How Product Growth Changes Your Commit

Most enterprises sign an Anthropic commitment based on a snapshot. They take last month of usage, multiply by twelve, add a little headroom, and call it a commit. That works when usage is flat. It falls apart the moment the product behind the spend starts to grow, because token consumption does not move in a straight line. It moves with adoption, with feature launches, and with how heavily each active user leans on the model. When you sign a commit you are not betting on today, you are betting on a curve you have not drawn yet.

This article is about reading that curve before you put a number in a contract. The goal is simple. You want a commit that earns you the best possible rate without exposing you to a shortfall on the downside or an overage cliff on the upside. Both failures cost real money, and both are avoidable with a little discipline up front.

Why a flat snapshot lies to you

Token spend on a growing product is driven by three multipliers stacked on top of each other. The first is active users. The second is sessions or calls per active user. The third is tokens per call, which itself splits into input tokens and the much more expensive output tokens. When a product is growing, all three of these tend to rise at the same time. More people show up, the people who are there use it more, and the experiences you ship to keep them tend to get richer and longer.

Because the multipliers compound, a product that doubles its user base can easily triple or quadruple its token bill in the same period. If you sized your commit off a flat month and signed a twelve month term, you can blow through the commitment in the first two quarters and spend the back half of the year paying overage at a rate you never negotiated. That is the single most common way a good rate turns into a bad deal.

A commit is a bet on a curve, not a snapshot. Size it for the shape of the curve, not the height of today.

Draw the curve in three scenarios

You do not need a perfect forecast. You need three honest ones. Build a low, a base, and a high scenario for the next twelve to twenty four months. The low case assumes adoption stalls and you ship nothing new. The base case assumes your current roadmap lands roughly on time. The high case assumes a launch lands well and adoption accelerates. For each, project the three multipliers month by month and convert to tokens, then to dollars at current model rates.

What you are looking for is the spread. If the high case is three times the low case, a single fixed commit number cannot serve you well. It will either be too small for the high case or too large for the low case. That spread is the signal that you should be negotiating a ramped or tiered commitment rather than one flat annual figure.

What the three multipliers actually tell you

Active users is the number most leaders watch, but it is the least useful for sizing a commit on its own. Tokens per call is where the cost hides. A product that adds retrieval context, longer system prompts, or chained calls can double its tokens per interaction without adding a single user. When you model growth, model the richness of the experience separately from the size of the audience. Teams that only track headcount of users routinely undershoot because their per call cost crept up while they were not looking.

Match the commit shape to the growth shape

Once you can see the curve, you can ask Anthropic for a structure that fits it. There are three shapes worth knowing.

The first is a ramped commitment. Instead of one flat annual number, you commit to a schedule that steps up over the term. You pay for less in the early quarters when your product is still climbing, and you commit to more later when the curve says you will be there. This protects you from overcommitting in month one and from the overage cliff in month nine.

The second is a tiered rate. You negotiate a unit price that improves as you cross volume thresholds during the term. This rewards the growth you expect without forcing you to promise all of it up front. The key term to win here is that the better rate applies to all volume in the tier, not just the marginal tokens above the threshold.

The third is overage at the committed rate. This is the quiet protection that matters most for a growing product. If you exceed your commitment, you want the extra usage billed at the same discounted rate you negotiated, not at list price. Without it, growth past your commit is the most expensive token you will ever buy. With it, a high case stops being a punishment.

Build the buffer in the right direction

Buffers are usually framed as protection against running short. For a growing product the more important buffer is protection against being penalized for success. You want the floor of your commit set near your low case so a stall does not leave you paying for tokens you never used, and you want the rate protection extended well above your high case so a great quarter does not cost you list price. The asymmetry matters. A product that grows fast is far more likely to overshoot than to undershoot.

