What Is Modelled Conversions?
Modelled conversions are estimated conversions that Google Ads, GA4, and Meta report when direct measurement is blocked by consent refusal, ITP, iOS 14 (ATT), or missing cross-device joins. The platform uses machine learning trained on observed conversions to infer how many un-observed ones almost certainly occurred, then adds the estimate to the reported total. In 2026, modelled conversions can be 20 to 60 percent of reported conversions on a modern account.
GA4 + Google Ads modelled conversions on an EU account
A German ecommerce brand runs Consent Mode v2 with 40 percent consent refusal. Directly observed Google Ads conversions were 1,240 for the month. Modelled conversions added another 810, taking the reported total to 2,050. Ads Manager showed a tROAS-hit campaign 'on target'. Only after inspecting the `conversions (by conversion time, model)` split did the operator see that pausing the campaign for a 10-day test showed real backend revenue drop matched the modelled number closely, so the model was accurate in this case.
Benchmarks
- Google Ads (Consent Mode v2 active): 15 to 45 percent of reported conversions are modelled.
- Meta (post-ATT, CAPI enabled): 20 to 60 percent of iOS conversions are modelled via AEM + statistical modelling.
- GA4 (blended data mode): up to 30 percent of sessions from unconsented users are behaviourally modelled.
- Directional accuracy of platform modelling vs holdout tests: usually within +/- 15 percent, occasionally worse when consent rate is very low or very high.
Why it matters
If you do not know which reported conversions are observed and which are modelled, you will misread every experiment. A drop in reported conversions might just be a modelling update. A rise might be a threshold change. Every serious 2026 account splits the two in reporting and treats modelled numbers as directional, not as ground truth.
Common mistakes
- 1.Treating modelled conversions the same as observed ones in a lift test. Modelled numbers are estimates conditioned on the campaign being live, so they cannot honestly evaluate a pause test.
- 2.Not enabling Consent Mode v2 in the EU. Without it Google has no basis to model, so reported conversions collapse.
- 3.Ignoring the Ads Manager 'attribution model' toggle in Meta. Modelled conversions look different under 7-day-click vs 1-day-click, and comparing across models is meaningless.
- 4.Assuming modelling recovers 100 percent of lost signal. It does not; a healthy CAPI + Enhanced Conversions setup still beats modelling for the same account.
Put Modelled Conversions to work
Free calculators
Related services
FAQs about Modelled Conversions
How can I see which of my conversions are modelled?
In Google Ads add the 'Conversions (by conversion time, model)' column. In GA4 open Admin then Attribution Settings, the Reporting Identity control shows blended vs observed. Meta does not expose a clean split, but any iOS conversion without a matching Pixel event id is functionally modelled via AEM.
Are modelled conversions accurate?
Directionally yes, within roughly +/- 15 percent when your Consent Mode, CAPI, and Enhanced Conversions setups are healthy. In pathological cases (very low consent, broken CAPI, missing click identifiers) the error grows quickly.
Should I optimise campaigns off modelled conversions?
Yes for tactical bid decisions inside the platform, because the auction itself uses modelled data. No for boardroom-level allocation; for that pair modelled numbers with a quarterly incrementality test or MMM.
Will disabling modelling improve accuracy?
No, it just hides the same missing conversions instead of estimating them. The right response to noisy modelling is to fix the underlying signal loss: Consent Mode v2, CAPI, first-party data, Enhanced Conversions.
Related terms
Google's consent signalling API; required in EEA for ad measurement.
Server-side event stream from your site/CRM to Meta or Google.
Google's hashed first-party data sent server-side to recover lost matches.
Meta's privacy-safe event system after iOS 14.5; caps you at 8 ranked events.
How credit for a conversion is assigned across ad touchpoints.
Apple's App Tracking Transparency; broke cross-app deterministic tracking.
Holdout study proving whether ad spend actually caused conversions vs stealing organic credit.
Turning ads on/off by region to isolate true incremental lift vs organic baseline.