CRO research methods: how to find out why visitors do not convert

The two questions research answers

Every method answers one of two questions: what are people doing, and why. Nielsen Norman Group, the usability research firm, describes this as the split between behavioural methods (what people do) and attitudinal methods (what people say), and between quantitative (how many) and qualitative (why, and how to fix it).

Neither side is enough alone, and conversion rate optimization depends on using both. Numbers tell you that 70% of visitors leave a page, and they cannot tell you the headline confused them. Interviews tell you why one person hesitated, and cannot tell you how common it is. Combining them is the point.

NN/g’s 2022 guide “When to Use Which User-Experience Research Methods” lists 20 methods and notes that most projects benefit from combining several. For conversion work, six cover almost every case, and they are below.

Analytics: find where the problem is

Start with your analytics tool, such as Google Analytics. NN/g classes analytics as a quantitative method that analyses recorded behaviour like clicks and form fills, and it requires that tracking was set up in advance. That last point matters: check it is correct before you trust it.

Set up your main action as a key event so it is counted. Then compare conversion rate by traffic source, device and landing page. Look for the biggest gap between a page’s traffic and its results. A mobile rate far below desktop, or one paid campaign converting at half the rate of the others, tells you where to look next.

Build a simple funnel: the steps from landing to completion, with the number of people at each. The step with the steepest drop is where to spend research time. Analytics cannot explain the drop. It tells you where to point the other methods.

Heatmaps and session recordings: watch what people do

A heatmap shows, in colour, where visitors click, tap or scroll on a page. A session recording replays one visitor’s visit, including their cursor, scrolling and clicks. Free tools, such as Microsoft Clarity, offer both. This is a vendor product, and the principle applies to any similar tool.

Use heatmaps to answer narrow questions. Do visitors scroll past your offer? Do they click an image that is not a link? Do they reach the form? Use recordings to see the story behind the number: a visitor tapping a button that does nothing, scrolling back and forth, or leaving at the form.

Treat both as pointers. A heatmap is an average and hides the individual, and a recording shows what happened, not why. Watch recordings of one page and one segment at a time. Twenty to thirty sessions is a practical starting point, though no source fixes that number. Stop when you see the same problems repeating.

Surveys and feedback: ask why

Surveys measure attitudes through mostly closed questions, and NN/g notes that an intercept survey is one triggered while someone is using the site. On a conversion page the most useful versions are short: one question on the thank-you page, such as “What almost stopped you from contacting us?”, and one on exit, such as “Is there anything you could not find?”.

Ask customers who converted why they chose you and what they compared you with. Their words often make better headlines than yours, and they show which objection nearly won.

Remember the limits. Surveys capture what people say, and what people say and do can differ. Self-selection also matters: those who answer are not a random sample. Use survey answers as hypotheses to check against behaviour.

User testing: see confusion happen

In a usability test you give a real person a task, such as “find out what it costs to hire us and request a quote”, and watch them try. You say as little as possible. You learn where they stall, what they misread and what they expected to find.

Jakob Nielsen’s widely cited 2000 article “Why You Only Need to Test with 5 Users” argues that five users find about 85% of usability problems, based on a model in which each user finds on average 31% of them. He recommends several small rounds with fixes between, not one large study. The 85% figure is an estimate, not a guarantee.

The same article notes the limits. If your audience splits into distinct groups, test each group, and quantitative studies need more people, around 20 for usability metrics. For a conversion page, five people matching your real customers, in a remote session you can record, is a sensible start.

Heuristic review: a fast expert check

A heuristic evaluation is an inspection of a page against a list of recognised usability principles. The most used list is Jakob Nielsen’s 10 usability heuristics, developed with Rolf Molich in 1990 and refined in 1994 through a factor analysis of 249 usability problems. They include visibility of system status, error prevention and consistency.

NN/g’s 2023 guide recommends three to five evaluators, each judging the same interface independently for about one to two hours, then comparing notes. A single reviewer will miss issues.

This method is quick and cheap, which makes it a good first step. Its own guide is blunt about the limit: heuristic evaluations cannot replace user research. A broken heuristic is a likely problem, not a proven one.

How to combine them

A sensible order for a single page is below. Each step narrows the question for the next.

• Analytics: choose the page and segment with the most lost value.
• Heuristic review: list likely problems in an hour or two, before spending on anything else.
• Heatmaps and recordings: see which of those problems real visitors actually hit.
• Survey: ask converters and leavers what nearly stopped them.
• User tests: watch five people attempt the key task, and note where they stall.
• Write findings as hypotheses and rank them by how many visitors they affect. Then change, and test if traffic allows. See conversion rate math for the traffic needed.

Research when traffic is low

Small sites cannot run a reliable A/B test, which is exactly when qualitative research pays off most. Five user tests need no traffic at all. Recordings of 100 visits still teach you a great deal, and a one-question survey needs only a few replies to show a pattern.

Be careful with privacy. Mask form fields and sensitive pages in recording tools, and follow the consent rules that apply where your visitors live. When in doubt, ask your developer or legal adviser.

Whatever you learn, check it against the message that brought visitors in. Many problems found in research are really message match problems: the page does not continue the conversation the visitor started. The overview in what is CRO shows where research fits in the loop.