Cohort economics: the only affiliate metric that predicts next year

Almost every affiliate reports performance by calendar month, and almost every affiliate is misled by it. A monthly figure blends new players who have just deposited with cohorts acquired two years ago, and it hides the single most important question in rev share: how quickly does a programme's player value decay?

Two programmes can produce identical revenue this month while being worth entirely different amounts. One retains 40% of first-month value at month twelve; the other retains 9%. Under a rev share deal you are effectively buying an annuity, and the decay rate is the price of that annuity. Judging it by this month's total is like valuing a bond by today's coupon.

Building cohort curves without full operator cooperation is more feasible than most affiliates assume. Tag every click with a month-stamped sub-ID, insist on sub-ID passthrough in the reporting, and you can reconstruct the curve from your own statement data even when the programme only reports aggregate revenue. It takes three or four months before the shape becomes readable and about nine before it becomes reliable.

Read the curve in three places. Month one to two tells you about bonus abuse and onboarding quality: a collapse here usually means the welcome offer is attracting the wrong players rather than that your traffic is poor. Month three to six tells you about product and reactivation. Month nine onward tells you about the operator's CRM competence, which is the part you cannot influence at all.

The comparison that matters is not curve against curve but curve against deal structure. A steep decay is entirely acceptable under a strong CPA; it is ruinous under a rev share you negotiated on the assumption of longevity. Conversely a flat curve makes a modest rev share far more valuable than a headline CPA, and it justifies patience through a weak quarter.

One practical warning. Cohort curves will occasionally show a step-change that has nothing to do with players — a tracking change, a deduplication rule, a silent attribution window adjustment. Any discontinuity that appears across all your cohorts on the same date is an infrastructure event, not a behavioural one, and it should trigger a conversation with the programme immediately rather than a strategy change on your side.