Stop optimizing creators on CPA. Start optimizing on cohort.
A creator with a slightly higher CPA and dramatically better cohorts is the right pick. Here's how to set up the reporting that lets you see it.
Creator-program reporting is dominated by CPA. It is easy to compute, easy to compare against paid-acquisition channels, and easy to put in a deck. It is also the wrong number to optimize on for creator commerce.
What CPA misses
Creators don't drive first orders the way a paid-search ad does. They drive cohorts — groups of buyers who came in through a creator's audience, and whose behavior over the next 30, 60, 90 days is qualitatively different from buyers acquired through other channels.
A creator with a CPA of €38 might produce buyers with a 31% repeat rate at 60 days. A paid- search keyword with a CPA of €22 might produce buyers with a 9% repeat rate at 60 days. The first is dramatically more valuable per acquired buyer. The second wins the CPA column.
The reporting that exposes it
A serious Creator Commerce OS reports, per cluster, per campaign, and per creator:
- First-order CPA — the headline performance number.
- 30-day repeat rate — early signal on cohort quality.
- 60-day GMV per acquired buyer — the number worth optimizing on for D2C.
- 90-day cross-category lift — buyers who became multi-category after the first order.
Together those numbers expose cohort quality as a per-cluster property. They are also what unlocks rev share as a compensation model — once you know which clusters produce better cohorts, paying them more on rev share is obvious math.
How to roll this out
- Tag every creator order with creator ID, cluster ID, and campaign ID.
- Join order events forward at 30/60/90 days against the original creator ID.
- Report at the cluster level, not the creator level — individual creators have noisy cohort data.
- Use the cluster signal to drive sourcing for the next campaign in the same vertical.
This is not a small lift if you're stitching it together yourself. It is the default report in a Creator Commerce OS — see also our attribution architecture whitepaper.
What changes for the team
The conversation moves up. Instead of "the campaign hit our CPA target, ship more like it," the team is asking "which clusters produce 60-day GMV-per-buyer above €X, and how do we double the spend in those clusters next quarter."
That is the conversation a CFO will fund.
Related field notes.
Most attribution tags are 30–80KB. cyrqle's is gated at sub-2KB gzipped. Here's why that gate matters, how we enforce it, and what we left out.
Third-party cookies are gone or going. The creator-commerce attribution stack that survives is first-party — tag, QR, and promo, reconciled against order data.
Follower count is a vanity input that still drives most influencer-marketing tools. Cluster signal — audience behavior, not creator self-description — is the only input that maps to attributable revenue.