Most large buyers do not run one Claude workload. They run several. A support assistant, a document pipeline, an internal coding tool, a customer facing feature, and a handful of experiments that may or may not survive the year. Each of these has its own usage pattern, its own growth curve, and its own tolerance for cost. When you put them all under one Anthropic agreement, the math stops being a single forecast and becomes a portfolio problem. Treated well, that portfolio is your single biggest source of negotiating leverage. Treated badly, it is the reason your commit is wrong in both directions at once.
This article walks through how to model a portfolio commit, how the products offset each other, and how to use the aggregate to win a better rate than any single product could earn on its own.
Aggregate the volume, separate the behavior
The first instinct of most finance teams is to forecast each product, sum the numbers, and commit to the total. That gets the volume right and the behavior wrong. Products do not all peak in the same month. A retail facing feature spikes in the fourth quarter. An internal coding tool dips when engineers are on holiday. A batch document pipeline runs heavy at month end and quiet in between. When you sum naive monthly forecasts you tend to overstate the peak, because you are stacking peaks that never actually land on the same day.
The correct approach is to model each product separately, then combine them at the portfolio level and look at the combined monthly curve. The portfolio curve is almost always smoother than any single product, because the peaks and troughs partially cancel. That smoothing is real money. A smoother curve means you can commit closer to the true average rather than padding for a peak that the portfolio as a whole never reaches.
A portfolio commit is not the sum of the product commits. It is the sum of the curves, and the curves cancel.
Use the portfolio to earn a higher band
Anthropic prices committed spend in bands. The rate improves as you move from the smaller commitments into the larger ones, and the jumps between bands are meaningful. A single product that spends in the lower band on its own may be sitting just under a threshold that would unlock a materially better rate. Combine three such products under one agreement and you cross the threshold. Every product then buys tokens at the better rate, including the small ones that could never have earned it alone.
This is the core portfolio play. You are using the volume of the whole to lift the rate of every part. We have seen this single move take a meaningful percentage off the blended cost of a portfolio without changing a line of code. The work is in the modeling, in proving the aggregate volume credibly, and in making sure the agreement applies the negotiated rate across every product rather than ring fencing it to one.
Watch for ring fencing in the contract language
When you consolidate, read the agreement carefully for language that limits the discounted rate to a named product, a named workload, or a named business unit. That language quietly defeats the entire portfolio play. You want the committed rate to apply to all qualifying Claude usage under the agreement, across products, across teams, and across regions. If a vendor wants to scope the rate narrowly, that is a negotiation point, not a given.
Model the offset, then size the floor
Once you have the combined curve, the commit floor should sit near the portfolio low case, not the sum of the individual low cases. Because the products offset, the portfolio almost never falls as low as every product hitting its floor in the same month. Sizing the floor too high is how buyers end up paying for committed tokens they did not consume. The unused commitment usually disappears at the end of the period rather than rolling forward, so an inflated floor is simply money handed to the vendor for nothing.
On the upside, the same rate protection rules apply as for a single product, only they matter more. With several growing products under one agreement, the chance that the aggregate overshoots the commit is high. Negotiate overage at the committed rate so growth across the portfolio is billed at your discount rather than at list. For a portfolio, the gap between those two outcomes can run into seven figures over a multi year term.
Decide what belongs in the commit and what does not
Not every workload should be inside the commitment. Experiments that may be cancelled, products with wildly uncertain futures, and one off projects can distort the floor if you fold them in. A clean approach is to commit confidently to the stable core of the portfolio, the products you know will run all year, and to keep the speculative work outside the floor where it is billed at the same committed rate but does not raise your minimum. You get the rate without taking on the risk of promising volume you may never deliver.
This is where a portfolio view earns its keep. You are not making one bet, you are making a managed set of bets, and you can choose which ones carry a floor and which ones simply ride the rate. That flexibility is hard for a single product owner to see and easy to see when you model the whole.
Govern the portfolio after you sign
A portfolio commit is not a set and forget number. The whole reason it is efficient is that the products offset, and offsets drift. A product you expected to stay flat takes off. A product you expected to grow gets cancelled. If you do not watch the blend, the commit you negotiated for one portfolio shape ends up serving a different one. Run a monthly review that rolls every product into one cost view, compares it to the committed curve, and flags divergence early. Negotiate a reforecast right so that if the blend moves past an agreed threshold you can adjust the commit rather than carry a wrong number to renewal.
Underneath all of this, the same optimization discipline applies that we run on every Claude workload. Route work across Opus, Sonnet, and Haiku so the expensive model only handles what truly needs it, which on its own typically moves aggregate spend 40 to 70 percent. Cache the repeated context that multiple products share, which can take up to 90 percent off that portion. Push anything that does not need a real time answer into batch at half the cost. A portfolio gives you more places to apply each lever, which is why portfolio buyers tend to see the largest absolute savings of anyone we work with.
