Guide

Your ad platforms claim more sales than you made.

Add up what Meta, Google and TikTok each report and the total beats the revenue in your accounts. Nothing is broken. Five separate mechanics produce the gap, and each one has a different fix. Here they are, and the number to budget on once you have them.

Every platform claims the same sale

What happens

One buyer sees a Meta ad on Monday, searches your brand on Wednesday and buys. Meta claims the sale. Google claims the sale. Neither knows about the other. Run three channels and you can book one order while three dashboards each record one conversion. The overlap is not a bug in any of them; it is what a siloed attribution model is designed to do.

What to do about it

Never add platform conversions together. Treat each number as that platform's claim on a sale, not as a count of sales. The only figure that counts orders once is the one from your own order table.

View-through conversions

What happens

Meta counts someone who saw an ad and did not click, then bought within a day. Display and video networks do the same over longer windows. Some of those people were always going to buy. The ad was in the room, not in the decision, and it still gets the credit.

What to do about it

Split click-through from view-through in every report and look at them separately. View-through is useful as a directional signal for upper-funnel work. It is not evidence of a sale caused, and it should not sit in the same total as click-through.

Attribution windows that do not match

What happens

Meta's default is seven-day click and one-day view. Google Ads defaults to a longer click window and reports the conversion on the day of the click, not the day of the sale. GA4 reports on the day of the sale. So the same order can land in three different weeks, and a week-on-week comparison across platforms compares things that are not the same thing.

What to do about it

Fix one window and one date basis for your own reporting, apply it to every channel, and state it on the report. If you also quote platform-native figures, label them as such. Most reporting arguments are two people using different windows and neither saying so.

The same conversion fired twice

What happens

A buyer refreshes the confirmation page and the pixel fires again. Or the Google Ads tag is installed directly and the same purchase is also imported from GA4, and nobody set one of them to secondary. This inflation is silent, consistent and inside a single platform, which makes it the one people find last.

What to do about it

Send an order ID with every conversion event and let the platform deduplicate on it. Then audit: pull one week of platform conversions and one week of orders, and match them line by line. If the platform has more rows than your order table for the same window, you have found a duplicate, not a discovery.

Modelled conversions

What happens

Cookie loss and consent mean platforms cannot observe every conversion, so they estimate the ones they cannot see and report the estimate alongside the ones they can, usually without separating them. The estimate is not random: it is produced by the party being graded.

What to do about it

Ask each platform what share of reported conversions is modelled and keep the figure. Judge trends on observed conversions where you can, and treat the modelled share as a confidence interval on the rest, not as a fact.

The reconciliation that settles it

Take one month. Put your real orders on one side, from the order table, net of refunds and cancellations. Put each platform’s reported conversions on the other, click-through only, on a single agreed window. The difference between the sum of the claims and the real total is your overlap. Divide it out and you have a discount factor you can apply to every platform figure until the next time you check.

That discount factor is not the truth either. It is a much better guess than the raw numbers, and it stops the budget drifting toward whichever platform reports most generously. For the channels big enough to matter, the only real answer is to turn one down and watch what happens, which is what a geo holdout test does properly.

The same problem, handled in the Paid and Analytics modules. Related reading: five marketing numbers that are probably lying to you.

One number for a sale, however many platforms claim it.

Connect your ad platforms and your order data, and Decifer reconciles the claims against what you actually booked, on one window, with the overlap shown rather than hidden.

Request access How the engine works