Setting Up Ecommerce Analytics You’ll Actually Use
Most stores have analytics. Far fewer have analytics anyone opens before making a decision. The gap is almost never caused by missing data — it is caused by data that was collected without a question in mind.
Start from decisions, not events
The useful exercise is to write down the recurring decisions the business actually makes: what to merchandise on the category page, which products to promote, whether a template change helped, where spend should go, what to fix next. Then ask what each decision requires you to know.
Most stores discover they need far fewer events than they are tracking, and that two or three of the ones they genuinely need were never implemented.
The events that earn their place
The funnel, properly delineated
View item, add to cart, begin checkout, purchase. This exists in every setup, but it is frequently wrong in ways that make it useless: add-to-cart firing on the variant selection rather than the button, or checkout steps that do not match the actual checkout. Verify the funnel against reality before trusting anything built on it.
Search and filter behaviour
What people search for and get no results on is the most direct product and merchandising feedback available, and it is routinely untracked. Filter usage tells you which attributes matter enough to structure the catalog around.
The failures
Failed payments, validation errors, out-of-stock views. Failure events are more actionable per unit of effort than almost anything else, because each one is a specific, fixable loss rather than a trend to interpret.
Where setups usually go wrong
Tracking is implemented once and never verified again. Templates change, apps update, and events silently stop firing — often discovered months later when someone questions a number. A periodic audit against real transactions is unglamorous and necessary.
Reporting is built on averages across dissimilar things. A site-wide conversion rate blends new and returning, mobile and desktop, branded and non-branded traffic, and moves for reasons nobody can isolate. Segment first, average second.
Dashboards are built to display rather than to prompt. If a chart has never caused anyone to do anything differently, it is decoration. Delete it — the clutter has a real cost in attention.
A reasonable standard
You do not need a perfect implementation. You need one where the funnel is correct, the failure points are visible, the segments you care about are separable, and someone can answer “did that change work” within a day. That is a much smaller build than most teams assume, and considerably more useful than what they currently have.