How we decide what a marketing channel actually earned

Ask two advertising platforms what they earned you last month and they will happily tell you. Add the answers up and you will often find they earned you more than the shop actually took. Nobody is lying. Each one is counting every order it can plausibly claim, and the same order is plausible to more than one of them.

That leaves you with a choice most reporting quietly makes for you. Either you believe one platform and ignore the rest, or you believe the shop and give the channels nothing. Both answers are wrong in a way that costs money, because both of them decide a budget.

Three numbers, not one

We report every channel as three figures, and we never collapse them into one.

The floor is what we can prove. Every order carries a customer journey, and the journey names the last click before the purchase. One order, one channel, no double counting. It undercounts, because tracking expires, people switch devices, and plenty of orders arrive with nothing attached at all. But nothing in it is invented.

The claim is what the platform says it earned. It is the ceiling, not the answer. A claim is worth reading because the platform can see things we cannot, such as a view that led to a search a week later. It is worth doubting because the platform is also the party being judged.

The estimate sits between them. It is the midpoint of the band, and it is paid for out of the pot of revenue that no channel could claim. That last part is the whole trick: a channel can never be granted more than the pot physically holds, however loudly it claims. So the estimates always add up to the revenue the shop actually took.

Why the pot matters

Without the pot, every reconciliation is just an argument about weightings. With it, the arithmetic is closed. If two platforms both want credit for the same order, they are competing for the same unclaimed revenue, and there is only so much of it. When the pot is too small to satisfy the claims, we scale the uplifts down rather than pretending the revenue exists.

The pot has a name in the interface: unknown. It is not a failure state. On most accounts it is the largest single bucket, and shrinking it is a tracking project, not a reporting one.

What this changes on a Monday morning

The point of all this is not tidiness. It is that budget decisions stop being decided by whichever platform shouts loudest.

A channel whose floor already clears your target does not need a debate. A channel whose claim clears the target but whose floor is nowhere near it needs a test, not a budget increase. And a channel that looks enormous in one dashboard and invisible in the journey data is usually a tracking problem sitting in front of a real result.

We have found organic social outperforming paid social by several multiples this way, on an account where the ads dashboard was the only report anyone had been reading. The founder's own link in bio was carrying more revenue than the entire ad account. Nobody had lied about that either. It simply was not in the report.

The rule underneath it

Every figure we publish carries its source and the window it covers, and a figure with two possible meanings gets two fields rather than one average. Platform return and blended return are different numbers. Stored in the same column, they become a third number that is true of nothing.

That is a database decision, made once, that keeps a whole class of argument from ever starting.