12 Jun 2026 · 9 min read

Measuring social without lying to yourself

Last-click attribution is a story your dashboard tells to make you feel better. Here is the model we use instead.

Every platform's attribution window is designed by the platform, for the platform. Meta will happily claim a sale that TikTok also claims, and your finance team will notice before you do.

We run three layers. Platform-reported numbers for optimisation, because the algorithms need them. A first-party model for reporting, built on a 90-day cohort of people who engaged, not clicked. And a quarterly incrementality test, geo-split, to keep both honest.

The number the board sees is the middle one. It is always smaller than the platform number and always larger than last-click. It is also the only one that has ever predicted next quarter's revenue.

The hardest part is not the maths. It is agreeing, before the campaign, what counts. We write that down in the first week and we do not change it when the results come in.

More playbooks