A conversion rate optimization process has six steps: research why visitors don't buy, turn what you find into a written hypothesis, prioritize the ideas, A/B test the best ones, analyze the results honestly, then roll out the winners and record every outcome. Repeat the loop. Most of the gains come from getting the research right and from refusing to call a result before the data supports it.
This guide walks through each step with real tests from the LiftArchive test database, so you can see what each stage looks like on actual ecommerce stores. If you're new to the topic, start with our CRO overview.
The six steps at a glance
| Step | Main question | Output |
|---|---|---|
| Research | Where and why do visitors drop off? | A list of problems backed by data |
| Hypothesis | What change should fix the problem, and how will we measure it? | A written hypothesis for each idea |
| Prioritize | Which ideas should run first? | A scored, ranked test backlog |
| Test | Does the change beat the control? | A live A/B test with a fixed sample size |
| Analyze | What really happened, and for whom? | A result with confidence level and segments |
| Roll out and document | What do we ship and what did we learn? | A shipped change and a test log entry |
Think of it as a loop, not a straight line. Each analysis feeds new research, and the test log gets more useful the longer you keep it.
Step 1: Research what is stopping people from buying
Research tells you where the problem is and why it happens. Skip it and you end up testing button colors because someone saw it in a blog post.
Good research mixes numbers and words. Analytics shows you where people drop off. Heatmaps and session recordings show you what they do on the page. Surveys, reviews and support tickets tell you what they were thinking.
- Analytics: funnel drop-off by page type and device, conversion rate by traffic source.
- Behavior tools: heatmaps, scroll maps and recordings on your key pages.
- Customer voice: post-purchase surveys, exit surveys, review mining and support logs.
- Expert review: a structured CRO audit of each step of the funnel.
Research pays off in the test results. Dave Diederen changed the above-the-fold USPs to the points heatmaps showed buyers engaging with and saw +6.23% CVR. In another test, Dave Diederen placed a countdown timer where heatmaps showed shoppers were focused, which brought +20% conversion rate.
Survey data works the same way. Oliver Kenyon added three survey-based reassurance lines above a collection page grid, worth $44,890/month. Kasper Dolk added benefit bullets to the hero that answered survey objections, which produced $763,679.14 in extra profit per year.
For a deeper look at survey and interview methods, see our guide to customer research for CRO.
Step 2: Write a hypothesis for each idea
A hypothesis turns a research finding into a test you can judge. It links the problem you saw, the change you'll make and the result you expect.
A simple format works well:
Because we saw [research finding], we believe [change] for [audience] will improve [metric]. We'll know it worked if [primary metric] moves by [amount] after [sample size or duration].
Kasper Dolk's survey test shows the pattern. The finding was that shoppers had specific objections. The change was to answer them in hero bullets. The metric was profit. That clear chain makes the result easy to interpret and easy to apply to other pages.
Write the hypothesis before you build anything. It stops you from moving the goalposts after the test ends. Our guide to writing A/B test hypotheses has more templates and examples.
Step 3: Prioritize the backlog
Research usually produces more ideas than you can test. Prioritization decides what runs first, based on likely impact, confidence in the evidence and the effort to build it.
Frameworks like PIE (potential, importance, ease) and ICE (impact, confidence, ease) give each idea a score. They're rough, but they make the trade-offs visible and keep the loudest opinion from winning by default.
A few rules of thumb help:
- Favor pages with the most traffic and revenue. A small lift on the product page often beats a big lift on a page few people see.
- Give higher confidence to ideas backed by more than one research source.
- Look at what has worked for similar stores. Browse product page tests or checkout tests to see which changes have a track record.
- Check traffic before you commit. If a page can't reach a useful sample in a few weeks, the idea needs a bigger change or a different method.
Our A/B test prioritization guide compares the main scoring models in detail.
Step 4: Run the test
Now build the variant and run it as a controlled A/B test. Decide the primary metric, the sample size and the run time before launch, and don't change them mid-test.
- Pick one primary metric. Revenue per visitor is often a better choice than conversion rate for stores, because it captures order value too.
- Size the test up front. Use the sample size calculator to work out how many visitors you need.
- Run full weeks. Weekday and weekend shoppers behave differently, so run at least one or two full business cycles.
- QA every variant. Check it on mobile, desktop and the main browsers before you send traffic to it.
