Stop buying AI seats. Start buying decisions.

Accenture bought 743,000 Copilot seats.

That was one line in a quarter where Microsoft passed 20 million paid Copilot seats, up from 15 million three months earlier — the fastest growth it has reported since launch¹. Bayer, Johnson & Johnson, Mercedes and Roche are each past 90,000.

Everyone read that as adoption. Adoption is not the interesting part. The interesting part is what happened next, measured somewhere else entirely: PwC asked 4,454 chief executives across 95 countries what twelve months of AI had returned, and 56% reported neither a revenue increase nor a cost reduction. One in eight — 12% — reported both².

Twenty million seats. Fifty-six per cent nothing. Those two numbers are not a contradiction to be explained away. They are the same fact, viewed from the purchasing end and the P&L end.

The seat is the bug

A seat is a licence to access a capability. It is not a change to how work runs.

That distinction sounds pedantic until you follow the money. When you buy a seat, you have bought a person the option to use a model, at their discretion, on whatever they decide to point it at. Nothing about the process changed. No step was removed, no handoff eliminated, no cycle time cut. Someone drafts an email faster. The email still gets written, sent, read, actioned and filed by the same people in the same sequence.

That is why the dashboards look healthy while the P&L doesn't move. Utilisation is not a business outcome. A well-adopted seat produces exactly one measurable thing: a bill.

And it is a strange bill, because it is priced per person rather than per unit of work. Your cost scales with headcount and with enthusiasm — how many people have it, how chatty they are — rather than with how many claims you process or how many jobs you quote. There is no denominator. You cannot say what a seat costs per transaction, because a seat is not attached to a transaction.

You already know this. You have looked at the licence count and the adoption chart and thought, quietly: what did we actually get? Nobody asks it out loud because seats are cheap enough that asking looks like penny-pinching, and because the answer would implicate the person who signed off.

Seats are about to stop being cheap enough.

The price you are paying is not the price

Here is the second half of the squeeze, and it is not a forecast — it is last year's audited accounts.

OpenAI earned $13.07 billion in 2025 and ran a $20.92 billion operating loss³. Round the numbers and the company spent something close to $2.60 for every dollar it took in. Its headline net loss of nearly $39 billion is mostly a one-time charge from its restructure, and quoting that would overstate the case. The operating loss is the honest figure, and the honest figure is enough.

That gap is not a rounding error in somebody else's business. It is your discount.

Investors are funding it, and they are funding it at a scale that cannot continue indefinitely: global venture funding hit a record $510 billion in the first half of this year alone, more than all of 2025, with over 70% of second-quarter capital going to AI companies⁴. That money buys market share, and the way it buys market share is by selling inference for less than it costs to serve.

I am not going to tell you when that ends, because nobody credible can. What I will tell you is what it means for a per-seat contract: your renewal price is set by somebody else's willingness to keep losing money. You do not control that variable, you cannot forecast it, and it is currently the largest hidden assumption in your AI budget.

The standard advice here is to negotiate harder, get a multi-year lock, push adoption up so the per-seat cost looks better. All of that treats the price as the problem. The price is not the problem. The unit is the problem. A cheaper seat is still a cost with no denominator.

Two questions, in order

So run the audit, and run it in two passes. The order matters — the second question is meaningless without the first.

Pass one: who actually needs the desktop tool, and can use it?

Go line by line down the licence bill and ask, for the people holding those seats: does any named piece of work run differently because they have this? Not "do they like it", not "do they log in". Does something a customer or a colleague receives arrive faster, cheaper, or better, in a way somebody would put a number against?

You will find three populations. A small group for whom the answer is obviously yes — they have both the work and the skill, and they will tell you exactly what changed. A middle group who use it pleasantly and produce nothing you can point at. And a group who were provisioned because provisioning everybody was easier than deciding.

Most licence bills are mostly the last two. That is not a failure of your people; it is what happens when access is bought instead of change.

Pass two: at full price, would you still buy it?

For every line that survived pass one, ask the harder question: if this cost three times what it costs today, would you renew?

Three is not a prediction. It is a stress test, and it is roughly the shape of what "no subsidy" looks like against a business spending $2.60 to earn a dollar. Some lines will pass instantly — the ones attached to work whose value obviously exceeds the price. Some will pass for a handful of people and fail for the rest, which tells you the seat count is wrong rather than the tool. And some will fail flatly, which tells you the thing was only ever viable while somebody else was paying part of it.

