<b>~28% of reported Facebook conversions are now modeled, not observed — and they cluster on your best ad sets

<b>~28% of reported Facebook conversions are now modeled, not observed — and they cluster on your best ad sets</b>

Modeled = Meta statistically estimates conversions it cannot directly track (post-iOS signal loss).

Delivery-breakdown audit, 30 accounts:
— Modeled share of reported purchases: median 28%
— Modeling skews toward high-volume ad sets — inflating their apparent ROAS vs small ones
— Conversions API (CAPI) cut modeled share to ~16% by feeding server-side events
— Accounts on browser pixel only sat at 35%+ modeled

Your 'best' ad set may just be the one Meta modeled most generously.

Read: install CAPI to shrink the modeled fog before you trust ad-set-level ROAS comparisons.