Consent Mode modelled conversions: report them as real, or keep them separate?

Say a lead gen site gets around 55% accept on its consent banner. GA4 will show conversions that look cleaner than that opt-in rate would suggest, because Consent Mode v2 models the missing ones. Those modelled numbers feed Smart Bidding anyway, so some people want to treat them as real in weekly reporting too. I'd keep them apart. Modelled totals are fine for bidding, but if you judge a landing page change on them, you're partly judging the model's assumptions about the users who declined. What I'd do is report observed and modelled in separate columns and base budget decisions on observed volume trends. Has anyone compared modelled totals against CRM leads over a few months? I'd like to see how far apart they drift.

I'd push back on setting budgets from observed trends. At a 55% accept rate you're only seeing the slice of users who opted in, so the series can move because of a banner change or a shift in who consents, not because performance actually changed. If Smart Bidding is spending against the modelled numbers, your budget logic should follow them too, and CRM leads are what you check the model against.

Agreed the observed series can drift with the banner, but the banner also changes what gets modelled. Consent Mode only models conversions where the user declined and there's enough consented traffic to learn from. So a page with a weak accept rate or thin conversion volume can end up with a model that's mostly guessing. Check the modelled share of conversions per landing page before you trust any page-level comparison.

Modelled share per page is a useful check, but it often tracks the traffic mix more than conversion quality. A high-consent landing page can look clean just because its visitors are more willing to accept cookies, not because those visitors convert any better. So it's worth splitting modelled share by traffic source before you compare pages. Otherwise you're mostly comparing audiences, not pages.