How to Do Conversion Rate Optimization: A Step-by-Step Workflow
You do conversion rate optimization by working a repeatable loop, not by guessing at button colors: set a measurable goal and baseline, gather quantitative and qualitative data to find real friction, turn that friction into testable hypotheses, prioritize them, run valid A/B tests, analyze results honestly, and feed what you learn back into the next round. The workflow below is the same one we walk students through in the SEO University Conversion Rate Optimization track, and it’s the sequence we still run on client sites at Salterra.
None of this requires a huge team. A small site with steady traffic can run this loop manually with a spreadsheet, GA4, and a free heatmap tool. What matters is doing the steps in order and resisting the urge to skip straight to “let’s just test a new headline.”
1. Set Your Goal and Establish a Baseline
Step 1: Define the one metric that matters
Pick a single primary conversion event for the page or funnel you’re optimizing — a purchase, a lead form submission, a demo booking, an email signup. Resist tracking five “conversions” at once early on; it splits your sample size and muddies every test result. Set this up as a proper event in GA4, not just a pageview proxy, so you’re measuring the action itself.
Step 2: Record your current baseline
Before you touch anything, document the existing conversion rate, average order value (if ecommerce), and traffic volume for the page or funnel over a representative period — at least two to four weeks, longer if your traffic is seasonal. This baseline is what every future test gets measured against, and it’s also what tells you whether your sample size is realistic for testing at all. A page getting 200 visits a month is not going to produce a statistically valid A/B test in any reasonable timeframe; plan for sequential testing or bigger structural changes instead.
2. Gather Quantitative Data
Step 3: Audit the funnel in GA4
Build a funnel exploration in GA4 for the path you care about — landing page to product page to cart to checkout to confirmation, for example. Look for the step with the steepest drop-off. That’s your first suspect, not the homepage, not the whole site. CRO works best when it’s targeted at the biggest leak, not spread evenly across every page.
Step 4: Segment before you conclude anything
Break the funnel down by device, traffic source, and new versus returning visitors. A “bad” conversion rate is often a mobile experience dragging down a healthy desktop number, or a paid traffic segment mismatched to the landing page. Segmenting first prevents you from optimizing for the wrong audience.
3. Gather Qualitative Data
Step 5: Watch heatmaps and click maps
Install a heatmap and scroll-map tool (Hotjar, Microsoft Clarity, or similar) on the pages in your target funnel. Look for dead clicks on non-clickable elements, rage clicks, and how far people actually scroll before abandoning. If your key CTA sits below where most visitors stop scrolling, you’ve found a hypothesis without needing a single test yet.
Step 6: Watch session recordings
Sample 20-30 session recordings of visitors who didn’t convert, filtered to the funnel step you identified in Step 3. Look for hesitation patterns: repeated back-and-forth between pages, form fields people abandon or re-enter, or scrolling back up to re-check price or shipping cost. Recordings surface friction that heatmaps alone won’t show, especially around form usability.
Step 7: Ask visitors and customers directly
Run an on-page exit-intent or post-purchase survey with one or two open questions — “What almost stopped you from completing this?” or “What were you looking for that you couldn’t find?” Pair that with a handful of short customer interviews if you can get them. Survey data is where a lot of the best hypotheses come from, because it tells you the “why” behind numbers heatmaps and recordings can only show you the “what” of.
4. Turn Friction Into Hypotheses
Step 8: Write hypotheses in a fixed format
Force every idea into the same structure: “Because we observed [data], we believe [change] will cause [effect] for [audience], and we’ll know this is true when we see [metric] move.” This format kills vague ideas fast — if you can’t fill in the “because we observed” blank with real data from Steps 3-7, it’s an opinion, not a hypothesis, and it goes to the bottom of the list.
Step 9: Pool hypotheses in a shared backlog
Keep every hypothesis in one running list — a spreadsheet is fine — with columns for the supporting data source, the page or funnel step it applies to, and an estimated effort to build. This backlog is your CRO roadmap; treat it the way you’d treat an SEO content calendar. Ideas don’t get run the moment someone thinks of them.
5. Prioritize What to Test First
Step 10: Score with ICE or PIE
Run every backlog hypothesis through a scoring framework so prioritization isn’t a debate. ICE (Impact, Confidence, Ease, each scored 1-10) is fast and good for smaller teams. PIE (Potential, Importance, Ease) works well when you’re prioritizing across many pages rather than many ideas on one page. Neither framework is precise science — their real value is forcing a consistent conversation instead of testing whatever the loudest stakeholder wants tested this week.
Step 11: Sequence high-traffic pages first
All else equal, test on your highest-traffic, highest-intent pages first. They reach statistical significance faster and the win, once validated, compounds against more visitors. A clever test on a page with trickle traffic is a slow way to learn anything.
