What Is the Exit Clause Actually Worth? Pricing AI Vendor Optionality
Every AI strategy says keep your options open. Almost none of them price the option. Real-options arithmetic on a foundation-model commitment, two worked cases where the answer goes opposite ways, and why your migration cost is the variable that decides it.
Every AI strategy document written in the last two years says some version of keep your options open. Avoid lock-in. Preserve the ability to switch providers. I have written that sentence myself, on this site, more than once. What almost none of those documents do, including mine, is say what the option is worth, and a strategic principle that has never been priced is not a principle, it is a preference.
The gap shows up the moment a real contract lands on the table. A foundation-model provider offers a three-year commitment at a serious discount. The alternative is a twelve-month term at close to list with a ninety-day exit. Procurement wants the discount, the architects want the exit, and the argument runs on adjectives, because nobody in the room has converted either position into a number. The discount is a number. The flexibility is not. The number wins, and everybody involved describes the outcome as commercial reality rather than what it actually was, which is one side of the argument arriving unarmed.
There is a standard way to price the other side, it is forty years old, and it does not require a finance function to run.
The method, in the only version that matters
Stewart Myers named the idea in 1977: part of what a company is worth is not its assets but its options, the things it has the right but not the obligation to do later. Timothy Luehrman turned that into something a manager could use in a pair of Harvard Business Review pieces in 1998, and the useful residue for our purposes is small enough to fit on a slide.
An exit right is an option. It has a price, which is whatever you gave up to get it. It has a value, which is what it will save you if you use it, discounted by the chance that you never do. Compare the two. That is the whole method.
Three inputs:
The premium. The difference in total contract value between the committed deal and the flexible one. This is the one number in the exercise that is not an estimate, because both figures are written on the two contracts in front of you.
The probability. The chance that something appears inside the term that is worth switching to. Not the chance that a better model ships, which is close to one and therefore useless. The chance that a better model ships and is enough better, on your workload, to justify moving to it.
The net saving. What switching actually saves over the remaining term, minus what migration costs you. This is where most of the argument lives and where most strategy documents stop paying attention.
Expected value of the option is probability times net saving. If it exceeds the premium, buy the flexibility. If it does not, the exit clause is a comfort purchase and the committed deal is the better commercial decision. The arithmetic is trivial. The discipline of doing it at all is what is missing.
Case one, where the answer goes against the conventional advice
An enterprise running around two million euros a year of inference on a single foundation-model provider. The numbers here are illustrative, chosen because they are the shape I keep seeing rather than because they are anyone’s actual contract.
Deal A 3-year commitment, 25% discount €1.50M/yr €4.50M total
Deal B 1-year term, 10% discount, 90-day exit €1.80M/yr €5.40M over 3 yrs
Premium for flexibility €900kNow the option. Say there is a fifty per cent chance that within the next eighteen months a substitute lands that is thirty per cent cheaper on this workload at equivalent quality. That is a generous assumption in the provider’s favour and roughly matches the last two years of release cadence.
If it happens at month eighteen, the remaining term is eighteen months of spend, so about €2.70M at Deal B pricing. Thirty per cent of that is €810k. Migration on a workload of this size, in an organisation with the usual amount of prompt logic and evaluation tooling welded to one provider’s behaviour, runs six to nine months of a small team. Call it €400k, which is conservative.
Gross saving if it happens €810k
Migration cost −€400k
Net saving €410k
Probability × 0.5
Expected option value €205k
Premium paid for the option €900kThe exit clause costs four times what it is worth. On these numbers the committed deal is not procurement steamrolling architecture, it is the right answer, and the architects arguing for flexibility were arguing for a €700k loss they had no way of seeing because nobody had written it down.
I want to be careful about what this case does and does not establish. It is one set of assumptions. Move the probability to eighty per cent and the option is worth €328k, still under the premium. Move the discount gap from fifteen points to five and the premium drops to €300k and the option wins comfortably. The conclusion is not that commitment beats flexibility. It is that the answer is computable and most organisations are guessing.
Case two, and the variable that actually decides it
Same company, same spend, same offers. One difference: this organisation put a provider abstraction layer in eighteen months ago, keeps its evaluation harness provider-neutral, and has already moved one workload between providers, which is the part that matters. Migration is now three weeks of one team, not two quarters.
