<b>Modeled conversions inflate reported results by 9–17%</b>

<b>Modeled conversions inflate reported results by 9–17%</b>
Across 26 accounts cross-checked vs backend, 90d. Modeled = conversions Meta estimates, not observed.
— Share of reported conversions that are modeled: median 14%
— Accounts under-deduplicating server+browser events: ▲ to 25%
— Backend-to-Meta variance: median 12% Meta-over
The Conversions API doesn't remove modeling — it just gives Meta cleaner signal to model from.
Read: hold a fixed Meta-vs-backend ratio per account and trust the trend, not the absolute number. The dashboard is always a little generous.