Most GA4 properties we audit are collecting plenty and answering nothing. The tag fires, the numbers move, and nobody can say which campaign produced a customer. That is a planning failure before it is a technical one.
The measurement plan comes first
We write down the decisions the business needs to make, the questions behind them, and only then the events that would answer them. A page view is not a question. “Which channel brings trials that convert to paid” is.
Naming that survives
One naming convention, applied everywhere, documented in a spreadsheet the client keeps. Parameters carry the detail — plan, value, step — instead of a new event name for every variation, because five hundred distinct event names is the same as none.
Conversions defined once
The events that count as conversions are agreed with whoever reports the numbers, so the marketing dashboard and the board pack cannot disagree. Values are passed where a monetary value genuinely exists, and left out where it would be invented.
Consent, properly
Consent mode wired to the banner so that a decline actually stops collection. Reporting reflects what consented users did, and the gap is acknowledged rather than quietly modelled away. We would rather hand over a number with a stated limitation than a confident wrong one.
Beyond the interface
Raw event data exported to BigQuery, because the interface’s sampling and thresholds start to hurt at exactly the point the questions get interesting. That export is also what lets marketing data sit beside product and revenue data.
Implemented through Tag Manager, reinforced with server-side tracking, and used every day by analytics and CRO.