Gross saving if it happens €810k
Migration cost −€40k
Net saving €770k
Probability × 0.5
Expected option value €385kStill under a €900k premium, which is worth sitting with rather than skipping past. Even with cheap migration, a fifteen-point discount gap is a lot to pay for one option. But now look at what else changed. The same abstraction that took migration from €400k to €40k applies to every provider relationship this company has, on every renewal, for as long as the layer survives. It is not buying one option. It is lowering the strike price on all of them at once.
That is the finding worth taking to a board, and it inverts the usual procurement conversation. The lever is not the clause. Negotiating a ninety-day exit into a contract changes one number in one deal. Spending the same effort making the switch cheap changes the value of every exit right you will ever hold, including the ones you have not negotiated yet.
Which reframes the platform and abstraction argument that this site has made on sovereignty grounds. The sovereignty case for provider neutrality is real and it is also, for most companies, not the case that moves a budget. The commercial case is that migration cost is the multiplier on every option in your vendor portfolio, and it is the only input in the whole calculation you control directly.
What to do with the three numbers
The temptation with any borrowed method is to make it more elaborate than it needs to be. Luehrman’s own papers go to Black-Scholes and option-value tables, and if your treasury function wants to run it that way they are welcome. Nothing above needs it.
Put the premium, the probability and the net saving on one slide and force the argument onto them. The precision is not the point. The point is that the assumptions become arguable. A CFO who believes the switch probability is twenty per cent rather than fifty can now say so, the number moves, and the disagreement is about something specific. Compare that with the version where architecture asserts that flexibility is strategically important and procurement asserts that the discount is real, and the loudest person wins.
Three ways this goes wrong in practice, all of which I have watched:
The probability gets set by whoever wants the outcome. If the person who wants the exit clause also supplies the switching probability, you have not priced anything, you have decorated a preference. Get it from someone with no position, or bracket it and show the range.
The migration cost gets estimated by the people who would do the migration, on the assumption that everything works. Take whatever they say and ask what it was last time. If there was no last time, the honest number is much higher than the estimate, and case one is what you are in.
The option gets bought and never exercised. An exit right nobody uses is a premium paid for nothing, and the pattern is common enough to have a shape: the clause survives the negotiation, the migration cost never comes down, and by renewal the switch is as impractical as it would have been under the committed deal. If you buy the flexibility, fund the thing that makes it exercisable in the same budget cycle. Otherwise take the discount.
The verdict
Keep your options open is not a strategy, it is a placeholder for arithmetic nobody has done. The arithmetic takes an afternoon, needs three numbers, and produces an answer that is frequently the opposite of what the AI-strategy literature assumes, because that literature almost never carries a migration cost.
For most enterprises today, on today’s discount structures and today’s switching costs, the committed deal is the better commercial decision and the flexibility premium is not worth paying. That is uncomfortable to write on a site that has argued for provider neutrality, and it is what the numbers say. The way out is not to argue harder for exit clauses. It is to make switching cheap enough that flexibility becomes worth its price, and then buy it.
What this connects to
The governance hub covers the wider decision set this sits inside. The strategy considerations page maps six risks onto six contract clause types, and this page supplies the valuation column that mapping was missing: an exit clause is worth pricing, not just drafting. The sovereignty page makes the provider-neutrality argument on independence grounds; read this alongside it for the commercial version, which is the one that moves budgets. The CTO page covers who owns this decision when procurement and architecture disagree, which on the evidence of the two cases above they will.
Sources
- Myers, S. (1977), “Determinants of Corporate Borrowing,” Journal of Financial Economics 5(2) — the origin of real-options thinking: part of a firm’s value is the options it holds, not the assets it owns
- Luehrman, T. (1998), “Investment Opportunities as Real Options: Getting Started on the Numbers,” Harvard Business Review — the manager-usable form of the method, and the source of the simplification used here
- Luehrman, T. (1998), “Strategy as a Portfolio of Real Options,” Harvard Business Review — the portfolio framing behind the case-two argument that migration cost is a multiplier across every vendor relationship, not a single deal input
- Related: governance hub, strategy considerations, AI sovereignty, AI for the CTO
Methodology: both worked examples are constructed, not client data, and are labelled as such in the text. The spend levels, discount spreads and migration ranges reflect the shape of foundation-model commitments I have seen across fractional CTO and CIO engagements rather than any single contract, and no client figures appear here. The method is public and cited above; the arithmetic is deliberately simple enough to reproduce in a spreadsheet in ten minutes, which is the point. If your own numbers produce the opposite answer, that is worth sending and I will fold it into the next revision.
