Full Answer

GA4 uses behavioural modelling to fill gaps created by consent rejection, cookie expiration, cross-device activity, and ad blocker interference. When a user rejects cookie consent but still browses and purchases, GA4 may model that session and attribute a conversion based on statistical inference from similar consented users. The modelled conversion appears in reports alongside directly measured conversions — with no visual distinction.

The reporting identity setting determines the modelling level. The default blended mode applies the most aggressive modelling, filling consent gaps and stitching cross-device journeys using Google signals data. Observed mode shows only data from users who were directly measured with cookies intact. Device-based mode uses only the device-level cookie with no cross-device stitching or consent modelling.

Switching between these modes reveals the modelling gap. A store that sees 1,000 conversions in blended mode and 650 in observed mode knows that approximately 35% of its reported conversions are modelled rather than directly measured. This is not inherently wrong — the modelling may be accurate — but it means one-third of the conversion data is an estimate rather than a measurement.

For WooCommerce stores making budget decisions based on GA4 data, the distinction matters. Modelled conversions carry uncertainty that directly measured conversions do not. A ROAS calculation built on 35% modelled data has a wider confidence interval than the same calculation built on fully observed data. Stores that independently capture conversions through server-side pipelines to BigQuery can compare their directly measured figures against GA4's blended view to quantify the modelling component and calibrate their confidence in GA4's reported performance.