A committed spend number is a bet on your future demand, and most buyers place that bet by guessing. Guess high and you strand money in a commitment you cannot use. Guess low and you pay overage on top. Here is the buyer side method for sizing an Anthropic commitment from optimized, measured demand instead of a hopeful forecast, so the number you sign is the number you actually need.
A committed spend agreement is a bargain: you promise Anthropic a level of spend over a term, and in exchange you get a better rate than pay as you go. The bargain is good when the commitment matches your real demand and bad when it does not, and the entire risk sits in that match. Most buyers size the commitment by guessing, taking a recent bill, applying a growth assumption, and signing a round number that feels right. That guess is where the money is lost, because a commitment is a floor you pay regardless of usage, so guessing high strands the difference in unused commitment, and guessing low sends your overflow into overage at rates that erode the discount you committed to get. Neither failure is necessary, because demand can be measured and shaped rather than guessed.
The buyer side method replaces the guess with a number built from evidence. It rests on a single principle that reorders the whole process: optimize demand first, then size the commitment to the optimized number, never the reverse. A commitment sized against unoptimized demand bakes every inefficiency you have not yet fixed into a number you are contractually bound to for the length of the term. Fix the inefficiencies first and you commit to a leaner, truer figure. The sequence is the difference between committing to what you actually need and committing to what you happen to spend today.
The reason sequence matters so much is that the optimization levers move demand by a large margin, and a commitment signed before they are applied locks the pre optimization volume into the contract. Model routing across Opus, Sonnet, and Haiku, prompt caching at up to ninety percent on repeated input, and batch processing at roughly half the rate on asynchronous work together typically cut aggregate spend by forty to seventy percent against a baseline of uniform top tier usage. A buyer who commits before pulling those levers commits to a number that the optimized deployment will never reach, and then either pays for unused commitment or, worse, has locked a high baseline that the savings cannot escape. A buyer who optimizes first sees the true floor of their demand and commits to that.
This is why we treat token optimization and commitment sizing as one connected exercise rather than two. The optimization sets the real cost base. The commitment is sized to that base. Done in the wrong order, the optimization saves money on the margin while the commitment overpays on the whole, and the net result is far worse than it should be. Done in the right order, the saving is captured inside the commitment rather than stranded outside it.
Once demand is optimized, sizing becomes an evidence exercise rather than a guess. The inputs are knowable and the method is straightforward.
The key judgement is the split between the stable base and the variable peak. Your base demand, the volume you are confident you will use regardless of how the year goes, is what you commit to, because committing to it is pure upside: you get the better rate on volume you were always going to consume. Your variable peak, the volume that depends on uncertain growth or seasonal swing, is what you should be cautious about committing, because committing to a peak you may not hit is how unused commitment is created. The instinct to commit high to get a better rate is exactly backwards when the high number is uncertain, because the rate saving on volume you do not use is dwarfed by the cost of the commitment you cannot meet.
Sizing the number correctly is necessary but not sufficient, because the terms around the commitment determine what happens when reality diverges from the plan, and reality always diverges somewhat. Two terms matter most. The overage rate governs what you pay for volume above the commitment, and a buyer who has sized conservatively to the base will run some volume into overage by design, so the overage rate should be negotiated, ideally to the committed rate rather than a punitive premium, so that exceeding the commit is not penalized. The treatment of unused commitment governs what happens if you fall short, and the goal is to avoid terms where unused commitment simply vanishes at period end with no ability to roll it forward or otherwise recover the value.
Negotiated well, these two terms turn a right sized commitment into a resilient one. You commit to your confident base, you run your variable volume into overage at a fair rate, and you are protected against losing value if you fall short. Negotiated poorly, even a perfectly sized number can be undermined by punitive overage that punishes the conservative sizing or by unused commitment terms that destroy the value of any shortfall. The number and the terms are one decision, and sizing the number without negotiating the terms leaves the commitment exposed.
We size commitments for a living and we do it in the right order. We optimize the demand first, applying routing, caching, and batch to find the true cost base, then build the commitment from measured, optimized demand split into its confident base and variable peak, and then negotiate the overage rate and unused commitment treatment so the right sized number is protected against the divergence that always comes. We sit between you and Anthropic and study nothing else, so we know how the commit bands, overage rates, and unused commitment terms actually work in these agreements and where comparable buyers have secured protection. The playbook below lays out the full method, the optimization levers, the base and peak split, and the terms that protect the number. Download it and stop sizing your commitment with a guess.
Buyers consistently underestimate the cost of committing high, because the logic of a volume discount pulls in the wrong direction. The bigger the commitment, the better the rate, so it feels prudent to commit generously and lock in the best price. The flaw is that the discount applies only to volume you actually use, while the commitment applies to volume whether you use it or not. Commit to a number above your real demand and you pay for the gap at full freight, and the rate saving on the volume you do use rarely covers the cost of the commitment you do not. The arithmetic almost always favors committing to the confident base and running the rest into overage, even at a slightly higher per unit rate, because paying a small premium on the volume you actually consume beats paying the full rate on volume you never touch. Committing high to chase a better rate is one of the most common and most expensive sizing mistakes buyers make.
The corollary is that the right commitment often feels uncomfortably conservative to a buyer used to thinking of a bigger commit as a better deal. Sizing to the confident base means deliberately leaving your uncertain growth outside the commitment, which feels like leaving the discount on the table, when in fact it is avoiding a liability. Reframing the commitment as a liability you take on rather than a discount you win is what corrects the instinct, because a liability is something you size carefully to what you can be sure of, not something you maximize.
A commitment is sized under uncertainty, and that uncertainty resolves over the term as real demand reveals itself, so the sizing exercise should not end at signature. A buyer who has sized conservatively to the base and watched demand grow into and beyond it has earned the evidence to commit more confidently at the next opportunity, while a buyer who finds demand softer than expected has learned to hold the line rather than grow the commit. The agreement should ideally leave room to revisit the commitment as this evidence accumulates, through expansion terms, true ups, or simply a term length short enough that resizing happens regularly. A commitment locked for a long term with no ability to adjust forces you to guess once and live with it, which is precisely the guessing this method exists to avoid.
This is why term length and adjustment terms are part of commit sizing rather than separate from it. A shorter term or a built in resizing mechanism lets you commit conservatively now and grow the commitment as the evidence justifies, which is strictly better than committing high under uncertainty and hoping demand catches up. We size the commitment to optimized, measured demand, negotiate the overage and unused commitment terms that protect it, and structure the term so the number can track demand as the uncertainty resolves, so that the commitment stays right sized across the whole relationship and not just on the day it is signed.
Download the playbook for the buyer side method that sizes an Anthropic commitment from optimized demand and protects you on overage and unused commitment.
Download the playbookWeekly intelligence on Anthropic pricing moves and the buyer side counters that work.