You'll need a testing tool that can split traffic and track revenue. Convert is one option that works on Shopify and other platforms. As of November 2026, its pricing page lists the Growth plan at $399/month billed monthly or $299/month billed annually, with 100K tested users per month and a 15-day free trial that doesn't need a credit card. That price is a lot for a small store with modest traffic, so it suits brands that already have enough visitors to run tests regularly.
Tools mentioned here
You only need one A/B testing tool to copy this result. Here's where we'd start.
Some of these are affiliate links. We only list tools practitioners actually use to run tests like these.
Step 5: Analyze the results honestly
Analysis is where many CRO programs go wrong. The goal is to find out what really happened, not to find a winner.
Don't stop early. Ruben tested an authority badge like a doctor or expert endorsement. After week one it looked like +17%. After four weeks it was +1.66% CVR, which was flat. The early spike was noise.
Check whether a result holds up in other contexts. The same idea can win, lose or do nothing depending on the store:
- Dave Diederen added a countdown timer to checkout and saw +3.16% RPV (+$139,029/month est.).
- On another store, the same change lost with -7.1% RPV.
- For a European market, a countdown timer lost with -35.61% RPV.
- Ruben tested a countdown timer across several brands: +~10% CVR on a low-priced brand, flat on a high-priced brand and -9% CVR on another.
Look at more than one metric, too. Dave Diederen introduced subscriptions with the option pre-checked by default. The result was -9.17% RPV upfront and +1,295% subscription share. Whether that's a win depends on customer lifetime value, which a short test window doesn't show.
Before you call any result, check it with a significance calculator and read our guide to statistical significance.
Want results like this on your store?
Dave Diederen at Syntra Labs. Dave builds and tests Shopify stores at Syntra Labs, and shares most results in public.
Step 6: Roll out the winner and document everything
When a variant wins, build it properly into the site rather than leaving it running in the testing tool. Then watch the metric for a few weeks after launch to make sure the lift holds.
Losing and flat tests are just as useful as winners. They tell you what not to build and stop the next person from repeating the test. Dave Diederen removed an email popup and lost -20.51% RPV, which is a strong reason to keep it. Another of Dave Diederen's tests, skipping the cart and sending shoppers straight to checkout, lost -10.23% CVR.
Record each test in a shared log:
- The research finding and hypothesis
- Screenshots of control and variant
- Dates, traffic, sample size and primary metric
- The result, confidence level and any segment differences
- What you learned and what to test next
Learnings feed the next round of research. Kasper Dolk's work shows how a pattern can travel: an outcome headline with a tagline handling the #1 objection at the top of the PDP gave about a 10% conversion increase across multiple clients.
Want results like this on your store?
Kasper Dolk at Maison Commerce. Kasper runs Maison Commerce, which builds CRO programs, landing pages and themes for fast-growing ecommerce brands such as ICON and HiStrips.
Common ways the CRO process breaks
- Testing without research. Ideas come from opinions, so the win rate stays low.
- Peeking and stopping early. Early results swing a lot, as the authority badge test showed.
- Copying a winner without context. The countdown timer tests above show how much results vary by store and market.
- Not writing things down. Without a log, the same ideas get tested again a year later.
Our list of common A/B testing mistakes covers more of these.
Questions
What are the steps of the conversion rate optimization process?
The CRO process has six steps: research why visitors don't convert, write a hypothesis for each idea, prioritize the ideas, A/B test the top ones, analyze the results, then roll out winners and document every result. The loop then repeats, with each round of learnings feeding the next round of research.
What is a CRO framework?
A CRO framework is a repeatable structure for running optimization work. It usually covers how you gather research, how you write hypotheses, how you score and prioritize ideas (for example with PIE or ICE), and how you run, analyze and record tests.
How long does one cycle of the CRO process take?
It depends mostly on traffic. Research and building a test can take a week or two, and most tests should run at least one to two full weeks to cover weekday and weekend behavior. Low-traffic stores may need longer tests or bigger changes to reach a reliable result.
Why is research the most important step in CRO?
Research tells you where and why visitors drop off, so you test changes that address real problems. In our database, tests based on heatmap and survey findings, such as Dave Diederen's heatmap-driven USP change (+6.23% CVR), show how research points to useful changes.
What should I do with losing A/B tests?
Record them just like winners. A losing test tells you what not to build and stops someone from repeating it. For example, Dave Diederen's test removing an email popup lost -20.51% RPV, which is a clear reason to keep the popup.