communication, behavioral adoption, and organizational alignment as it does on technology capability. McKinsey on organizational transformation

Technology Is Rarely the First Failure Point

Modern tax technology has improved dramatically over the last decade. Most large organizations now have access to:

• Sophisticated tax provision systems

• Cloud ERP environments

• Automation capabilities

• Workflow platforms

• Powerful data and AI tools, and

• More implementation partners than ever before

Yet despite this progress, many finance and tax functions remain constrained by:

• Fragmented data

• Manual workarounds

• Spreadsheet dependency

• Disconnected processes, and

• Inconsistent operating models

Why?

Because technology alone rarely changes organizational behavior.

In many instances, organizations are layering modern technology on top of operating models originally designed for a very different era.

The result is often digital complexity rather than true transformation.

The Human Infrastructure Is Often Missing

Many transformation efforts begin with detailed technology roadmaps.

Far fewer begin with equally detailed discussions about:

• Organizational readiness

• Communication strategy

• Governance structures

• Change management

• Role re-design, or

• Long-term adoption

These are not secondary considerations.

They are central to success.

Transformation Is Still Treated Like a “Night Job”

One of the most common patterns I’ve observed is that organizations frequently attempt transformation while expecting already overextended teams to absorb the effort on top of their existing responsibilities.

Tax professionals are already operating under:

• Compliance deadlines,

• Audit demands,

• Regulatory complexity,

• Staffing constraints, and

• Constant business change

Then transformation becomes an additional responsibility layered on top of the existing workload.

Not surprisingly, progress becomes sporadic.

Projects lose momentum.
Meetings become status updates.
Adoption slows.
Competing priorities win.

In too many organizations, transformation is still treated as a temporary initiative rather than a sustained organizational capability that, frankly, doesn’t end but exists indefinitely to support constant change that on some scale is inevitable.

That distinction matters.

The organizations making the most meaningful progress tend to institutionalize transformation rather than treating it as episodic project work.

Communication Matters More Than Most Organizations Realize

Successful transformation efforts require relentless communication.

Not once at kickoff.

Continuously.

People need to understand:

• Why change is occurring,

• How priorities are evolving,

• What success looks like,

• How their roles may change, and

• How leadership intends to support the transition

Silence creates uncertainty.

Uncertainty creates resistance.

Resistance slows adoption long before leadership recognizes the problem.

Research from Prosci and other organizational change specialists has consistently shown that communication frequency, sponsorship visibility, and frontline engagement materially improve transformation adoption outcomes. Prosci change management research

Organizational Structure Matters

Large-scale transformation efforts require dedicated organizational support.

That may include:

• A transformation office(s)

• PMO structure(s)

• Executive steering committee(s)

• Embedded change leaders

• Cross-functional governance, and

• Clearly defined accountability

Without sufficient structure, transformation efforts often drift into ambiguity.

Projects continue but real transformation stalls.

Many organizations underestimate the scale, duration, and organizational complexity associated with sustained transformation efforts.

Technology changes quickly.

Organizations do not.

Training should begin earlier

One of the most overlooked aspects of transformation is training.

Too often, organizations wait until rollout to begin educating teams on new processes and systems. By then, resistance and anxiety may already be deeply embedded.

The most successful organizations tend to involve users much earlier in the process.

They create familiarity during development rather than forcing adaptation after deployment.

People support what they help build.

Executive Sponsorship Is Not a Kickoff Event

Many organizations correctly recognize the importance of executive sponsorship.

But executive sponsorship is frequently treated as:

• Initial approval only

• Kickoff participation, or

• Occasional steering committee visibility

That’s not enough. Transformation requires sustained executive reinforcement throughout the life of the initiative.

Leadership should consistently communicate these four things:

• This effort matters

• Priorities will remain aligned

• Modernization is not optional, and

• Organizational learning will be supported throughout the process

Prosci’s benchmarking research has repeatedly identified active and visible executive sponsorship as one of the single greatest contributors to successful organizational change. Prosci on executive sponsorship

The strongest transformation cultures often send a message that sounds something like this:

We are going to modernize.

This effort matters.

It will require change.

We will make mistakes.

We will learn as we go.

But we are committed to the outcome and committed to supporting the people helping us to move through the transformation.

That balance matters enormously.

Organizations that create fear around transformation often unintentionally suppress transparency, experimentation, and learning. These are the ingredients for self-sabotaging a transformation program.

Organizations that lack accountability often find themselves adrift.

The most effective transformation environments tend to combine:

• Disciplined commitment

• Organizational support, and

• Permission to learn

I’ve Seen This Movie …

My observation - organizations often launch transformation initiatives with strong enthusiasm and detailed technology plans…

…but quietly continue rewarding the same behaviors, priorities, and risk avoidance patterns that existed before the transformation began.

Employees notice the inconsistency immediately.

And when they do, adoption slows long before leadership realizes it.

AI Is Increasing the Pressure, Not Eliminating the Problem

Artificial intelligence is accelerating the urgency surrounding transformation efforts. It’s also exposing foundational weaknesses that have existed for years:

• Fragmented data environments

• Inconsistent governance

• Unclear ownership, and

• Disconnected operating models

AI may improve productivity dramatically in some areas.

But organizations that have not addressed the underlying human and operational dimensions of transformation may find that AI simply magnifies existing organizational complexity.

Technology can accelerate an organization. It can also accelerate dysfunction.

Subscribe now

What Successful Organizations Tend to Do Differently

The organizations making real progress in tax transformation often share several characteristics.

They:

• Treat transformation as a long-term organizational capability

• Dedicate real capacity to change efforts

• Invest heavily in communication and training

• Align transformation to broader business priorities

• Build governance structures that survive leadership turnover

• Focus relentlessly on process improvement, and

• Recognize that adoption is just as important as implementation

Most importantly, they understand that transformation is not purely a technology exercise.

It is an organizational exercise.

And organizations are ultimately made up of people.

Final Thought

The tax function at global multinational companies is entering a period of enormous change.

AI will matter.
Data will matter.
Technology will matter.

But organizations that focus exclusively on tools while underinvesting in leadership, communication, governance, training, and organizational readiness may continue to struggle to achieve durable transformation outcomes.

The future of tax transformation will not be determined solely by technology capability.

It will also be determined by how effectively organizations help people move through change.