Which events are actually worth measuring on a site

When analytics goes onto a new site, the default behaviour is to switch everything on at once: page views, scroll depth, click heatmaps, session duration. Six months later the dashboard holds fourteen charts. Nobody looks at any of them, and the question “what is happening on the site” is still unanswered. What is missing is not data.

The one criterion

Before an event goes on the list, write down one question: if this number rises or falls, what will I do?

Every event where you cannot write that answer comes off the list. This single exercise takes most setups from fourteen charts down to four or five events, and the ones that survive actually get read.

Page views fail the test. Nothing follows from them going up or down.

Five events cover most sites

The primary action. A quote form submitted, an appointment booked, a purchase completed. This is why the site exists. It needs to be measured as one event so the ratio maths stays simple.

The start of that action. The form was opened, or the first field was filled. The gap between starting and completing shows directly whether the form itself causes trouble, points at what to fix, and no other measurement gives you that.

Secondary contact. The phone number was tapped, a WhatsApp link opened, the address copied. In small businesses most conversion runs through those three actions and none of them shows up in form tracking.

A sign of deep interest. The pricing page was opened, a case study read to the end, a file downloaded. Intent is read here.

Errors. A form that would not submit, a payment refused, a search returning nothing. This is the most skipped group. It is also the one whose fixes pay back fastest.

Naming is decided once

Event names will be read six months from now, most likely by somebody else; when form_submitted, Form Submit and submit-form sit next to each other in one dashboard it stops being clear which one counts what, and nobody dares delete any of them.

Pick one shape and write it down. Lower case, underscores, verb last. Keep the list of names in the same place, because looking there when adding an event stops the same thing being added a second time under a different name.

What not to collect

Putting personal data inside an event turns a simple measurement job into a legal one. It also buys nothing on the analysis side.

Email address, phone number, free text from a form: none of them are event parameters. The question is “how many people wrote”. No identity is needed.

URLs need care too: addresses carrying a search query or a coupon code go into the event data exactly as they are, and cleaning that data afterwards is far harder than you expect.

The one check after setup

Once the setup is finished, wait a week. Then open the dashboard and look at the count on every event one by one.

An event still at zero is either wired up wrong or genuinely unused, and both possibilities carry information. If a count is higher than you expected, the same action is most likely firing from two places at once. That distorts every ratio until it is fixed.