Consent Mode v2 on a small site: does modeled conversion data help bidding?

Setting up Consent Mode v2 is the easy part. The real argument is about what it buys you. Basic mode blocks tags until a visitor consents, so you get nothing from people who decline. Advanced mode sends cookieless pings and lets Google model the conversions it can't see.

What I'm not sure about is whether that modeling is worth much at low volume. Take a dental clinic with 20 form fills a month and about 40% of visitors rejecting cookies. Modeled conversions only feed into Google Ads once the traffic thresholds are met, so for a while nothing may visibly change.

I'd still run advanced mode, but I wouldn't judge the account by the modeled figures. Compare the consented count against the real lead count in the CRM each month and treat the gap as a rough estimate. How do you check that the modeled numbers are in the right range?

Agreed on the CRM comparison, with one wrinkle. The modeled numbers only cover conversions Google can tie to a click, so calls and walk-ins at a clinic never enter the model at all. If you set total CRM leads against Google's count, most of the gap is just your offline leads, and it looks like a modeling problem when it isn't. Split the check by channel instead: form fills against form conversions, calls against call tracking. Once you do that, whatever gap is left actually tells you something about the model.

Kai's split by channel is right, but it only works if the clinic's calls get tracked back to a click. If people are dialing a plain number on the site, those calls are invisible on both sides of the comparison, so the gap will look like modeling error even after you split it.

I'd push back on the last line. Untracked calls are a measurement gap, not a modeling error, and if we lump them into the model check we end up blaming the model for data it never saw. Once you've established the gap is untracked calls, the model isn't really being judged at all.

The fix is usually small. Either put a call tracking number on the site, or fire a conversion from a click listener on tel: links in GTM. The listener counts clicks, not completed calls, so it overstates things a bit, but it at least shows how many people tried to phone.

For a clinic with 20 form fills a month, I'd also track the share of form fills that are consented over time. That ratio doesn't depend on Google's thresholds, so it moves for reasons you can actually explain to the client.

I disagree on the click listener. A tel: tap isn't a lead. On mobile a lot of those taps are accidental, or just people checking opening hours. If you import it as a conversion action, smart bidding will learn to chase whoever taps the number, and that's not necessarily the people who book. I'd keep the listener as a GTM observation, or as a secondary action that stays out of bidding. If you want calls in the model, use call tracking with a minimum call duration so only real conversations count.

Agreed that the tap listener shouldn't feed bidding, but I'm not sure a duration cutoff holds up either. A caller booking a cleaning at a dental clinic can be done in 40 seconds, and a long call is just as likely to be a supplier or someone asking about insurance. Duration tells you about effort, not whether someone actually booked.

The better signal is whether the CRM marked the caller as booked, which you can send back as an offline conversion if your setup supports it. Until that's in place, I'd only let calls into bidding after checking the threshold against a few weeks of real bookings.

Booked calls are the right signal, but I wouldn't let them into bidding until the CRM can match calls to bookings for a few weeks and you're seeing roughly 15 booked calls a month. Below that, keep calls as a monthly review metric. Smart bidding needs a steady stream of conversions to learn from, and a thin, lumpy signal means one busy week of cleaning bookings can pull the whole campaign in a direction that doesn't last.

A clinic with 20 form fills a month probably sits under that call threshold too, so let the form conversions do the bidding and use the calls for context. Once booked calls cross the line consistently, import them as an offline conversion with a value, and the algorithm finally has something real to optimise for.

Agreed on keeping calls out of bidding until the volume is there. But check the CRM match rate before you trust the booked-call count at all. If only half the calls can be tied back to a click or call-tracking ID, you're hitting that 15-a-month threshold with a biased subset, so measure match rate first.

Match rate is the right check, but the bias usually hides in the source mix. Click-tied calls skew toward paid search, so if most of your unmatched bookings come from organic or direct visitors, bidding only learns from the paid slice of demand.