True ROAS, and how to get to it

If you add up what every ad platform claims it earned you, the total will exceed your revenue. That is not fraud — it is four systems each counting a conversion they can plausibly claim, none of them subtracting a refund.

Why the numbers diverge

Platform-reported ROAS is the platform's best case: its own attribution window, its own modelled conversions, gross of refunds. Two platforms can both count the same order. Your analytics tool applies a different rule again, over only the sessions it observed. Finance counts cash. All four are internally consistent and mutually irreconcilable.

The adjustments that matter

• De-duplicate across platforms. One order, one owner. Decide the rule and apply it in one place.
• Subtract refunds and chargebacks against the cohort that generated them, not the month they landed in.
• Handle failed and cancelled renewals — subscription revenue that never arrives still sits in most ROAS reports.
• Add revenue the pixel never saw: offline closes, phone orders, anything finished outside the browser.
• Use margin, not top-line, once the above is stable and you want to compare across products.

Blended or channel-level?

Blended ROAS — all revenue over all spend — is honest and nearly useless for allocation: it cannot tell you which channel to cut. Channel-level ROAS is useful and easy to fake, because every platform overstates its own. Run both: blended as the reality check, channel-level for decisions, and investigate whenever the two move in opposite directions.

When the number is trustworthy

A true ROAS number is ready to spend against when it reconciles to your payment processor for the same period, when the cohort has had time to mature, and when you can trace any single order in it back to a click. Until all three hold, treat it as directional.

Questions people ask

Read next