The headline is straightforward. Model the curves, not the totals. Use the aggregate to lift the rate on every part. Floor the stable core, ride the rate on the rest, and govern the blend after you sign. That is portfolio commit math done on the buyer side, and it is exactly the work we sit between you and Anthropic to run.
A worked portfolio, in numbers
Consider four workloads under one agreement. A support assistant that runs steadily all year. A document pipeline that runs heavy at month end and quiet between. A retail facing feature that triples in the fourth quarter and is modest the rest of the year. And an internal coding tool that dips every time engineering takes leave. If you forecast each one's peak month and sum the peaks, you get a number that no single month of the portfolio ever reaches, because the document pipeline's month end peak does not land on the retail feature's fourth quarter peak, and neither lands during the coding tool's holiday dip.
Model them together and the combined curve is visibly flatter than the sum of the individual peaks. That flatness is the portfolio's gift. It means the true average sits well below the summed peak, and the commit floor can sit near that average rather than near the inflated peak. Buyers who skip this step routinely commit to ten or twenty percent more than the portfolio ever consumes in a single month, and that excess is money handed over for nothing.
The consolidation conversation with Anthropic
Bringing several products under one agreement is itself a negotiation, and it is one the vendor has reason to welcome, because a consolidated account is a larger and stickier account. Use that. The pitch to the account team is that you are willing to commit the combined volume of the portfolio in exchange for a rate that reflects the band that combined volume reaches. You are offering them a bigger, cleaner relationship, and asking, in return, for the rate that volume deserves and for that rate to apply across every product.
The points to win in that conversation are specific. The committed rate must apply to all qualifying usage under the agreement, not a single named workload. Overage must bill at the committed rate. The treatment of unused commitment must be clear, ideally with some carryover rather than a hard forfeiture at period end. And you want a reforecast right so that as the blend shifts you can adjust rather than carry a stale number to renewal. Each of these is ordinary in a well negotiated portfolio agreement and absent in a poorly negotiated one.
Governing the blend so the offset holds
The efficiency of a portfolio commit depends on the offset between products holding roughly as modeled. Offsets drift. A product you floored as stable surges. A product you expected to grow gets cancelled and its volume vanishes. If you do not watch the blend, you can find the portfolio drifting toward a synchronized peak you never planned for, or collapsing below the floor you committed.
Run a single monthly cost review that rolls every product into one view against the committed curve. Watch for two things. First, divergence in the aggregate, which tells you whether the total is tracking the commit. Second, change in the shape, which tells you whether the offsets that justified your floor still exist. When either moves past the threshold you negotiated a reforecast right against, use that right. A portfolio commit is a managed instrument, not a number you set once and forget, and the management is where the modeled savings either materialize or quietly leak away.
Why portfolios are where the biggest savings live
Across the engagements we run, portfolio buyers tend to capture the largest absolute savings, for a simple reason. Every lever applies in more places. Model routing across Opus, Sonnet, and Haiku can be tuned per product, so the cheap models carry more of the aggregate. Caching benefits multiply when several products share repeated context such as a common system prompt or a shared knowledge base. Batch processing can be applied across every asynchronous workload at once rather than one at a time. And the consolidated volume lifts the rate underneath all of it. A single product owner sees a slice of this. A portfolio owner sees the whole board, and the whole board is where the real money is.
Deciding the floor product by product
A portfolio commit does not have to treat every product the same way, and the best ones do not. The disciplined approach is to decide, product by product, how much of each one belongs inside the committed floor. A product that has run steadily for a year and shows no sign of stopping can carry most of its volume in the floor with confidence. A product that launched last quarter and is still finding its audience should carry only its proven base, with the upside riding the committed rate rather than raising the minimum. An experiment that might be cancelled should carry nothing in the floor at all, while still benefiting from the negotiated rate on whatever it happens to consume.
Sorting the portfolio this way protects you from the most common portfolio mistake, which is letting the optimism around a few new products inflate the floor for the whole. The floor should be built from volume you are confident will materialize, drawn mostly from the stable core. Everything speculative rides the rate without raising the minimum. Done well, this gives you the rate of a large committed buyer with the downside risk of a much smaller one, which is precisely the asymmetry a buyer side advisor is trying to engineer.
The renewal advantage a portfolio creates
A consolidated portfolio also changes the renewal in your favor. A single product on a single agreement arrives at renewal as a small, replaceable line item that the vendor can afford to push on price. A portfolio of products under one agreement arrives as a large, strategic relationship that the vendor has every reason to protect. That difference in posture is worth real money at renewal, because an account team will work much harder to keep a consolidated portfolio than to retain a single workload. You carry more weight simply by virtue of having brought the volume together, and that weight is leverage you can use the next time the contract opens.
This article is part of our Token Optimization Playbook. Read it for the full buyer side method behind everything above.