Measurement
Part of How to set up GA4 data-driven attribution for UK sites
How to set up GA4 data-driven attribution for UK sites
Set up GA4 data-driven attribution on UK sites: the exact Admin path, Consent Mode v2, PECR and ICO consent guidance, and a settings checklist.
What to take away
- GA4 data-driven attribution only credits touchpoints the property can see, so tag coverage and consent state are settled before model choice.
- Conversion modelling estimates conversions from unconsented users, but Google applies eligibility rules and it does not replace consent.
- Lookback window, reporting identity and key event selection change the numbers more than the model name does.
- Keep a dated log of consent-mode settings, model changes and exclusions, because someone will ask for the basis of a figure later.
Why consent mode comes first
Data-driven attribution splits credit between ad clicks, organic sessions and other touchpoints. It learns from conversions recorded in your GA4 property. If tags fire before a visitor makes a choice, the model learns from a population that includes people who never consented.
Consent mode passes the visitor's consent state to Google tags and adjusts what is stored. Consent Mode v2 adds the ad_user_data and ad_personalization signals alongside analytics_storage and ad_storage, and Google requires it for UK and EEA traffic served through Google Ads. Implement it in Google Tag Manager by setting a default denied state before tags fire and attaching consent settings to each tag, or directly through gtag.
Your banner must capture that state accurately, and the categories it offers must match the tags you run. Consent management platforms such as OneTrust, Cookiebot and CookieYes pass the signals that consent mode reads; a banner that bundles analytics and advertising consent hides which data you actually hold.
For UK sites the processing rules come from the UK GDPR, overseen by the Information Commissioner's Office (ICO), whose UK GDPR guidance hub covers lawful basis, data minimisation and retention. Cookies and similar technologies sit under the Privacy and Electronic Communications Regulations 2003 (PECR), and the ICO's cookies and consent guidance sets the standard the banner must meet. In attribution terms, that means deciding what you measure, how long identifiers persist and what you could defend if a visitor asked.
Apple's App Tracking Transparency gives iPhone users a choice over cross-app tracking, so mobile ad identifiers are frequently missing. Consented and unconsented groups therefore differ by device as well as by preference.
The GA4 settings that change the result
| Setting | Where it sits | What to check |
|---|---|---|
| Attribution model | Admin, Attribution settings | Data-driven chosen, and enough conversions to train it |
| Lookback window | Admin, Attribution settings | Whether the 30-day default matches your buying cycle |
| Reporting identity | Admin, Reporting identity | Blended or observed, and who can audit the join |
| Key events | Admin, Events | Only events you would act on |
| Consent mode | Tag setup | Basic or advanced, and tested in debug view |
Data-driven attribution needs enough path data to train, and in GA4 it is enabled under Admin > Attribution settings > Attribution model, where you select Data-driven and save. Google's Analytics Help documentation on attribution models and conversion modelling sets out how models are selected and which properties qualify, and Google Ads Help documents the consent mode requirements. Check your converting volume before treating a channel split as a finding.
GA4 attribution settings checklist
- Attribution modeldata-driven, enough conversions
- Lookback windowmatch buying cycle
- Reporting identityblended or observed, auditable
- Key eventsonly actionable events
- Consent modebasic or advanced, debug tested
Reporting identity decides how GA4 stitches sessions together: user ID, device ID or modelled data. Blended identity joins more sessions and can inflate cross-device results without showing which join did the work. That is why an attribution workflow that starts with a decision ticket suits teams sharing one property.
A setup sequence for UK sites
- Inventory every tag, event and destination in Google Tag Manager, and mark which ones depend on consent.
- Implement consent mode in Google Tag Managerset a default denied state for analytics_storage, ad_storage, ad_user_data and ad_personalization, update the signals from the banner, then confirm the denied state registers in DebugView.
- Confirm Data-driven is the attribution model in Admin > Attribution settings and save; record the date.
- Set the reporting attribution lookback window in the same panel against your real buying cycle rather than the 30-day default.
- Under Admin > Reporting identity, choose blended or observed and write down the choice; then mark key events in Admin > Events so the model learns from events you would act on.
- Leave the model running for a full cycle before comparing it with a holdout or incrementality test.
Test the sequence in a staging property before you touch live settings. Model and lookback changes create a break, so record the date and keep reports either side of it separate.
UK GA4 attribution setup sequence
- Inventory tags, events, destinations, consent dependencies
- Configure banner and consent mode, test denied state
- Set reporting identity and document choice
- Choose lookback window against buying cycle
- Mark only actionable key events
- Run model full cycle before comparison
Recording the model and its limits
GA4 does not give you a presentable change history for attribution settings. Keep a dated record of the consent-mode version, banner wording, model, lookback window, key events and exclusions.
Consent mode cannot close a measurement gap by itself, and conversion modelling only estimates conversions for properties that meet Google's eligibility rules: consent mode implemented, enough consented events to train on, and enough unconsented traffic to model. Where a large share of visitors never consent, a cookieless attribution approach explained for UK teams covers the alternatives, including server-side collection and media mix work.
The model output is an allocation of credit, not proof of incrementality. Test one channel at a time before you move budget on the strength of a GA4 channel report.
Common questions
Does data-driven attribution work without consent mode?
It can run, but it credits only what the property measures. On a UK site with low consent rates, that means a model trained on a biased sample and reports that understate channels reaching cautious visitors.
How many conversions does data-driven attribution need?
Google's Analytics Help documentation sets out the data requirements for data-driven attribution and conversion modelling rather than a single number that fits every property, and explains when other models apply. The practical test is eligibility: enough path data to train the model, and consent mode in place with sufficient consented and unconsented events. Check the model status in Admin > Attribution settings and your own converting volume rather than assuming the model is live.
Is conversion modelling allowed under UK GDPR?
Modelling estimates behaviour rather than identifying individuals, but the collection underneath still needs a lawful basis under UK GDPR and, for cookies and similar storage, consent under PECR. Keep clear information and a retention limit, and read the ICO's UK GDPR guidance hub and cookies guidance before using estimated conversions in client reporting.
Which lookback window suits UK B2B sites?
Compare 30, 60 and 90 days over the same period and see whether the click path changes. If most deals close after 60 days, a 30-day window will credit the last click and hide the first.
