With less individual identification, measurement depends more on first-party data (collected directly by the company, with consent or another legal basis) and on behavioural modelling (statistical estimates of what observed data do not show). Preparing GA4 for this means collecting less, with more quality, and interpreting with humility.
The essentials in a definition
First-party data and behavioural modelling are the foundations of measurement with less individual identification. They require collection with a legal basis (GDPR), correct Consent Mode implementation and, for own analyses, tools such as BigQuery.
First-party data
These are data the company collects directly: purchases, accounts, subscriptions, interactions. They have a clear legal basis if collection is transparent, and are the most reliable for decisions and for feeding platforms. Define what you collect, why and for how long, in compliance with the GDPR, and avoid data you do not use.
Good data hygiene reduces risk and improves quality.
Behavioural modelling
When users decline consent, platforms may use modelling to estimate conversions and behaviours from patterns of those who consented. This modelling is aggregate and probabilistic, and depends on volume and a correct Consent Mode implementation. Modelled values are estimates, not observations.
Track the proportion between observed and modelled data.
Measuring well with less identification
- Define first-party data and purpose
- Implement consent correctly
- Validate against the database
- Interpret with trends and tests
Preparing GA4
Define clear key events, ensure consent is communicated to tags, maintain a consistent data layer and validate against the database. Consider exporting data to BigQuery for own analyses and lawful combination with first-party data (sales, CRM).
Document sources, rules and limitations.
Less individual data demands more discipline with the data you have.
Interpreting with humility
Accept there will be differences between platforms and between periods. Use trends and comparisons instead of absolute values, validate important decisions with incrementality tests and keep a reference in the database. Communicate uncertainty to management.
Review the data strategy every year.
Decision table
| Data type | Nature | Caution |
|---|---|---|
| First-party | Collected directly with a legal basis | Minimise and protect |
| Observed | Measured directly | Covers only those who consented |
| Modelled | Statistically estimated | Uncertainty and volume dependence |
| Database | Real sales | No attribution by channel |
Common mistakes to avoid
- Collecting everything “just in case”. It increases risk and noise.
- Treating modelled values as observed. It overstates certainty.
- Not validating. Errors go unnoticed.
How to measure the result
Track the observed/modelled proportion, differences from sales and trend stability.
Limits and exceptions
Modelling and privacy rules evolve, and this content is not legal advice. Modelled values are estimates with limits. Keep regular documentation and review.
Illustrative example
Illustrative example: a store reduces collection to the essentials, implements consent with communication to tags and exports GA4 data to an analysis base, where it cross-references with sales. It starts evaluating trends and validating decisions with tests, instead of chasing exact values.
In practice: checkpoints
Create a map of first-party data sources, with purpose, legal basis, retention period and owner. Review it once a year and whenever you launch a new collection. Compare observed and modelled data every month and note the differences. Test the consent flow on several devices, because small implementation faults can reduce observed data more than the refusals themselves. Create a simple dashboard showing the proportion of sessions with consent, the volume of key events and the gap with the database. Agree with management how modelled numbers will be reported, so that expectations are set from the start, and keep the documentation where new team members can find it.
Frequently asked questions
Does modelling replace real data?
No; it estimates what is missing, with uncertainty.
Which first-party data are most useful?
Purchases, subscriptions and interactions with consent or a clear legal basis.
Do I need BigQuery?
Not always; it is useful for advanced analyses and combining data.
How do I communicate uncertainty?
With ranges, trends and the database reference.
What should I prioritise?
Quality and legitimacy of first-party data.
Prepare GA4 for less identification: lawful, minimal first-party data, well-implemented consent, validation in the database and interpretation by trends and tests.
Sources and recommended reading
- General Data Protection Regulation (GDPR)
- Google for Developers: set up consent mode
- GA4 BigQuery export
Last editorial review: October 2026. Informational content; always confirm current conditions in the official sources before deciding.