Modeled Conversions

Modeled conversions are sales Meta did not directly observe, so instead of leaving a gap it estimates them with statistics and folds them into the numbers you see.

Since Apple's tracking prompt and the wider privacy changes, a real slice of what happens after the click never reaches Meta cleanly anymore. Meta had a choice: show you a hole where that missing data should be, or fill the hole with its best statistical guess. It chose to fill it. Using aggregated, privacy-safe signals and modeling, it estimates the conversions it could not see directly and adds them to your reported totals. So the number in Ads Manager is now part measured and part estimated, blended into one figure that does not tell you which parts are which.

This matters because a modeled conversion and a confirmed one look identical on your screen, yet they are not the same kind of fact. One is a sale Meta actually recorded. The other is a probability dressed up as a count, Meta saying "based on the patterns, roughly this many likely happened." On a healthy, well-fed dataset the modeling is reasonable. On a thin one, it is guessing from very little, and the softer your signal, the more of your number is estimate rather than record.

The mistake is treating a modeled number like money in the bank. It is a planning input, useful for direction, not a receipt you can reconcile against your accounts to the exact sale. This is a large part of why Ads Manager and your real revenue rarely line up, and why the fix for signal loss, feeding Meta stronger first-party data through the Conversions API, matters: better signal means less of your reported number has to be invented.

Some of the sales Meta reports were never seen, only estimated, so trust the direction of a modeled number, not its last digit.