Google Ads attribution: data-driven modelling, and what it hides
Google moved everyone to data-driven attribution and a 30-day default window. The model is better than last-click — and it is still a model, computed by the party selling you the clicks, on data it partly estimates.
Google-reported conversions = observed gclid matches + modelled (consented-traffic estimate)
Google Ads typically reports 15–30% more conversions than a first-party server-side count, and it credits fractionally — a single sale can appear as 0.4 conversions in one campaign and 0.6 in another. Modelled conversions (from consent mode and cross-device estimation) are counted, but never itemised.
Fractional credit is honest about uncertainty. It is also impossible to reconcile against a bank statement.
gclid
30-day click, data-driven model
90-day click
+22%
Where the gap comes from
Four mechanics, each independently defensible, that compound into a number you cannot reconcile against your bank.
Modelled conversions
With consent mode, Google estimates conversions for users who declined cookies. They appear in the same column as observed ones.
What to do Segment by 'conversions' vs 'all conversions' and look for the delta; the modelled share is often 10–25%.
Fractional data-driven credit
One order becomes 0.37 conversions here and 0.63 there. Useful for bidding, unusable for cash reconciliation.
What to do Keep a whole-number, server-confirmed count of orders in your own dataset for finance.
Cross-device stitching from Google accounts
Google can stitch signed-in users across devices; you cannot see that stitch or verify it.
What to do First-party identity resolution on your own email gives you a stitch you can inspect.
The gclid is dropped constantly
Redirect chains, link shorteners, consent banners and Safari's link-decoration handling strip or lose the gclid before the landing page records it.
What to do Capture the click id server-side on the first request, before any redirect can eat it.
Enhanced conversions send hashed first-party data back to Google for better matching. Same trade as Meta's CAPI: it makes Google's model better, not yours.
Frequently asked questions
What are modelled conversions in Google Ads?
Statistical estimates for conversions Google believes happened but could not observe — typically because the user declined cookie consent or converted on another device. They are reported alongside observed conversions with no separate column by default.
Why do Google Ads conversions have decimals?
Data-driven attribution splits credit for one conversion across multiple touchpoints. A single order can appear as 0.4 conversions on one keyword and 0.6 on another. It is fine for bidding and meaningless for revenue reconciliation.
Is a 30-day window better than Meta's 7 days?
Only if the identifier survives 30 days. In Safari, a JavaScript-set cookie is capped at 7 days regardless of what the ad platform's window says, so most of the extra window applies to Chrome traffic only.
How do I reconcile Google Ads with actual revenue?
Record the gclid server-side on the landing request, keep it on a durable first-party visitor id, and confirm the purchase from your backend. Then count whole orders whose journey contains a gclid.
Related guides
See the receipt for every conversion.
Flowsk Signals stitches the anonymous click to the email to the purchase — first-party, server-side, de-duplicated on a key you choose. One snippet, $29/mo, and a journey you can inspect event by event.
More free tools
Same deal — instant, no signup.