Read the Numbers

Analytics answers one question well: what happened. It is very bad at why, and every misuse of it comes from forgetting that. Numbers narrow down where to look; they do not tell you what to think.

What GA4 is counting

Google Analytics 4 is event-based. Everything is an event with parameters — a page view, a click, a scroll, a purchase — which is a better fit for how people actually use sites than the older session-and-pageview model. The vocabulary worth knowing is short.

• Users — distinct people, approximately, because it depends on cookies, devices and consent.
• Sessions — a visit; it ends after inactivity or at the day boundary.
• Events — anything you chose to record. Some are automatic, the useful ones you define.
• Conversions (key events) — the events that mean something happened: a form sent, a purchase, a booking.
• Engagement rate — the share of sessions that lasted, scrolled or converted. It replaced bounce rate, which measured the opposite and was routinely misread.
• Traffic source — where the visit came from: organic search, direct, referral, social, paid, email.

The four questions worth asking of any report

1. Compared with what? A number with no comparison period is not information. Sixty visitors is good or terrible depending entirely on last month.
2. Is the change bigger than the noise? Small sites are noisy. A 15% weekly swing on 200 visits is usually weather, not a trend.
3. Did the measurement change, or did reality? Whenever a metric moves sharply on one day, suspect the tag, the consent banner or a deploy before you suspect your customers.
4. What would I do differently if this number were double? If the answer is nothing, you are collecting it out of habit.

Diagnosing a conversion drop

Conversions fall. The instinct is to redesign something, and the correct move is to find out where the loss actually happens before changing anything a customer can see.

1. Confirm the tracking still works. Submit the form yourself and watch the event arrive. An astonishing share of "conversion drops" are a deploy that removed a script.
2. Split traffic from conversion rate. Fewer conversions with steady traffic is a site problem; steady conversion rate with less traffic is an acquisition problem. They lead in opposite directions.
3. Segment by source. A drop confined to one channel is a channel problem — an ad group paused, a ranking lost, a newsletter not sent.
4. Segment by device and browser. A drop confined to mobile Safari is a bug, not a market shift.
5. Look at the step before the conversion. Most losses happen one screen earlier than where you noticed them.
6. Only now, look at the page itself.

Following that order costs twenty minutes and routinely finds a broken form, an expired card on an ad account, or a consent banner change that stopped recording half your visitors — none of which a redesign would have fixed.