Making Sense of GA4: A Setup and Migration Guide

GA4 replaced Universal Analytics, and the transition frustrated a lot of teams. The reports moved, the metrics changed names, and the underlying model is genuinely different. Once the model clicks, though, GA4 is more flexible than what it replaced. This is the setup that gives you numbers you can trust.

The core shift is that everything is an event. Universal Analytics had distinct hit types for pageviews, events, transactions, and so on. GA4 has one type: the event. A pageview is an event. A purchase is an event. This is why the interface feels unfamiliar, and why custom tracking is more consistent once you adjust.

Plan your events before you implement. List the actions that matter to the business — form submits, sign-ups, add-to-cart, purchase, key content views — and name them consistently, using GA4's recommended event names where they exist. A written measurement plan prevents a messy account later.

Use Google Tag Manager. Implement GA4 through GTM rather than hard-coded tags. It makes changes safer, keeps a version history, and lets marketing adjust tracking without a deploy.

Mark the right events as key events. Key events, formerly called conversions, are the handful of events that represent real business value. Marking too many dilutes the concept; be selective.

Configure data retention and PII controls. Set event data retention to the maximum 14 months, and make sure you are not sending personally identifiable information into GA4, which is against its terms and a compliance risk.

Link the other Google products: Google Ads, Search Console, and BigQuery. The BigQuery export is free with GA4 and gives you raw, unsampled event data for serious analysis.

Rebuild your key reports as Explorations. The default reports are thin. The Explorations section is where GA4 gets powerful — funnels, path analysis, segment overlap, and cohort tables you configure yourself.

Expect your numbers to differ from Universal Analytics. Different session definitions, different attribution, and different bot filtering mean the two will never reconcile exactly. Pick a switchover date, annotate it, and compare trends going forward rather than chasing the old figures.