Analytics and Behaviour Tools for CRO: The Working Guide

A useful measurement stack for a small business has three layers and no more. Analytics counts the funnel: sessions, product views, adds, checkout starts and orders, split by device and source. Behaviour tools show the experience behind those counts: recordings, heatmaps and frustration signals on the pages the counts indict. Questions close the loop: short surveys and a post-purchase question asking buyers what nearly stopped them. Each layer answers a different kind of question, and none can substitute for the others. Set them up to answer the decisions you actually face, keep the reports to a handful you open on a rhythm, and let everything else go uncollected. This bench is the map for that stack.

Analytics answers where and how much: where the funnel loses people, how a segment compares with itself last quarter, whether a change coincided with a movement. Recordings and heatmaps answer what it looked like: the hesitation at a field, the scrolling for a delivery cost, the button nobody notices. Surveys answer why, in the visitor's own words, with all the tidiness and bias that self-reporting carries. Confusing the layers produces the classic frustrations: staring at a dashboard demanding it explain a drop it can only locate, or watching recordings with no funnel to tell you which page deserves watching. Work the layers in order, counts first, behaviour second, words third, and most questions resolve before they become arguments.

The setup guide in this bench, analytics setup that answers questions, starts from the decisions: what will you do differently if this number moves? Events, goals and reports are then built to serve those decisions and no others. The failure mode it prevents is the default install, which collects hundreds of numbers nobody has a use for, while the three counts the business actually steers by, its step rates by device and source, are missing or wrong. A small stack configured deliberately beats a large stack configured by default on every measure that matters: accuracy, comprehensibility and the chance its owner opens it next month.

Most analytics value for a small business sits in a short, stable set of views: the funnel, the source table, the device split, the top landing pages read as answers, and the site search report where one exists. The bench names them in the five reports to open monthly, and adds the deeper views, funnel reports and a one-page dashboard, for the weekly routine. The principle is rhythm over depth. Five reports opened monthly for a year teach the business its own patterns; fifty reports opened once teach nothing except that analytics is overwhelming, which is a lesson about report design, not about the business.

Recordings and heatmaps earn their place after the counts have chosen a page. The bench covers them in session recordings and heatmaps, including the frustration signals in Microsoft Clarity's documentation, such as repeated rapid clicking, and the discipline of counting patterns across sessions rather than narrating single dramatic ones. Heatmaps suit questions about attention on a page; recordings suit questions about sequence and struggle; neither estimates an effect size. Choosing between recordings, heatmaps and surveys is a real decision with different costs and biases, compared directly in choosing behaviour tools. The common error is buying all three before the funnel has asked a question any of them can answer.

A measurement stack can be neatly set up and still lie: double-counted conversions, test orders inside the revenue, consent choices quietly changing who is measured, a tag firing twice after a theme update. The bench devotes a full guide to tracking traps because a wrong baseline does more damage than no baseline; it manufactures confidence in a fiction. The consent dimension is covered in consent, privacy and measurement: measurement must be lawful and honest about its own gaps, especially when a share of visitors declines and disappears from the counts. Build verification into the routine: place test orders, check the numbers land once, and re-verify after every platform or theme change.

Illustration of the whole bench working as one system. Monday's funnel review shows mobile checkout completion has fallen while other steps hold. The week's recordings on the checkout step show repeated hesitation at a newly added field. A handful of post-purchase answers name the same field as confusing. The decision writes itself: simplify the field, log the change, watch the step. Thirty minutes, three layers, one decision. That routine, formalised in from data to decision, is what the tools are for. A stack that does not end in a weekly decision is not a measurement system; it is a subscription to other people's behaviour, delivered as charts.