Some of Your 'Conversions' Were Never Observed. They Were Modeled.
Ad platforms fill gaps in broken conversion tracking with statistical estimates, then report the total as one number, indistinguishable from a conversion that actually happened.
What 'modeled conversions' actually means
When a conversion can't be directly observed, a visitor declined cookie consent, a device blocks cross-app tracking, a cross-device path breaks the chain, ad platforms don't simply drop it from the count. They estimate whether a conversion likely happened, using patterns learned from the conversions they could observe directly, and add that estimate into the reported total alongside the real ones.
The total doesn't tell you which is which
A reported conversion count blends genuinely observed events with modeled estimates into a single figure, with no visible split anywhere in the standard dashboard view. A month showing 500 conversions could be 350 real and 150 estimated, or 480 real and 20 estimated. The headline number looks identical either way, and nothing in the interface flags which version you're looking at.
Why the blend matters for budget decisions
Optimizing toward a number that's partly a statistical estimate means partly optimizing toward the platform's own model of what a conversion looks like, not toward verified reality. The fix isn't demanding a platform stop modeling; every platform does it, for understandable reasons. The fix is reconciling the reported total against your own order records, so the gap between modeled and observed becomes visible instead of sitting invisibly inside a single trusted-looking number.