Key takeaways
- Tie each collected field to a stated purpose, decision, owner, lawful basis, and retention period.
- Set and update consent state before dependent tags and page transitions.
- Separate observed, attributed, imported, and modeled quantities in reporting.
- Design revocation, deletion, access control, and vendor offboarding before launch.
01
Start with the decision, then work backward to data
“We might need it later” is not a measurement requirement. Name the decision, minimum evidence, population, retention horizon, and person accountable for use. The W3C privacy principles call for data minimization and purpose limitation: collect and use what is necessary for the stated context rather than treating availability as permission.
A campaign-quality decision may need source, consent state, lifecycle outcome, coarse geography, event time, and value. It rarely needs unrestricted free-text sales notes or a permanent record of every page action attached to a named person.
Swipe to compare every column
| Control | Question | Evidence to retain |
|---|---|---|
| Purpose | Which decision needs this field? | Approved measurement specification |
| Consent | What did the person allow, where, and when? | Versioned consent state and update event |
| Access | Who needs identifiable or row-level data? | Role grants and access review |
| Retention | When does the purpose expire? | Deletion schedule and execution log |
02
Make consent timing part of the architecture
Google’s consent-mode implementation guide requires a default state before commands that send measurement data and an update when the person changes their choice. It also warns that consent updates should occur on the page where the interaction happens, before navigation.
Test first visit, grant, partial grant, denial, later revocation, return visit, cross-domain navigation, slow banner loading, and tag failure. A polished consent interface is not enough if the underlying tags race ahead of it.
03
Keep modeled data visibly modeled
Privacy controls and technical loss create gaps. Platforms may use modeled conversions or aggregate estimates. Keep observed events, matched imports, attributed credit, and modeled quantities distinct in the semantic layer and dashboard labels.
Modeling can support decisions under missingness; it does not restore the exact journey of a person who declined tracking. Publish the method, eligibility thresholds, uncertainty, and changes that affect comparability.
04
Plan the end of the data before collecting it
Define deletion and revocation propagation across the warehouse, CRM, analytics tools, ad destinations, logs, backups, and vendors. Minimize debug logs and prohibit raw personal data in error messages. Review access and retention on a recurring schedule.
This is not legal advice, and obligations vary by jurisdiction and context. The practical standard is still useful: if the team cannot explain why a field exists, who can see it, when it leaves, and what happens after a person withdraws permission, the measurement design is unfinished.
Primary sources and further reading
Use the source material to validate details against your own context and current platform configuration.
This field note follows the XenGrowth editorial policy: primary sources where available, visible limitations, material review dates, and no invented first-hand experience.
Stay with the problem
