Reading heatmaps and session recordings: what to look for in user behaviour
Heatmaps show what happened, not why. Value appears only when maps and recordings are read together against a specific hypothesis.
Four maps and what each one shows
The click map shows where people tapped. The main finding is usually not where they hit buttons but where they hit non-clickable things: gallery images, headings, numbers in a price table. That is a direct signal of expectation.
The scroll map shows how far down the page people get. Look for a sharp cliff: if 60 % leave before the pricing block, something above it is losing interest or convincing them they already know enough.
The link map counts navigation separately and often reveals that half the menu is never used. Form analytics show which field people stall on and abandon.
Reading the click map without errors
First error: looking at a site-wide map. Look device by device and segment by segment — behaviour from mobile ads and from desktop organic differs enough that averaging is meaningless.
Second: treating many clicks as success. Repeated clicks on one element often mean it is not responding — the person taps three times because nothing happens. That is a bug, not engagement.
Third: ignoring clicks into nothing. If people keep tapping a product photo they expect a zoom. If they tap a line in a price table they expect it to expand. Each such hotspot is a ready-made development ticket.
The scroll map and page structure
Look beyond the percentage who reach the bottom: examine where attention lingers. Metrica shows average time spent on each band, which tells you which block deserves to move up.
A frequent finding on local projects: the pricing or payment terms block sits at the very bottom while the scroll cliff falls in the middle. Moving that block up usually beats any amount of copy editing.
Mobile and desktop scroll maps differ radically. On mobile the page is two or three times longer in screens, and a block visible immediately on desktop ends up on the third screen.
Session recordings without wasting a day
Never watch recordings sequentially. Always filter by segment: bounced sessions on a service page, sessions over two minutes with no lead, sessions that started the form but never submitted. Each segment answers a different question.
Fifteen to twenty recordings from one segment is usually enough — patterns repeat fast. If nothing emerges after twenty, the segment was chosen badly rather than the problem being absent.
What to watch for: where the cursor or finger stalls, where someone scrolls up and down over the same spot looking for something, where they close abruptly. Returning repeatedly to one block almost always means information is missing there.
Turning findings into changes
State each finding as a hypothesis with an expected effect: 'people click portfolio photos expecting a zoom — add a lightbox, expect deeper browsing and longer time on page'.
Prioritise by implementation cost and proximity to revenue. A fix on the hero or in the form almost always beats one in the 'about us' block, because it sits in everyone's path rather than a fraction's.
And do not ship fifteen edits at once if you want to know what worked. On modest traffic you must trade speed against certainty — usually it is wiser to change one or two bottlenecks and watch the result for a month.
Limits of these tools
Recordings are not captured when blockers are active, so part of the audience never appears. Maps also average: on a responsive page where blocks reflow, the overlay on the layout can be imprecise.
Conversion data inside the maps is secondary — they show behaviour, not outcome. A decision should rest on the chain: the map showed an anomaly, a recording explained it, the goal number confirmed the effect.
The main limit is volume. On a page with fifty visits a month any map will draw a pretty blob that means nothing. Below a few hundred visits, talking to customers directly beats reading maps.