TL;DR Every committed-spend contract embeds a growth assumption, usually last year’s rate carried forward, and nobody writes the rate down. The assumption is satisfied by growth being sustained, not by usage being high, so the commit breaks on deceleration: the account can grow every quarter and still land under the number. The miss stays invisible while committed revenue prints records, then surfaces at renewal as an argument. Ask what growth rate the sizing assumed, and whether the account is growing slower than that.
An account misses its AI usage commitment after a year in which its consumption rose every quarter. Nothing declined. No seats churned, no workloads left, adoption widened. The number was missed anyway, because the number was never only a volume. It was a bet on a growth rate, written into the contract as a volume. The commit is the artifact carrying the most weight in AI software pricing right now, and the assumption inside it receives less scrutiny than any other term on the paper.
The Commit Is a Discount Attached to a Forecast
Committed consumption is the dominant AI deal shape: the customer commits to a volume or a spend level, earns a discounted rate for the commitment, and the contract defines what happens around that number. The discount mechanics of that layer, the schedules, the calibration, the path back when usage lands under the number, belong to our article on volume discounts in AI consumption pricing, and nothing here re-derives them.
Both sides have good reasons to sign. The vendor plans against committed revenue; the buyer earns a better rate and a known bill. Peer-reviewed research on plan choice finds that buyers pay for predictability itself, the insurance of knowing the number in advance. The market recommends the structure for the same reason: a minimum commitment creates a predictable revenue baseline while leaving room for usage to scale.
A forecast can miss in two directions. The industry has an instrument list for overrun. The other direction is this article.
Every Commit Embeds a Growth Assumption
Minimum spend, annual commitments, and ramp pricing are standard custom terms in enterprise software deals. Each is denominated in a value metric, tokens consumed, API calls, or a credit built on top of them. And each was sized somehow.
Sized on the trailing record
The sizing almost always inherits the trailing record. Last year’s consumption plus last year’s growth becomes next year’s number, whether the arithmetic ran in a spreadsheet or on instinct in a renewal meeting. Multi-year ramps make the assumption explicit and compounding: year two sits above year one because growth is presumed to continue. Flat annual commits carry the same assumption silently. A commit matching the account’s run rate plus its recent trajectory has a growth rate inside it whether anyone named it or not. Either way, the assumption is in the contract.
Forecasts miss. Economic research on usage-based plans finds that buyers systematically misestimate the variability of their own future usage, and the misses run in both directions. Nothing about a committed-spend structure repeals that. The commit simply decides in advance who carries each direction of the miss.
Satisfied by sustained growth, not by a high level
AI consumption makes this acute for a mechanical reason: the amount of usage a given task produces is not stable, and the vendors are the ones moving it. OpenAI said its GPT-5.6 models use fewer output tokens, naming customer backlash over cost as the reason. Editors and agent runtimes ship efficiency work of their own. So an account’s usage growth can slow sharply while its adoption widens, its seat count rises, and its satisfaction improves. The usage curve bends for reasons that have nothing to do with the customer’s health.
A committed-spend contract sized to last year’s growth rate does not need usage to fall to become a dispute. It only needs usage to grow more slowly than the commit assumed.
Deceleration Breaks It. Decline Is Not Required
The account grew every quarter and still missed the number
Walk the shape. An account growing fast signs a commit sized to that pace. Growth slows, not to zero, but to a rate most vendors would celebrate. Every quarter’s usage runs higher than the last. The account never declines. And cumulative consumption still lands under the committed number, because the commit was sized to the old rate and the account is compounding at the new one. The gap opens a little in the first quarter, more in the second, and compounds until the term ends with an account that grew every quarter and still missed the number.
No formula is needed to see this; it falls out of compounding. What is needed is the right series. Commit utilization reads differently against the growth rate the sizing assumed than against last quarter alone. Quarter-over-quarter growth of metered revenue is a series most vendors can produce on demand; whether that growth is itself slowing is the second derivative of consumption, plain arithmetic on numbers already in the reporting stack, and it is the quantity the commit’s sizing staked itself on. The pace at which the usage mix is shifting is a third series in the same family. Every vendor has these numbers; almost nobody reads them mid-term.
