By · Updated

folders, a data table and magnifying glass on a desk
Practical method

Set up proportionate audience measurement

Measurement should answer useful questions: information found, interrupted journeys or content needing correction. Define decisions before collecting events. A dashboard does not by itself justify everything a tool can collect.

01

Connect each measure to a decision

List questions and possible actions: improve a page, simplify a form or clarify an empty result. Define units, period and denominator. A visit and a person are not interchangeable units.

Separate technical counting from user success. A page view proves neither reading nor a qualified enquiry. Compare signals and report limitations.

02

Limit data and dependencies

Describe events, parameters, recipients and retention. Exclude free text, email addresses and unnecessary values in tracked URLs. Check information automatically sent by the tool too.

Review privacy claims against your actual configuration. Consent-exemption conditions are specific; a tool name is insufficient. Consult current references and have uncertain points reviewed.

03

Check what is actually transmitted

In a trial environment inspect network requests and browser storage. Test user choices, different pages and a later return. Compare observations with the collection plan.

Check events are not sent twice and errors do not include entered fields. Use synthetic data. Record version, configuration and check date.

04

Interpret without false precision

Document blockers, bots, tracking changes and incomplete periods. Growth may result from a new counting method. Keep the method alongside the chart.

Review measures leading to no decision. Remove unnecessary events and verify retention. Smaller collection is easier to understand and maintain.

05

Separate intent, receipt and processing

A button click, technical receipt and qualification of an enquiry are different events. Decide which answers your question. Define an observable trigger and a rule preventing retries from being counted as new enquiries.

Test success, failure and double clicking with fictional references. Compare measured events with test-environment records. Where measurement and business processing have different scopes, explain the difference rather than adjusting figures to force agreement.

06

Compare periods that are actually observable

Changes in consent, bot filtering or event triggers can alter observed counts without altering activity. Annotate configuration versions and examine differently configured periods separately before concluding the service changed.

Check time zones, incomplete days and reporting delays too. An enquiry received late in the day may be processed tomorrow; state whether records are grouped by receipt or processing date. Interpret changes only under comparable conventions.

A record to keep with the decision

Minimum evidence for a review
ItemEvidence
QuestionIntended decision and denominator
CollectionFields, recipients and retention
CheckObserved requests and configuration

Frequently asked questions

Does an exempt tool make everything compliant?

No. Check configuration conditions, other processing and any transfers. Exemption for a tracker does not constitute overall service approval.

Reference material

CNIL · Solutions de mesure d’audience

The practical checklist is an editorial synthesis to adapt to your service. It does not constitute a certification or an audit result.

Continue with a related decision