6. Run the Test
Step 12: Choose the right test type
Use a standard A/B test when you’re testing one variable against a control with enough traffic to reach significance in a reasonable window. Use multivariate testing only when traffic is high and you genuinely need to test combinations of elements. For low-traffic pages, a before/after sequential test with a longer observation window is often more honest than forcing an underpowered split test.
Step 13: Calculate sample size before you launch
Use your testing tool’s built-in sample size calculator before launch, based on your baseline conversion rate and the minimum lift you’d care about detecting. Committing to that number up front is what stops you from peeking at day three and calling a winner too early.
Step 14: Let it run to statistical significance — and full business cycles
Don’t stop a test the moment it crosses 95% significance if it hasn’t also run through at least one full weekly cycle, and ideally a full billing or purchasing cycle for your business. Traffic behaves differently on weekends, paydays, and around promotions. Stopping early on a lucky Tuesday is the single most common way CRO programs fool themselves.
7. Analyze the Results
Step 15: Check significance and practical impact together
A statistically significant result isn’t automatically a meaningful one. A variant that lifts conversion by 0.3% with high confidence may not be worth the engineering cost of shipping it permanently. Weigh statistical confidence against the actual revenue or lead volume impact before declaring a win.
Step 16: Segment the results the same way you segmented the problem
Break the test results down by the same segments you used in Step 4 — device, traffic source, new versus returning. A test that “loses” overall sometimes wins decisively on mobile, which tells you to ship a mobile-only change rather than discard the idea entirely.
8. Iterate and Build Institutional Knowledge
Step 17: Document every test, win or lose
Log the hypothesis, result, segment breakdown, and your interpretation in a shared test log, regardless of outcome. Losing tests aren’t wasted — they rule out an explanation and often point toward the next, better hypothesis. Teams that only document wins keep re-testing ideas they already disproved.
Step 18: Roll the loop forward
Ship the winning variant, update your baseline, and pull the next hypothesis off the backlog. CRO isn’t a project with an end date — it’s a standing loop that runs continuously against your highest-value pages, the same way you’d continuously refresh content for search.
A Note on the AI-Search Era
As more discovery traffic arrives via AI-powered search summaries rather than classic ten-blue-links results, visitors who click through tend to be further along and more intent-driven — which changes what “friction” looks like. Check whether these visitors expect a more direct answer or comparison above the fold than your page currently gives them; that’s often a faster win than a broad redesign.
Frequently Asked Questions
How long should an A/B test run before I call a winner?
Run it until you hit your pre-calculated sample size and it has completed at least one full weekly cycle, ideally a full purchase or billing cycle. Stopping early because a result looks good on day two or three is the most common source of false winners in CRO programs.
What's the difference between ICE and PIE prioritization?
ICE scores Impact, Confidence, and Ease for individual hypotheses and is fast to apply to a backlog of ideas. PIE scores Potential, Importance, and Ease and is better suited to prioritizing across many different pages or funnel steps rather than many ideas on a single page. Either is fine as long as you use one consistently.
Do I need a big testing tool budget to start CRO?
No. GA4 for funnel data, a free-tier heatmap and recording tool, and a simple on-page survey will surface most of your early hypotheses at no cost. Paid A/B testing platforms become worth it once you have enough traffic to run frequent, well-powered tests.
What if my site doesn't get enough traffic for a valid A/B test?
Lean harder on the qualitative steps — heatmaps, recordings, and surveys — to build strong hypotheses, then make the change directly and evaluate with a longer before/after comparison rather than a split test. Structural, higher-confidence changes are a better use of low traffic than a test that will never reach significance.
Should I test one element at a time or redesign whole pages?
Both have a place. Single-element tests (a headline, a CTA, a form field) isolate cause and effect cleanly and are ideal when traffic is moderate. Full-page redesigns move faster but make it harder to know which specific change drove the result — treat them as a bigger bet informed by everything your qualitative research already told you, not a replacement for the hypothesis process.
How does CRO relate to SEO work?
SEO earns the visit; CRO decides what that visit is worth. A page that ranks well but converts poorly is leaving the value of that ranking on the table, which is why we treat conversion work as a standing part of any content or landing page strategy rather than a one-off project tacked on at the end.
Terry has 30+ years in software and SEO. He’s the founder of Salterra Digital Services and SEO Spring Training, host of the Roundtable SEO Mastermind, and lead instructor at SEO University — teaching the exact tactics his team uses on client work.
Continue learning
Ready to master this?
This guide is one lesson from the Conversion Rate Optimization course. Get every lesson, framework and checklist — plus the full 38-course catalog — inside SEO University.
Practitioner-focused training across the full digital marketing stack — from technical SEO to conversion optimization and the AI search era. By Salterra Digital Services, since 2011.