Session recordings and heatmaps: what they show and what they do not

Reports say what, recordings say why

Analytics tells you eighty per cent leave this page. That is a fact, and on its own it is useless: no edit follows from it.

A session recording shows how they leave: the person scrolled to the form twice, poked at a non-clickable element, went back up, hunted for a price and closed the tab. From that an edit follows on its own.

This is a division of labour between two tools, not a competition. Reports find the place, recordings explain the cause. Trying to find the place through recordings costs a week; trying to explain the cause through reports means inventing one.

What follows is how to watch recordings so it takes half an hour a week rather than becoming a box set. And where recordings mislead.

What they are and how they differ from reports

A session recording is a replay of what happened inside the browser window: cursor movement, scrolling, clicks, typing into fields, moves between pages.

It is not screen video and it is not surveillance of a person. Interface events are recorded and a replay is assembled from them afterwards. No camera, no microphone, nothing from other tabs.

A click map is derived from the same data: every click over a period, laid over a single snapshot of the page. One page, a thousand visitors, all their activity at once.

The tools have different names in different systems — session replay, heatmaps, visitor recordings — but the principle is the same everywhere, and everything below works regardless of which one you use.

The key difference from reports: a report aggregates and loses the particulars, a recording keeps the particulars and does not aggregate. So you cannot count with recordings and you cannot understand with reports.

Where to start: do not watch everything

The beginner's mistake is switching recordings on and watching them in order. Within an hour it is dull; within a day the tool is abandoned.

Recordings are watched by hypothesis, not in sequence. First a report shows a problem area, then you select the recordings that contain it.

The order is this: find the page with the worst figure among the ones that matter, filter recordings to it, and watch five to seven in a row. Half an hour.

Watch at increased speed and skip the pauses. What matters takes seconds; the rest is reading, which is visible at speed anyway.

And write observations down as you watch rather than afterwards. Five recordings later you will not remember which one held the hesitation.

Which recordings to watch first

Five filters that return more than the rest.

Sessions where the person reached the form but did not submit. The most valuable set: the intent was there and something got in the way.

Sessions longer than three minutes with no action at all. Usually either reading or somebody searching and not finding.

Sessions from mobile devices. If more than half your visits come from phones and you watch on a computer, half your problems are invisible to you.

Sessions from the search queries you care about. That is how you test the match between what somebody searched for and what they saw.

What recordings show that reports cannot

A list of the standard findings that surface within the first half hour almost every time.

Clicks on things that are not clickable. A heading that looks like a link, an image that looks like a gallery, a phone number that does not dial when tapped. Every such click is a small disappointment.

Scrolling past the important part. The price block sits exactly where people reached and turned back. A report will only tell you the page was viewed.

Ping-ponging between sections. Somebody circling two or three pages means the answer is on none of them while they are sure it exists.

Hesitation on a form field. Five seconds in a phone field means the format was unclear, or they are deciding whether to give the number at all.

And repeated submissions. If somebody presses the button three times, they never saw a confirmation, and you will get three enquiries or none.

Click maps: three kinds

Different systems name them differently, but there are exactly three, and they answer different questions.

The click map — where people pressed. It answers what people take for interactive and what they ignore.

The scroll map — how far people got. It answers whether anybody saw your main block at all. The answer is usually worse than expected.

The attention or cursor map — where people lingered. The most indirect of the three: the link between cursor and gaze exists but is weak, and conclusions from it need the most caution.

The practical order: scroll map first, click map second, cursor map last and carefully.

The main mistake in reading a click map

A bright patch on the map does not mean the element works. It means people pressed it.

Clicks come from confusion too. A red patch on a non-clickable image is not a success but a signal that people expect a link there and do not get one.

And the reverse: a cold zone around a button does not always mean the button is bad. They may simply never scroll that far, which shows on the scroll map rather than the click map.

So the two maps are read together. A click map taken alone regularly leads to edits that change nothing, because they fix the symptom.

How many recordings support a conclusion

A question rarely answered honestly, because the honest answer is inconvenient.

Recordings are a qualitative tool, not a quantitative one. They do not prove, they suggest. Five people tripping over the same thing is enough reason to examine that place, not enough to claim everybody does it.

A practical rule: if the same behaviour appears in three recordings out of ten, it deserves attention. If in one of ten, you have looked at an exception.

And the conclusion has to be checked against a report. Recordings gave the hypothesis; a report shows the scale — how many people a month behave that way and what it costs.

The reverse order — an edit based on one recording, then surprise that nothing changed — is the most common waste of time with this tool.

What must not be recorded

A short section, and the most expensive one to skip.

The contents of fields holding personal data must not reach recordings. Passwords, card numbers, identity documents, message text — all of it is masked on the site's side rather than in the service's settings.

Most systems mask password fields by default and do not mask arbitrary text fields. Mark those manually — it is one attribute in the markup.

Check the result: open one of your own recordings where you filled in a form and confirm asterisks appear instead of the text you typed.

And tell visitors you collect behavioural data in the site's privacy policy. Exactly what to write there and which consents your jurisdiction requires is a question for a lawyer, not for a developer.

Turning an observation into an edit

Observation without action is entertainment. A short procedure that gets to a result.

Write the observation as one sentence in the form people do X because they expect Y. For example: they click the card heading because they expect a link to the details.

Check the scale in a report: how many such people a month.

State the edit in one sentence and name in advance what should change in the numbers. Not improve the card but make the heading a link; visits to the detail page should rise.

And return to the recordings two weeks after the edit. Half of the changes that look obvious produce new behaviour you did not expect.

When recordings are useless

Three situations where the tool will not help and the time is wasted.

Too little traffic. Twenty sessions a month give no material for observation: you will watch random people and draw random conclusions.

The problem is at the entrance, not on the page. If the wrong people find you, recordings will show the wrong people leaving. True, but useless: the queries are what needs fixing.

No readiness to change anything. A tool that surfaces problems nobody intends to solve works as a source of irritation and gets switched off within a month.