Why platform conversions do not match your real sales

You meet it the first time anyone compares the ad platform's conversion column against the system that takes the money. A finance lead notices the monthly report claims more sales than the accounts recorded. A founder adds up what Google and Meta each say they produced and gets a number larger than the whole business. Somebody asks why the dashboard says the campaign paid for itself when the bank does not agree.

The mirror image is quieter and more expensive. An account whose offline conversion import stopped weeks ago shows a cost per acquisition several times the real one, everything looks unaffordable, and budgets get cut on a number that was never true.

A platform-reported conversion is not a count. It is a causal claim: this platform asserts that it caused an outcome, under its own attribution rules, inside its own window, using its own modelling for the conversions it could not observe. Your booking system is doing something completely different. It is counting events that happened. The two were never designed to agree and there is no setting that makes them.

Over-claiming comes from generous windows, view-through credit, statistical modelling filling the gap left by consent and app-level tracking restrictions, and two platforms both taking credit for the same person. Under-claiming comes from the opposite: a broken offline import, a consent banner blocking the tag, a sales cycle longer than the window, or an outcome that simply happens somewhere the pixel cannot see.

Both directions can be live in one account at once, on different campaigns, which is why a single correction factor never fixes it.

Every automated bidding decision the account makes is made against the platform's own count. If that count is inflated, the target you set is effectively far stricter than you intended and the account buys less volume than you are paying for. If it is deflated, the target is far looser and the account spends into outcomes you would never have approved.

The reporting cost is worse than the bidding cost, because it compounds. A monthly report built on platform figures, sent to somebody who also sees the real revenue, spends its credibility the first time the two are put side by side. And the decisions taken off the back of it, which channel to scale, which to cut, whether the whole programme is working, are all taken on the wrong number.

The rule that follows is simple and it is not optional: pull the client's real outcome series for the same periods and compute a claim ratio per platform BEFORE interpreting any movement in conversions, cost per acquisition or return on ad spend. Not after, not when something looks wrong. Before.

This matters because the alternative is not neutral. Reading a platform number without its ground truth is not a slightly less informed version of the right analysis. It is a confident analysis of a number that may be off by a factor, and confidence is exactly what makes it expensive.

It also changes what you do when something moves. A drop in reported conversions is an attribution question first and a performance question second, because the cheapest explanation is usually that credit moved rather than that demand did.

It is not a claim that platform data is useless. Platform numbers are the only thing that can tell you which keyword, audience or creative produced a result, and no booking system will ever know that. They are the right instrument for relative questions inside one platform, and the wrong instrument for the absolute question of how much business happened.

It is not a tracking bug to be fixed. Better tracking narrows the gap and never closes it, because the gap is mostly attribution philosophy rather than lost data.

It is not a number you can borrow. The ratios above are one account. Yours will be different, will move, and is the thing worth measuring rather than assuming.