Advertising judged on paid, retained subscribers.

Paid activation is the target, not signup

You define activation, first invoice and retained subscriber as distinct outcomes, and each channel is told which one it is working toward. The business metrics service computes cost per outcome from CRM stage records and collected revenue, so a channel that fills the trial list with accounts that never activate stops looking cheap.

Four truth layers, never averaged together

Platform-reported ROAS, independent journey attribution, CRM account quality and collected revenue stay four separate numbers. When a network reports more conversions than your invoice ledger records, the gap is raised as a discrepancy with both figures shown, rather than blended into a single score that hides it.

Churn-prone segments get suppressed at source

Suppression lists are built from account state: refunded subscriptions, trials that never activated, plans that cancel before the second renewal. Those lists are pushed to Google, Meta, LinkedIn and Microsoft as exclusions, so the networks keep running their own targeting and simply stop being rewarded for lookalikes of accounts you do not want more of.

Holdouts for the questions attribution cannot answer

Attribution distributes credit under the rule you selected. It cannot tell you whether a trial would have started anyway, so brand and retargeting campaigns are tested with geo holdouts and paused cells that have a defined readout date. The experiment result and the attributed result are reported side by side, never merged.