Take rate compresses between onboarding and the third year, concentrated almost entirely in the largest customer decile. Blended take rate can hold steady while the mix shifts underneath it — which is why the blended number is the wrong one to diligence.
What is a take rate?
The share of transaction value a payments or marketplace business keeps as revenue. It is the most load-bearing assumption in most fintech models and the one most frequently extrapolated flat — as net revenue retention is in software ones.
Definitions vary enough between businesses that cross-company comparison requires normalizing, and the normalization is rarely shown.
The definitional variance is worth pinning down before anything else. Whether interchange, scheme fees and payment-processing costs sit inside the take rate or outside it changes the number by more than most of the operational differences between businesses, and different companies make that choice differently and legitimately. A comparison run without normalizing it is comparing accounting conventions.
When does take rate compress?
When customers get large enough to negotiate, which is a function of their volume rather than of tenure. A merchant that doubles its throughput acquires leverage it did not previously have and uses it at renewal.
Compression is therefore a success problem. A business growing into larger customers will see take rate fall even while every operational metric improves.
Compression also arrives in steps rather than smoothly, because it is triggered by renewal dates. A business can hold its blended rate for six quarters and then move sharply when three large contracts come up in the same period, which makes a smooth downward extrapolation wrong in both directions — too pessimistic in the quiet periods and far too optimistic across a renewal cluster.
Which customers renegotiate first?
Those with an alternative and enough volume to make switching worth someone's time. Negotiating leverage predicts compression far better than tenure or satisfaction do.
The corollary is that a long tail of small merchants is more durable on price than the headline customer list suggests, which cuts both ways in a valuation and is testable with a pricing study on the tail.
The tail also behaves differently on churn. Small merchants leave more often and negotiate less, so the tail is durable on price and fragile on retention, while the large accounts are the reverse. A model that applies one set of assumptions across the base will misstate both, and the direction of the error depends entirely on where growth is coming from.
Where does the floor sit?
At the point where the business's cost to serve plus its required margin meets what the customer can get elsewhere. That is knowable from the competitive set rather than from the company's own history.
Assets that have not yet met a well-capitalized competitor in their segment have not yet found their floor, which is a diligence question rather than a modeling one.
There is a second floor worth establishing, which is the customer's own economics. Below some level of take rate the service stops being worth providing at the quality the customer expects, and the negotiation shifts from price to scope. Establishing where that sits is a question for the operators who have run the service rather than for the commercial team.
How do you test it?
Ask the customers, particularly the large ones, what they pay and what they have been offered. Cohort data from the company tells you what happened; customers tell you what happens at the next renewal.
Adjacent providers are the second source. They know what they have quoted into the same accounts, which is the clearest available read on where the floor actually is.
Cohort data and customer conversations answer different questions and should be read together rather than in sequence. The cohorts tell you what compression has already happened and to whom; the customers tell you what is about to happen at their next renewal. A diligence using only the first is describing history, and one using only the second has no baseline to place it against.