One more piece of discipline. Tie your commit to a reforecast checkpoint. Negotiate the right to revisit the commitment at a defined point in the term if usage diverges from plan by more than an agreed percentage. Anthropic account teams will often grant this because it lets them grow with you rather than lose the relationship at renewal. It turns a static bet into a managed one.

Where the savings come from

None of this works if your underlying spend is inflated. Before you commit a single dollar you should reduce the tokens you are buying. Model routing across Opus, Sonnet, and Haiku typically cuts aggregate spend 40 to 70 percent versus running everything on Opus. Prompt caching can take up to 90 percent off the cost of the repeated context that growing products lean on heavily. Batch processing takes 50 percent off any workload that does not need an answer in real time. Optimize first, then commit to the lower, cleaner number. Committing to inflated usage just locks in waste for the length of the term.

This is the buyer side order of operations we run on every engagement. Cut the spend, draw the curve, then size a commit that fits the shape of the growth rather than the height of a single month. Done in that order, growth becomes leverage instead of a liability.

A worked example of the curve bending

Picture a customer facing assistant that launches to a single business unit. At launch it serves two thousand active users, each running about ten interactions a day, at roughly four thousand tokens an interaction once you count the system prompt, the retrieved context, and the response. That is a manageable number, and the team sizes a commit around it with a comfortable feeling of headroom.

Now run the curve forward six months. The assistant rolls out to three more business units, so active users reach eight thousand. The experience got better, so interactions per user climb from ten to fourteen. And the team added richer retrieval, so tokens per interaction rise from four thousand to six thousand. None of these jumps is dramatic on its own. Together they take monthly token consumption up by a factor of roughly six. The commit that felt generous at launch is now exhausted before the third quarter, and every token after that is overage.

This is not a worst case. It is an ordinary case for a product that succeeds. The lesson is that success, not failure, is the scenario most likely to break a naively sized commit. You should size for the version of the future where the product works, because that is the version you are building toward.

The two failures a growth aware commit avoids

There are exactly two ways a commit goes wrong, and a growth aware approach is designed to avoid both. The first failure is the shortfall, where you committed to more than you used and the unused portion simply evaporates at the end of the period. This is the failure that catches buyers who padded a flat forecast out of caution. The second failure is the overage cliff, where you committed to less than you used and the excess is billed at list price rather than your negotiated rate. This is the failure that catches buyers whose product grew faster than the snapshot suggested.

A ramped commit with overage at the committed rate neutralizes both. The ramp keeps the early floor low so a slow start does not strand committed dollars. The committed overage rate means that if growth runs ahead of the ramp, the extra usage is still billed at your discount. You stop choosing between two bad outcomes and instead build a structure that survives whichever way the curve breaks.

How to defend the forecast to Anthropic

An account team will push back on a commit that ramps slowly, because a flat high number serves the vendor better. Your defense is the data. When you can show the three multipliers, the historical trend in each, and the scenario range built from them, you are negotiating from evidence rather than caution. A vendor can dismiss a buyer who simply asks for a lower floor. It is much harder to dismiss a buyer who arrives with a month by month model and says, here is the curve, here is why the floor belongs where I have put it, and here is the volume you capture as the ramp climbs.

That framing also gives the account team something to say internally. You are not asking them to give you a smaller deal. You are showing them a credible path to a larger one as your product grows, with a structure that protects you on the way up. Aligning your forecast with their incentive to grow the account is often what unlocks the ramp and the committed overage rate together.

Review the commit on a fixed cadence

A commit sized for a growing product should never be left alone for twelve months. Set a fixed cadence, monthly at the line level and quarterly at the commitment level, where you compare actual consumption against the ramp you committed to. The monthly view catches drift in the three multipliers early, while there is still time to act. The quarterly view is where you decide whether to invoke a reforecast right, accelerate the ramp, or hold. Buyers who review on a cadence almost never get surprised at renewal. Buyers who set the commit and look away almost always do, because by the time the annual invoice tells the story, the year is already over and the leverage is gone.