Consent Mode modeled conversions: trust them for bidding, or only for reporting?

Consent Mode v2 keeps coming up in measurement threads, and I still can't decide how much weight modeled conversions deserve. Google estimates conversions from users who declined cookies, so Google Ads often shows more conversions than the consent rate alone would explain. For bidding that's arguably useful, since Smart Bidding gets signal it would otherwise miss. For reporting it's a problem, because the number you show a client now includes estimates.

Take a dental clinic where 40% of EU visitors decline analytics. The modeled count can look healthy while booked appointments in the CRM stay flat. I'd keep modeled conversions on for bidding, but reconcile against the CRM every month before trusting any CPA trend. How do you handle it when the two numbers drift apart?

Keeping modeled conversions on for bidding makes sense, but a small clinic may not have enough consented traffic for Google to model much, so the modeled share could be close to zero. If that's the case, the drift probably comes from duplicate form fills or calls that never became bookings, not from cookie decline. Splitting the conversion action by source (form, call, booking confirmation) will show which one is drifting, and a single monthly total hides that.

I don't think that last step holds. Saying the drift "probably comes from duplicates, not cookie decline" assumes a small clinic's consent gap doesn't matter, and it does, just in the opposite direction. Declined users are invisible in GA4 unless Google models them. If the clinic has enough volume for modeling to kick in, the estimate fills part of the gap. If it doesn't, the gap stays open. Either way the platform undercounts, while duplicate form fills push the count up. So a flat CRM next to a healthy Ads number can come from two errors that partly cancel each other out.

To separate them, compare the share of declined clicks in your consent reporting against the CRM over the same period. If the platform number roughly matches bookings after modeling, duplicates are the likelier culprit. If it doesn't, the source split you mentioned is the right next step.

Agreed, and the cancelling errors are what make this so hard to read. For a clinic, a big share of leads usually comes in as phone calls, and those conversions often don't go through the same tag as form fills, so consent modeling may not estimate them at all. That means the gap you're measuring is really two blind spots stacked together: users who declined consent on the site, and calls that never reach the tag. Split calls from forms before you compare the platform number with the CRM, otherwise you won't know which gap you're looking at.

Calls matter a lot here, and for call-heavy clinics I'd take it further: the modeling question matters less than whether the call conversion is firing at all, because a missing call tag can't be modeled back in. Which call tracking setup are you looking at, a forwarding number or call extensions reporting straight to Ads?

It comes down to whether the front desk logs each call against the number it came in on. If calls land in the CRM with a source field, you can match them against the Ads call count and see where the tag is missing calls. If reception only keeps them in the phone system or types them in by hand, the Ads number will look low no matter what modeling does, so the fix is in how calls get recorded before anything changes on the tag side. Pull one week of raw call logs and count them before you trust either number.