Research · April 2026
Research methods that actually find wins
A research phase is meant to generate hypotheses that win in test. Not "insights". Not "personas". Hypotheses that, when you ship them to half your traffic, move the revenue number.
After running a few hundred tests across ecommerce and online education, some methods pull their weight. Some don’t. Here’s the ranking.
1. Email surveys of recent buyers
The single most productive hour of CRO research you can do. One open question: "What almost stopped you from buying?" Sent to buyers within 48 hours of purchase, ideally with a discount code as an incentive.
Why it works: recency. The objections that almost cost you the sale are still in their head. They type them in their own words. Those words become the counter-objections on the next PDP.
2. Session recordings, but only of converters
Most people watch non-converter recordings. This is a waste of time. Non-converters leave for a thousand reasons, most of them unreadable from a video.
Converters are different. Watching a converter tells you what they had to push through to buy. Where they hesitated. Which page they re-read. Which tab they opened (a comparison site, almost always). That’s your hypothesis list.
3. On-site exit surveys
A single-question popup on exit intent for non-converters: "What stopped you from checking out today?" Free-form answer, no dropdown.
The response rate is low: 1-3%. Doesn’t matter. Even 200 answers a month on a top-of-funnel page will cluster into five themes. Those themes are your next quarter’s tests.
4. Review mining
Only useful if you have 500+ reviews. But when you do, it’s gold. Tag each review for the top 2-3 things the customer mentions. Count the mentions. The top five are your value propositions. The five negative mentions are your friction points.
This one is almost entirely replaceable by AI now, which tips the ROI further in its favour.
5. Heatmaps (clicks only)
Useful for one thing: discovering which elements people try to click that aren’t actually clickable. Almost every site has 2-3 of these. Every one is a free win.
Don’t waste time on scroll maps. They mostly tell you the fold is the fold.
6. User testing (moderated)
Last on the list because it’s expensive and the returns are diminishing. One round of 5 users at the start of an engagement is worth the money. A second round rarely is.
When it does earn its keep: testing a new concept before you build it, where the cost of building the wrong thing is high.
What we rarely use
- Persona exercises. They generate vibes, not tests.
- Analytics dashboards built from scratch. The dashboard tells you where the leak is. It does not tell you why. You still need surveys and recordings for that.
- Competitor audits. Useful for context, not for hypotheses. Your competitor might be getting their own thing wrong.
The rule of thumb
If a research method doesn’t have a line of sight to a specific test you could run next week, stop doing it. Research should be shortening the list of things to try, not padding it.
If your research phase has produced a deck instead of a test queue, we should talk. Book a 15-minute call or see the full loop.