Every line lands in one of three places.

• Theatre. No named work runs differently. It fails pass one, and pass two never applies. Cancelling it changes nothing except the bill.
• Countdown. Real work runs differently, but the economics only hold at today's subsidised price. There is a clock on this line that you did not start and cannot see. This is the dangerous column, because it looks exactly like success on every dashboard you own. You are renting an advantage on somebody else's balance sheet.
• Load-bearing. Real work runs differently, and it would still be worth buying at three times the price. The business leans on it and it does not move.

Then add up the seats in each column, and the bill takes the name of whichever holds it: theatre if most of your seats buy access rather than outcomes, load-bearing if half or more earn their keep at any plausible price, countdown for everything in between — real work, real value, running on a clock somebody else set.

Where the money should go instead

Now the part that makes the audit worth running, because cancelling licences is not a strategy either.

The alternative to a seat is a judgment point: a single bounded decision, made by a model, at one specific step inside a process — classify this, match that, flag this exception, summarise this for the next step. Not a model sitting across the whole workflow reading everything. One call, on one question, with an escalation to a person when confidence is low.

That shape fixes exactly the thing the seat broke: it has a denominator. A judgment point costs a predictable fraction of a cent per transaction, because you have told it precisely what to decide. You can put that number in a spreadsheet next to the cost of the step it replaced, and you can defend it at any token price, because if prices triple you know precisely what triples and precisely what it saves. A seat estate has no such row. It has a headcount and a hope.

There is a second benefit that shows up a year later. Because the model only ever sees one bounded question, you can change which model answers it. Judgment points are portable in a way that a vendor's desktop assistant deliberately is not. When prices move — and they will move in both directions — you are choosing, not renegotiating.

But the denominator is not the real prize. The real prize is what the judgment point unlocks.

What you are actually buying

Think about the processes your company has never automated. They were rarely blocked by the mechanical work — moving data, filling fields, routing, chasing, reconciling against a rule. That has been automatable for twenty years and you have almost certainly done it wherever it was worth doing.

They were blocked in the middle, at one or two moments where somebody had to read something and decide. Does this invoice match this contract. Is this complaint about the product or the delivery. Does this application need a human. Is this photo showing the damage the claim describes. One unautomatable moment in the centre of an otherwise mechanical chain was enough to keep the entire process manual — and so it stayed manual, year after year, and eventually stopped being discussed as a candidate at all.

That is the category that just opened. Not "AI can do knowledge work" — the vague version that sold twenty million seats. The specific version: entire classes of process that were permanently off the automation list can now run end to end, because the one step that blocked them has a machine answer with a confidence signal and a human escalation behind it.

So when you fund a judgment point, you are not buying a faster step. You are buying the process that step was holding hostage.

Front-load the tokens into the build

There is a second thing that changed, and it is the one most budgets have not caught up with: building the automation itself got dramatically cheaper.

Agentic tooling now does the construction work — reading the existing system, writing the integration, generating the tests, wiring the exception paths — that used to be a six-month project with a vendor and a statement of work. That is genuinely token-hungry. It is also the right place to be token-hungry, because you spend it once.

Which gives you the shape of a sane AI budget, and it is nothing like a seat licence:

• The build bill. Large, deliberate, one-off. This is where you want your tokens going: using agentic tooling to construct the automated process around the judgment points. Spend generously here. It is capital, and it is the cheapest this work has ever been to do.
• The run bill. Small, per-transaction, forever. Once built, the process runs the way processes have always run: deterministic code doing the mechanical steps at effectively no cost, calling out to a model only at the judgment points, only when it reaches one, only for as long as that decision takes.

That split is the whole argument for why rising token prices should not frighten a company that has done this work, and should frighten one that hasn't. Your run bill touches a model at three or four moments per transaction instead of continuously. If prices double, a per-seat estate doubles across every person who has one. A built process barely moves, because most of what it does was never a model call in the first place.

The uncomfortable corollary: the build is cheapest to do now, in the noisy part of the cycle, while agentic tooling is priced the way it currently is. The companies that convert seat money into build money this year end up with processes that run cheaply for a decade. The ones that keep renewing per-person licences are buying the same thin capability again every twelve months, at a price somebody else sets.

The best argument against everything above

I want to hand you the strongest counter-argument myself, because it is a good one and somebody in the room will make it.

