Reading heatmaps without inventing intent

A red blob on a heatmap is a location, not a motive. Teams still narrate it as “users are confused by the CTA” because confusion is a story that licenses a redesign. Conversion drop-off diagnostics require a colder reading: density without a hypothesis is just temperature.

Sample before you stare

In Session Replay Review Practice we cap the sample. Twenty recordings for a claimed drop, drawn across device and locale, then stop. If you watch until you find the rage click that matches the story you walked in with, you will find it. That is not evidence. That is auditioning anecdotes.

Tag only what you can point at: overlay covering a control, keyboard obscuring submit, error string in the wrong language, latency above a threshold you defined in advance. “Looks hesitant” is not a tag.

Heatmaps lie about scrolling cultures

Thai-language checkout copy is often longer than the English mock the design system was built on. A heatmap that was calibrated on the English mock will show “nobody reaches the submit button” when the button has simply been pushed below a legal paragraph. The diagnostic is to measure button exposure by locale, not to add a pulsing arrow.

The same blob can mean hit-target failure on a small Android OEM and genuine comparison shopping on iOS. Split the heatmap. If you cannot split it, you do not yet have a diagnostic tool. You have a poster.

Fold replay back into the funnel claim

Replay never replaces the leak map. It can only support or weaken a sentence you already wrote. “Android 12 WebView users drop after the error string introduced in design-system v4” can be checked with a handful of recordings. “Users don’t trust us” cannot.

If your vendor’s replay coverage is thin, say so. Pim’s review of our evening room was fair: the first week feels empty until instrumentation exists. Do not fake coverage with three recordings and a confident workshop. Fix the pipe, then look.

More on how we run that room sits on the courses page. The method essay is on drop-off diagnostics.