We have used this arithmetic in public before: our read of GitHub Copilot’s record metered quarter argued that a record quarter which decelerates sharply the next quarter is the tell confirming itself. Here it points at a contract instead of a revenue print.
The asymmetry keeps all of this invisible. The level of usage and the velocity of usage sit on every dashboard in the company. The change in the velocity sits on almost none of them, and it is the number the commit’s sizing depends on.
Is Your Commit Architecture Built for the Pace You Assumed?
When growth decelerates, commits sized to a faster trajectory become anchors. A few diagnostic questions reveal whether your licensing, packaging, and pricing decisions can survive a slowing curve before renewal pressure exposes it.
Every Headline Number Still Prints a Record
Committed revenue is recognized on the commitment, so during exactly the quarters an account’s growth is undershooting the assumption, the vendor’s own reporting is at its best: record revenue, record consumption, healthy expansion. A renewal cohort can be accumulating a shortfall underneath a record year. The record is real. It is just silent on the assumption.
The volume-discounts piece made the schedule-level version of this argument: when the discount schedule has gone stale, record revenue tells you the quarter went well and tells you nothing about whether the schedule still fits. The commit version of that gap runs longer and ends harder. It opens the quarter growth first lands under the assumed rate, and it closes at the renewal where someone finally says the shortfall out loud. Renewal is the first scheduled moment anyone is contractually required to look.
Commits are not signed one at a time; they are signed in seasons. A book of commitments sized in the same exuberant quarter turns over in the same later quarter, and what we observe across our pattern library is that the misses arrive as a cohort, not as one awkward account. The sizing errors were correlated from the day the contracts were signed.
What the Renewal Conversation Sounds Like
The miss surfaces at renewal, and it surfaces as an argument about the number. The number that closed the deal becomes the number the buyer argues down the first term it stops climbing the way everyone assumed. That is not bad faith. It is what an unpriced assumption sounds like when it resolves: nobody priced the possibility that growth would land below the presumed rate, so the resolution happens in negotiation, where unpriced things are settled.
The market’s own promise sharpens the point. Consumption structures are sold on frictionless expansion: usage and revenue grow without anyone renegotiating the deal. The promise is true in the direction it means, which is up. In the other direction, deceleration is a renegotiation nobody scheduled.
One direction is designed. The other is where the disputes live
The overrun direction has a published instrument list: usage bands, overage terms, true-up provisions. A structure with one designed direction and one undesigned direction will always produce its disputes in the undesigned one.
Both directions are failures of the same embedded forecast. Overrun means the account grew past what the sizing presumed; deceleration means it grew short of it. One forecast, two ways to miss, one of them designed for. If your commitments carry considered answers for overrun and silence for deceleration, that asymmetry is the conversation to have with a pricing expert before the renewal cohort arrives, not after.
The frame for the fix: the miss should land in a structure that anticipated it. At the schedule level, the volume-discounts article owns that idea as the true-down with a defined path back onto the pricebook. At the architecture level, a pricing surface gives a commitment that needs to change somewhere defined to move. What either structure should say for a given vendor, which misses it absorbs and which it reprices, is design judgment, and precisely the decision most commit structures never made.
The Questions to Ask Before the Next Commit Is Sized
Four questions for your own pricing and revenue team, askable today, against your own book.
Ask what growth rate the sizing assumed, and whether the account is growing slower than that rate right now. If nobody can name the rate, the assumption exists anyway, unexamined, and it resolves in negotiation.
Ask which number in your own reporting would surface the gap mid-term, and who reads it. If the answer is commit utilization, reviewed at renewal, then the answer is nobody, too late.
Ask what the contract says happens when growth lands below the assumption. Most contracts answer the overrun direction in detail and this direction not at all. Whatever the paper leaves unsaid, the renewal meeting will improvise.
Ask whether the commit’s unit is bounded and stable. A commitment denominated in a surrogate unit whose meaning can move is carrying a second assumption underneath the first, and the two compound.
If you cannot name the growth rate your commitments assumed, start there. Describe your commit structure and where renewals are landing through our talk to an expert form, and a pricing expert will read it and reply.
