CRO Research: How to Find What to Test Before You Run a Single Experiment

Most teams start their A/B testing programme the same way — with a list of ideas. Someone saw a competitor do something interesting. A designer has a hunch about the hero image. The CEO wants to test the button colour. And so the tests begin, without anyone asking the more important question first: why would this change the way visitors behave?

Testing without research is expensive in time and traffic. It produces results, but rarely insight. CRO research is the step that comes before the test — the structured process of understanding what's actually happening on your site, where visitors are struggling, and why. Done well, it makes every test you run more likely to matter.

What CRO research is — and why most teams skip it

CRO research is the process of gathering evidence about how visitors behave on your website before forming a hypothesis or designing a test. It answers the questions analytics alone can't: not just where people are dropping off, but why they're dropping off, and what change would be worth testing as a result.

The reason most teams skip it is that research takes time and testing feels like progress. Running an experiment produces a number. Research produces questions — which can feel less satisfying, especially when there's pressure to ship tests and show results.

But the teams that build a research practice before they build a test backlog tend to run fewer, better tests. Their hypotheses are grounded in actual observed behaviour rather than opinion, which means the tests they run are more likely to produce signal worth acting on.

CXL's ResearchXL framework is the most widely referenced structured approach to CRO research — an eight-step process that moves from technical analysis through to hypothesis formation, designed to be completed before a single test goes live.

The five CRO research methods

CRO research draws on five core methods, broadly split into two categories. Quantitative research tells you the what and the where — qualitative research tells you the why and the how. A strong research process uses both.

1. Analytics audit

The analytics audit is where most CRO research starts. The goal is to identify the pages and funnel steps with the most friction — not by assumption, but by data.

CXL's conversion research guide describes this as finding the "leaks" in your funnel: pages with high exit rates, steps where conversion drops significantly, segments that behave differently by device, source, or browser. These are the areas where a test is most likely to make a measurable difference.

What you're building at this stage is a prioritised list of problem areas — not hypotheses yet, just observations about where the numbers point.

2. Heuristic evaluation

A heuristic evaluation is an expert review of your pages against a set of conversion and UX principles. It doesn't require user data — it's a structured way of looking at your own site and identifying friction, confusion, and missed opportunities that a trained eye can spot.

CXL's guide to heuristic analysis for CRO positions it as a faster pre-testing research step, specifically designed to generate hypotheses before running experiments. Common frameworks evaluate pages for relevance, clarity, friction, distraction, and anxiety — each of which points to a different type of test.

The advantage of heuristic evaluation is speed. It can be done in a few hours on any page and produces a useful list of potential test ideas without requiring traffic data.

3. Session recordings and heatmaps

Where analytics tells you where people drop off, session recordings show you what actually happened before they did. Contentsquare's guide to session replay describes this as capturing real user behaviour — clicks, scrolls, hesitations, rage clicks, error encounters — to reveal the friction that quantitative data only hints at.

A visitor who rage-clicks a non-linked element is looking for something that isn't there. A visitor who scrolls past the CTA without clicking might not have read it. A visitor who abandons a form halfway through encountered something that made them stop. These are the kinds of observations that turn an analytics finding into a testable hypothesis.

Heatmaps — click maps, scroll maps, and move maps — sit alongside session recordings as a qualitative layer on top of your analytics data. They don't tell you why something is happening, but they show you patterns across many sessions at once.

4. On-page surveys

Surveys are the most direct form of CRO research: you ask visitors what's stopping them. Contentsquare's guide to on-site surveys explains that on-page and exit-intent surveys capture voice-of-customer data that no analytics tool can surface — the specific objections, confusions, and hesitations visitors have in the moment they're experiencing them.

Questions like "what almost stopped you from completing your purchase?" or "was there anything you couldn't find?" surface the kind of insight that a well-designed test can directly address. CXL's voice-of-customer research guide also notes that survey responses often reveal the exact language your visitors use to describe their problems — which is as useful for copywriting tests as it is for UX changes.

