The biggest mistake in CRO is jumping straight to testing without knowing what to test. Research is not optional. It is the multiplier on everything else. A team that tests random ideas will occasionally land a winner. A team that researches first will land winners consistently - and understand why they won.
CRO research divides cleanly into two types: quantitative (what is happening) and qualitative (why it is happening). You need both. Quantitative methods find the problem. Qualitative methods explain it.
Quantitative methods: the 'what'
- Funnel analysis: track the specific steps from landing to conversion and find where the biggest drop-offs occur. For example, if 60% of visitors who add to cart abandon at the shipping page, that's your first test.
- Heatmaps: show where users click, move, and focus attention. Often reveals ignored CTAs, misleading non-clickable elements, or above-the-fold elements getting zero engagement.
- Scroll maps: show what percentage of users see each section of a page. If your CTA is below where 60% of users stop scrolling, that's a significant and easily fixable finding.
- Session recordings: watch anonymised real user sessions. A single afternoon of recordings can surface friction patterns that no analytics report would show - hesitation, back-navigation, zooming in on confusing text.
- Form analytics: identify which form fields cause the most abandonment. Long forms kill conversions; field-level analytics tell you which ones are doing the killing.
Qualitative methods: the 'why'
- On-site surveys: exit-intent surveys ('What stopped you completing this today?') and post-conversion surveys ('What almost stopped you?') are high-signal and low-cost. Even a 2% response rate yields powerful insight at scale.
- User interviews: 5–8 interviews with real customers or prospects will surface recurring objections and motivations that no amount of analytics can. The same concerns will emerge again and again - those are your test hypotheses.
- Usability testing: give participants a task ('Find and buy a blue medium t-shirt') and watch them attempt it. Where they hesitate or fail is your test backlog.
- Customer support analysis: your support team sits on a goldmine of conversion intelligence. The questions customers ask before purchasing reveal exactly what the website failed to communicate. Mine your support tickets before running any test.
"A useful CRO insight can come from reading support conversations and finding the question the homepage did not answer."Crow Editorial Team
Editorial context: Crow editorial team, methodology and corrections
Combining quant and qual
Neither type of research is sufficient alone. Quantitative methods tell you that 70% of users abandon the checkout at the payment step. But they can't tell you whether the abandonment is because the page is slow, the form is confusing, the trust signals are missing, or the shipping cost was a surprise. That's what qualitative research answers.
A practical research sequence for most teams:
Start with funnel data
Identify the pages and steps with the highest drop-off rates. These are your research priorities - the places where improving conversion would have the biggest impact.
Add heatmaps and session recordings to the problem pages
Understand the behavioural pattern. Are users clicking on non-links? Ignoring the CTA? Rage-clicking a broken element? Reading a section and immediately leaving?
Run a targeted survey
On the specific problem page, deploy an exit-intent survey. Ask one question: 'What stopped you completing this today?' Even 50 responses will surface a pattern.
Interview 5–8 recent customers
Ask them to walk you through the decision. What almost stopped them? What questions did they have? What made them finally click? This is where hypotheses are born.
Write your hypothesis
Structure it as: 'Because we observed [X], we believe changing [Y] will improve [Z] for [audience].' A hypothesis with this structure is testable. Vague ideas aren't.
The research-to-hypothesis ratio
A well-researched CRO programme generates 1 high-confidence hypothesis for every 2–3 hours of research. An under-researched one generates 10 low-confidence guesses in the same time. The guesses lose more often - and when they win, nobody knows why.