Most A/B tests lose because they test a guess. You change a button color, the lift is flat, and you decide CRO is a scam. The test was never the problem. Nobody did the research first. Conversion research is how you find, with evidence, exactly where people drop and why they hesitate, so every test you run is aimed at a real objection. Do it right and your win rate climbs, your wasted dev time falls, and the changes you ship survive contact with a P&L.
It has two halves. Quantitative data tells you what is happening and where. Qualitative data tells you why. You need both. Analytics shows 60 percent abandon at the shipping step. A one-line poll tells you they left because the shipping cost was a surprise. One without the other is half a hypothesis.
Start with the math, not the heatmap
Before you watch a single session replay, find where money actually leaks. Open analytics, build a funnel for your primary conversion, and find the step with the worst drop relative to its traffic. That is your highest-leverage page. A 20 percent lift on a page 100,000 people see beats a 200 percent lift on a page 400 people see.
- Define the conversion that pays the bills: a purchase, a qualified lead, a trial start. Not a soft metric like time on page.
- Build the funnel in GA4, Mixpanel, or your platform analytics: landing, key intermediate step, conversion.
- Calculate step-to-step rate and absolute drop count. Drop count = visitors entering the step minus visitors completing it.
- Rank pages by drop count times traffic. The top one or two get deep research first.
- Split by device. Mobile usually leaks more than desktop, and it is often a different problem entirely.
Worked example: 10,000 land on product, 4,000 reach checkout (60 percent drop, 6,000 lost), 3,200 hit shipping, 1,600 purchase (50 percent drop at shipping, 1,600 lost). Two steps leak. The product-to-checkout step loses more bodies, so it ranks first. Fix the biggest pool, not the worst-looking percentage.
If your funnel and tracking are a mess, fix that before anything else. You cannot research what you cannot measure. Clean attribution and a clear path is the foundation for a real growth marketing strategy.
Watch real sessions on the leaky pages
Numbers tell you where. Session replays and heatmaps tell you how the page behaves. Hotjar, Microsoft Clarity (free), Lucky Orange, and Contentsquare reconstruct individual visits so you can see scrolling, rage clicks, dead clicks, and form abandonment. Do not watch at random. Filter so you compare winners against losers.
- Watch 10 to 15 sessions that converted and 10 to 15 that abandoned on the same page. Note what differs in behavior.
- Filter for rage clicks and dead clicks. People clicking things that are not buttons reveals a broken mental model.
- Read scroll maps. If your CTA or key proof sits below where 50 percent of people stop, that is a layout fix, not a copy fix.
- Watch form sessions specifically. Which field do people abandon on, refill, or stall at?
- Capture baseline heatmaps before you change anything, so you can prove the change worked later.
Get the words from the people who hesitated
This is the half most teams skip, and it is where the gold is. You cannot infer an objection from a click. You have to ask. The goal is voice of customer: the exact language people use to describe their doubt, their need, and their decision. That language becomes your headlines, your CTAs, and your proof. Run these four sources in parallel.
- On-site poll on the leaky page. One question: "What almost stopped you from buying today?" Trigger on exit intent or after 30 seconds.
- Post-purchase survey to new customers: "What nearly made you not buy?" and "What problem were you trying to solve?" Aim for 100 to 200 responses, where answers start repeating.
- Customer interviews. 5 to 8 recent buyers, 20 minutes each, focused on the moment they decided. Five is enough to surface the most common patterns in qualitative work, per Nielsen Norman Group.
- Mine existing words. Support tickets, sales notes, chat logs, and reviews (yours and competitors'). Copy exact phrases. Do not paraphrase.
You are not inventing arguments. You are collecting the ones your customers already make, then putting them where the hesitation happens.ADGY
The output is a swipe file of real objections and real desire, in customer language. This is the raw material for a value proposition that actually lands and for trust elements that answer doubts before they kill the sale.
Turn findings into ranked, testable hypotheses
Research is worthless until it becomes a prioritized test queue. Write each finding in one structure: because we observed X, we believe changing Y will cause Z, measured by [metric]. Vague hypotheses produce vague tests. Then rank with a scoring framework so you work on the right things first. PIE scores each idea 1 to 10 on Potential (room to improve), Importance (qualified traffic through it), and Ease (effort to ship), then averages. ICE swaps Importance for Confidence. Pick one and stay consistent.
- Do: write the hypothesis before you build the variant. If you cannot state the expected metric move, you are not ready.
- Do: batch low-effort, high-confidence fixes (a confusing label, a missing shipping line) and just ship them. Not everything needs a test.
- Don't: test trivial changes on low-traffic pages. You will never reach significance.
- Don't: stack five changes into one variant. You will not know what worked.
For the structural side of friction, pair this with landing page optimization and clarity and specificity in your copy, where most research-driven wins actually come from.
Test with discipline, then keep what survives
Now run the experiment. The research bought you a sharp hypothesis. Do not waste it with sloppy testing. Rule of thumb: a primary-conversion A/B test needs roughly 1,000 conversions per variant to read a small lift at 95 percent confidence. Micro-conversions like clicks reach significance faster. If your traffic cannot get there in four weeks, test a bigger swing or test higher in the funnel.
- Calculate sample size and duration up front. Run full weeks to cover weekday and weekend behavior.
- Run to your pre-set sample. Never call a test early because it looks like it is winning. Peeking inflates false positives.
- Judge winners on revenue and contribution margin, not conversion rate alone. A higher rate at a lower order value can lose you money.
- Document every test: hypothesis, result, what you learned. Losing tests are research too.
- Re-run research on the next leaky page. CRO is a loop, not a project.
If you want this run as a system rather than a one-off, that is what we do. If you are sitting on traffic that converts below where it should, talk to us. For the testing engine itself, our guide on maximizing performance with testing strategies goes deeper.
The conversion research checklist
- Funnel built, highest drop-off page identified by drop count times traffic.
- 10 to 15 converting and 10 to 15 abandoning sessions watched on that page.
- Heatmaps and scroll maps captured as a baseline.
- On-site poll live on the leaky page; post-purchase survey collecting responses.
- 5 to 8 customer interviews done; support, chat, and reviews mined for exact language.
- Findings written as structured hypotheses and ranked with PIE or ICE.
- Top hypotheses queued; quick wins shipped without a test.
- Sample size and duration set before any A/B test goes live.
- Winners judged on margin, results documented, loop restarted on the next page.
Frequently asked questions
What is the difference between conversion research and CRO?
Conversion research is the diagnosis. CRO is the treatment. Research is the work of finding where people drop off and why, using analytics, session replays, and customer voice. CRO is the broader practice that also includes testing and shipping the fixes. Skip the research and your CRO is just guessing with extra steps.
How much traffic do I need before conversion research is worth it?
Qualitative research works at almost any traffic level: interviews, surveys, and watching sessions need only a trickle. Traffic matters for A/B testing. If you cannot get roughly 1,000 conversions per variant in a month, ship high-confidence fixes from research directly instead of running underpowered tests that will never reach significance.
Which conversion research tool should I start with?
Start free. Microsoft Clarity gives you session replays and heatmaps at no cost, and GA4 covers funnels. Add a simple on-site poll or a tool like Hotjar for voice of customer. Do not buy an expensive stack before you have a research habit. The method matters more than the tooling.
How do I know my research found a real problem and not noise?
Triangulate. A finding you trust shows up in more than one place: the analytics drop, the session behavior, and the customer's own words all point at the same friction. If only one source flags it, treat it as a hypothesis to test, not a fact to ship.