5. User testing

User testing means observing real people attempting tasks on your site while thinking aloud — the most direct way to watch where friction occurs in real time. CXL's user testing guide describes it as the method that surfaces usability problems and conversion blockers that analytics, recordings, and surveys all miss: the things people don't notice enough to report, but that visibly slow them down or cause them to give up.

Even a small number of user testing sessions — five to eight participants — can reveal patterns that no amount of quantitative data would have pointed to.

How to turn CRO research into test hypotheses

The output of CRO research isn't a list of changes — it's a list of observations. The step from observation to hypothesis is where most teams either get it right or waste their test.

A well-formed hypothesis connects a specific observation to a predicted outcome and a reason. CXL's hypothesis formation guide credits the following structure to CRO practitioner Craig Sullivan: "We believe that doing [A] for people [B] will make outcome [C] happen. We'll know this when we see data [D] and feedback [E]."

What this structure forces you to do is articulate not just what you're changing, but who it will affect, what you expect to happen, and how you'll know if it worked. A hypothesis without all of those elements is an assumption with a test attached to it — which is a different thing entirely.

The strongest hypotheses come from research that triangulates across methods. When an analytics drop-off, a session recording showing confusion, and a survey response all point at the same moment in the user journey, you have a high-confidence starting point for a test. When only one data source is pointing at a problem, you have a hypothesis worth forming but one that needs more scrutiny before you commit traffic to testing it.

Building a test backlog from CRO research

Once you have a set of research-backed hypotheses, the next step is prioritisation — deciding which tests to run first based on the potential impact of the change, the confidence you have in the hypothesis, and the effort required to build the test.

Frameworks like ICE (Impact, Confidence, Effort) and PIE (Potential, Importance, Ease) are designed for exactly this. They give you a structured way to score and rank hypotheses so your testing programme works through the highest-value opportunities first, rather than defaulting to whatever is easiest to build. (We covered all three frameworks side-by-side in our test prioritization frameworks guide.)

The research you've done should feed directly into both the Impact and Confidence scores. A hypothesis backed by three converging data sources deserves a higher confidence score than one backed by a single observation.

Ready to start testing your research-backed hypotheses? Try Mida free — no credit card required.

FAQs

Q: What is CRO research?A: The structured process of gathering evidence about visitor behaviour on your website before running A/B tests. It combines quantitative methods (analytics, heatmaps) and qualitative methods (session recordings, surveys, user testing) to identify where and why visitors struggle — so that tests are based on evidence rather than assumption.

Q: Why do CRO research before testing?A: Testing without research produces results but rarely insight. Research ensures the hypotheses you test are grounded in actual observed behaviour, which makes tests more likely to produce meaningful signal and compounds learning over time.

Q: What are the main CRO research methods?A: Analytics audit (to find funnel drop-off and high-exit pages), heuristic evaluation (expert review against conversion principles), session recordings and heatmaps (to see how visitors actually behave), on-page surveys (to capture voice-of-customer data), and user testing (to observe real people completing tasks on your site).

Q: How do you turn CRO research into a hypothesis?A: A well-formed hypothesis connects a specific research observation to a predicted outcome and a reason. A useful structure is: "We believe that doing [A] for people [B] will make outcome [C] happen. We'll know this when we see data [D]." The strongest hypotheses triangulate across multiple research methods pointing at the same friction point.

Q: How much CRO research do you need before testing?A: Enough to form a hypothesis you can defend with evidence. For most teams, a combination of an analytics audit, a heuristic review of the target page, and at least one qualitative method (recordings or surveys) provides sufficient grounding to run a well-justified test.

Sources

• CXL — ResearchXL: The Most Complete Conversion Research Framework
• CXL — Google Analytics for Conversion Optimisation
• CXL — Heuristic Analysis for CRO
• CXL — Quantitative vs Qualitative Research in CRO
• CXL — Voice of Customer Research
• CXL — User Testing for CRO
• CXL — From Data to Test Hypotheses
• Contentsquare — Session Replay
• Contentsquare — On-Site Surveys