It runs like this. Tokens are about to get much cheaper anyway, so optimising now is premature. Open-weight models have flooded the market and commoditised what the frontier labs were charging a premium for. Underneath the models, every layer of the stack is on a downward cost curve — chips, power draw per unit of work, cooling, the useful life of the hardware, and eventually whatever exotic answers get built for the energy problem. Richard Sutton named the underlying dynamic in 2019 as the bitter lesson: general approaches that scale with computation beat cleverly hand-built ones, precisely because they ride the falling cost of computation⁵. Betting against cheap compute has been a losing trade for seventy years.

I think that argument is right. I do not think it gets you out of this.

You do not get to spend the long horizon's prices today. Every month between now and then runs on this year's bill, and that is the bill your board is looking at. A company operating an efficient process and a company operating a wasteful one both benefit when prices fall — but only one of them is still in a position to press the advantage when it happens. Optimisation is how you win the short and medium term while you wait for the long term to arrive and prove you right.

And the horrific bills are available at today's prices. This is the part people underestimate. The failure mode is almost never the price per token; it is the shape of the usage — a model re-reading whole documents on every transaction, an agent looping because nothing told it when to stop, a workflow calling a frontier model for a decision a small one would have made correctly. Companies discover this monthly, at current, subsidised, historically-cheap prices. If prices fall ninety per cent and the shape never changed, a horrific bill becomes a merely bad one.

Now the concession, and it is a real one. You should absolutely run an experimental phase, and you should run it early. You cannot reason your way from a whiteboard to which judgment points matter in your business; you find out by trying things, cheaply, in places where being wrong is survivable. Spend the money to learn. And shortcut the phase wherever you can by learning from everyone who has already run the experiment — the failures are widely published now, and there is no prize for discovering them personally.

But a phase is a phase. It ends, and what it produces is decisions: which processes, which judgment points, which model for which decision, what gets built once and what gets paid for forever. That is the same discipline this audit is asking for. The experiment is how you earn the right to make those calls with evidence instead of vendor slides.

Cheap compute rewards whoever is positioned to use it. It does not retroactively refund whoever wasn't.

And the people

This is the part that decides whether the whole exercise is worth anything. When a built process removes hours from your week, those hours need a named destination or they evaporate into slack while the dashboard stays green. The destination is the work you currently cannot deliver enough of: the difficult conversation, the bespoke job you turn away, the follow-up nobody has time for, the quality pass that never survives month-end. That work is what customers choose you for, and it is the only part of this that a competitor cannot buy from the same vendor you did.

Seats spread a thin capability across everybody, forever, at a price you don't set. A built process puts a deep one exactly where the work was stuck, pays for itself once, and puts your people where the process was never the point.

One boundary, honestly

This audit cannot tell you what tokens will cost next year. Nobody can, and no vendor is going to volunteer it. What it does is remove the assumption from your budget — it tells you which lines are already earning at any price, which lines are living on somebody else's losses, and which lines were never doing anything at all.

Run it on one licence bill. Give yourself thirty minutes and the honest answers.

By the end you will have every line labelled, the seat count you actually need, and one word for the renewal conversation.

Start with your largest line by seat count. Ask pass one — what runs differently? — and if the room cannot name a workflow in a sentence, you have found the money for your first judgment point.

References

Footnotes

1. No Jitter. (2026, April 29). Microsoft 365 Copilot hits 20 million paid seats. https://www.nojitter.com/ai-automation/microsoft-365-copilot-hits-20-million-paid-seats ↩

2. Fortune. (2026, January 19). 56% of companies getting nothing out of AI, PwC research says; chairman blames forgetting the basics. https://fortune.com/2026/01/19/pwc-global-chairman-mohamed-kande-ai-nothing-basics-29th-ceo-survey-davos-world-economic-forum/ ↩

3. TechSpot. (2026, June 17). OpenAI made $13 billion in 2025 and lost $21 billion doing it. https://www.techspot.com/news/112794-openai-made-13-billion-2025-lost-21-billion.html ↩

4. Teare, G. (2026, July 2). Global startup investment hit record $510B in H1 2026 as AI boom accelerates funding and exits. Crunchbase News. https://news.crunchbase.com/venture/global-startup-exits-ipo-ma-soar-ai-q2-h1-2026/ ↩

5. Sutton, R. (2019). The bitter lesson. Summarised at https://en.wikipedia.org/wiki/Bitter_lesson — "General approaches that scale with available computational power tend to outperform ones based on domain-specific understanding because they are better at taking advantage of the falling cost of computation over time." ↩