How to recover the conversions your cookie banner is costing you

What you actually lose when consent is denied

Rejecting cookies does not delete the visitor. They still land, browse and, often, buy. What disappears is the thread that ties that purchase back to the ad they clicked. Without cookies, the identifiers that stitch click to conversion never get written, so Google Ads sees the spend but not the sale, and your reports quietly stop crediting the campaigns that are working.

That is why so many teams watch conversions fall overnight without touching a single campaign. On a typical site around a third of sessions sit behind the banner, so the shortfall is large and it looks like a performance collapse when it is really a measurement gap. The good news is that a gap this structured is exactly what modelling is built to close, provided the pipeline underneath it is wired correctly.

The recovery, in order

Recovery is not one setting, it is a short chain where every link has to be right. Skip step two and the rest starves quietly. Do them in this order and check each one before moving on.

1. Turn on advanced consent mode

   Switch Consent Mode from basic to advanced. Basic mode blocks every Google tag until consent is granted, so a declined visitor sends nothing and recovers nothing. Advanced mode still loads the tags in a restricted, cookieless state when consent is denied, and those anonymous pings are the raw material modelling runs on. This one choice is the difference between recovering a share of the loss and recovering none of it.

2. Send all four signals before tags load

   Set ad_storage, analytics_storage, ad_user_data and ad_personalization to their default denied state in the page head, before any Google tag initialises, then update them the instant the visitor chooses. All four have to be present. A common failure is shipping only the two analytics signals and wondering why Ads never models anything.

3. Check the denied pings leave the browser

   This is the step almost everyone skips, and it is where recovery silently dies. Open the network panel, decline consent, and confirm the cookieless collect request actually fires with gcs=G100 and the four signals attached. If your banner initialises after the Google tag, the ping is never sent and the model has nothing to learn from. No ping, no recovery, no error message.

4. Add enhanced conversions for the yes group

   Advanced mode recovers the visitors who said no. Enhanced conversions protect the ones who said yes. Hash a piece of first party data such as an email at the moment of conversion, so attribution survives cookie deletion and cross device journeys. It runs independently of Consent Mode, so the two stack rather than compete, and together they cover both halves of your traffic.

5. Move reporting to the server

   Route your tags through a server container so events reach Google and Meta from your own server rather than the visitor's browser. Server side reporting sits out of reach of ad blockers, browser privacy limits and third party cookie loss, which makes every recovered signal above it more durable. This is the step that carries a measured +37% ROAS in server side reporting across Amplio Data client implementations, a range that depends on your setup, not a guarantee.

6. Confirm modelling actually switched on

   Modelling only activates once you clear Google's daily traffic thresholds, and below them it fails with no alert at all. Check the modelled conversions column in Google Ads and the modelling behaviour flag in GA4 to confirm recovery is genuinely running. A pipeline that looks perfect but sits under the threshold recovers nothing and tells you nothing, so this final check is not optional.

The honest limits

Anyone promising to recover everything is selling you a story. Two limits are worth stating plainly. First, modelling estimates the conversions it cannot see from the ones it can, so it is a statistical fill, not a replay of lost data. Across our client work it brings back 20 to 40 percent of the conversions consent costs you, which is the difference between a campaign that reads as dead and one that reads as profitable, but the rest stays gone. Second, none of this reaches backwards. Modelling only works on data collected while consent was configured correctly, so conversions lost during a broken or missing consent period cannot be recovered at all. That is the real cost of leaving it unfixed, and the real case for closing the gap sooner.

The recovery you get is also only as honest as your banner. A banner that traps people into accepting produces consent that is legally fragile and behaviourally noisy, which poisons the very model you are trying to feed. Clear choices in, clean signal out.

This is the whole reason Velo exists: it ships the advanced Consent Mode wiring, the four signals in the right order, server side reporting and the audit trail as one configured layer, so the chain above is set up correctly by default rather than assembled by hand and hoped over. You stay compliant, and you keep the conversions the banner would otherwise quietly cost you.

Common questions