Qualitative research: heatmaps, recordings and user testing

0:00 Qualitative research

Analytics told you that sixty percent of mobile visitors leave the product page. Now what? You could guess why. Or you could watch. Qualitative research is how you see behaviour and hear reasons: heatmaps, session recordings, user tests and first-impression tests. In this lecture you'll learn to use each method well, how AI summaries fit in, how to protect privacy, and how to turn observations into findings your team will trust. Then you'll watch me run a focused recording review with a tally sheet.

0:37 Why qual matters

Why does qualitative research matter so much? Because it's where most good hypotheses come from. You can't A/B test your way to insight if every variant is a guess. Watching people struggle reveals things no dashboard shows: a button that looks disabled, a delivery date hidden in a tab, a size chart that opens in a new window and loses the basket. These are specific, fixable problems. And specific problems lead to specific hypotheses, which lead to tests that actually win.

1:12 Heatmaps done well

Heatmaps first. Click and tap maps show where people click, including on things that aren't links, which tells you what they want. Scroll maps show how far people get. Three rules. Segment by device, always, because mobile and desktop layouts differ completely. Collect enough views, a few hundred per device as a sensible minimum. And look for surprises: clicks on images that can't zoom, clicks on text people expect to expand, important information below where most people stop scrolling.

1:46 Recordings with a purpose

Session recordings next. Watching random recordings wastes hours. Always filter to answer a question: mobile sessions from paid social that viewed a product but didn't add to cart. Watch twenty to thirty, and stop when you stop learning new things. Look for rage clicks, repeated taps on the same element. U-turns, going back and forth between pages. Long pauses at prices or form fields. And error messages. Keep a tally sheet so patterns become counts, not anecdotes.

2:19 The 2026 toolkit

The tools have changed. Microsoft Clarity offers free heatmaps and recordings, with AI-generated session summaries. Hotjar, now part of Contentsquare, combines behaviour with on-page surveys. Enterprise platforms like Contentsquare and FullStory add journey analysis and frustration scoring. And product analytics tools like PostHog include session replay alongside experiments. Most now offer AI summaries and automatic frustration signals. They save hours. But summaries can miss context, so always watch a sample of the underlying sessions before acting.

2:52 User testing

User testing is the highest-value method. Watch five to eight people who match your customers attempt realistic tasks while thinking aloud. Give scenarios, not instructions. Bad: click the pricing page. Good: you want to train a team of ten; find out what it would cost. Probe gently: what are you thinking now? What did you expect to happen? Never lead: isn't that button clear? Record issues with severity, blocker, major or minor, and frequency, like five of six participants. And first-impression tests: show the page for five seconds, then ask what it offers and who it's for.

3:34 Simple example: Riyadh dental booking

A simple example. A dental clinic group in Riyadh sees many visitors reach the booking page but few book. Recordings filtered to booking page exits show people tapping clinic names that aren't clickable, and scrolling up and down. Six user tests reveal why. People can't tell which branch is closest, and the calendar shows no availability until you choose a dentist. Findings: blocker, no availability visible until dentist chosen, five of six. Major, branch location unclear, four of six. These become hypotheses in the next module.

4:11 Moderated vs unmoderated

Let's talk about moderated versus unmoderated testing, because you'll choose between them often. In moderated tests, you're present, in person or on a video call, and you can ask follow-up questions when someone hesitates. You get depth, but each session takes an hour of your time. In unmoderated tests, participants complete tasks on their own and record their screen and voice through a testing platform. You get results in hours, from more people, but you can't ask why in the moment, so tasks must be crystal clear. A practical pattern: use moderated sessions to explore a problem you don't understand yet, and unmoderated sessions to check a specific fix at speed.

4:59 Realistic example: Manchester fashion (illustrative)

Now a realistic scenario with illustrative details. A fashion brand in Manchester found a leak between product view and add to cart on mobile. The team filters recordings to mobile, paid social, product pages, no add to cart, and sessions longer than twenty seconds. They watch twenty-five, using a tally sheet. Eleven tap the size selector repeatedly, then leave. Nine scroll to the footer looking for delivery information. Seven try to pinch-zoom images. Four rage-click a greyed-out add button, because no size is selected and nothing says so. An on-page poll asking what's stopping you shows the same themes: fit and delivery. Two methods, one story. That's a finding worth acting on.

5:48 Watch me do it: focused recording review

Watch me run the review. I write the question at the top of my sheet: why do mobile visitors from paid social leave product pages without adding to cart? I set the filter in my recording tool. I read the AI summary first, which says users struggle with size selection. Useful, but I don't trust it yet. I open sessions one by one, note each pattern in a row, and add a count and example session IDs. After about fifteen sessions, no new patterns appear. I watch ten more to confirm counts, then stop. Finally I mark severity. The greyed-out button with no explanation might be a blocker, so I flag it for the team straight away.

6:39 Privacy in behavioural tools

Privacy matters. Behavioural tools collect data about real people. Mask all text inputs by default, and never capture payment fields. Where the law requires it, like under UK and EU cookie rules, load the tool only after consent. Mention it in your privacy notice and keep retention short. Restrict who can view recordings, and don't paste recordings or transcripts into general AI tools. Good privacy practice isn't just compliance. It's respect for the people whose behaviour is teaching you.

7:13 Observations → findings

Turning observations into findings. User scrolled up and down is an observation. A finding says what you saw, how often, where, the likely reason, and severity. For example: on mobile product pages, eleven of twenty-five sessions tapped the size selector repeatedly then left; likely cause, no fit guidance; severity, major. Give it an ID in your research log. And prefer findings supported by two methods, like recordings and polls. That triangulation turns anecdotes into evidence your team will act on.

7:48 Common mistakes

Common mistakes. Unsegmented heatmaps. Watching recordings without a question. Trusting AI summaries without watching sessions. Testing with colleagues or friends. Leading participants. Recording personal data without masking or consent. And stopping at observations without writing findings. Each of these wastes time or produces conclusions you can't defend.

8:08 Recap

Recap. Qualitative research explains why. Use heatmaps segmented by device, recordings filtered to a question, and user tests with real customers and realistic tasks. Use AI summaries to save time, but verify with real sessions. Protect privacy. And turn observations into findings with counts, severity and triangulation. Try this now: pick the biggest leak from your funnel diagnosis, write one question, filter twenty-five recordings, fill the tally sheet from the lesson text, and add one on-page poll to confirm or challenge what you see.