CRO Process: The 7-Step Framework I Use With Every Client

Warning: fileperms(): stat failed for /www/.wordpress/ in /wordpress/wp-admin/includes/file.php on line 2225

Most businesses approach conversion rate optimisation as a series of disconnected experiments. Move a button, change a headline, add a testimonial, hope something works. The result is unpredictable: occasionally a big win, often nothing measurable, frequently a regression. The pattern that actually compounds is a structured process applied to every page that matters.

After a decade running CRO and SEO across ecommerce, local service and national service businesses, the highest-impact work I’ve done has always followed the same seven-step process. The wins are not about button colours or headline tweaks. They come from reducing friction at every step of the funnel, and you only see the friction once you look at the data properly. This guide is how I approach CRO when I take on a new client.

This is the seven-step CRO framework I use with every client, from ecommerce stores doing seven figures to local service businesses with a single contact form. It covers how to audit your baseline, how to prioritise where to focus, how to design tests that actually answer a question, and how to document results so the next test builds on what you already learned. It pairs with our guide to what CRO is and connects to our broader work on SEO and Google Ads.

Why CRO Needs a Process, Not a Tactic List

The reason most CRO efforts produce inconsistent results is they skip straight to tactics. “Add social proof”, “use action verbs in CTAs”, “remove form fields”. Those are sometimes right, but applied without context they are guesses. The right tactic depends on which page, which audience, and which step of the funnel is leaking.

A repeatable process forces you to:

• Look at the data before forming an opinion
• Identify where the actual friction is, not where you assume it is
• Prioritise the changes that have the highest expected impact
• Design tests that produce answerable results
• Document outcomes so future tests build on existing knowledge

Done properly, each cycle of the process compounds. The second test you run is informed by what the first test told you. After six months you have a documented body of evidence about what works on your site specifically, which is far more valuable than industry-wide CRO best practice that may or may not apply to your audience.

Step 1: Baseline Audit (What Is Actually Happening?)

Before you change anything, you need a clear picture of current behaviour. This is the step most businesses skip and it is the reason their CRO efforts feel directionless. The audit pulls from at least five data sources:

Google Analytics 4

Three reports matter most:

• Landing pages by conversion rate — sort by sessions descending, look for the high-traffic pages with the lowest conversion rate. Those are your biggest opportunities.
• Funnel exploration — build a custom funnel from landing page → key page → conversion. The biggest drop-off step is where to look first.
• Conversion events by source/medium — paid traffic and organic traffic often convert very differently. Don’t average them.

Google Search Console

GSC tells you what people search for to find your pages. Compare the query intent to the page content. If you’re ranking for queries the page doesn’t directly answer, that’s a CRO problem disguised as an SEO problem.

Heatmaps and Session Recordings

Heatmaps show you where people click, scroll and pause. Session recordings show you how real users navigate your pages, where they hesitate, what they ignore. Hotjar and Microsoft Clarity are the two I install most often. Clarity is free and good enough for most businesses. Watch ten recordings on your top three landing pages. You will see things in the first viewing you cannot get from analytics.

On-Site Search Queries

The most under-used data in CRO. What people search for inside your store or website is a direct signal of unmet intent. Customers searching for product attributes you don’t have, services you don’t list, or information that isn’t on the page are telling you exactly what’s missing.

Customer Interviews or Survey Data

Quantitative data shows you what is happening. Qualitative data tells you why. Five customer interviews or a single targeted on-page survey often surface insights you would never find in analytics.

Step 2: Identify the Highest-Leverage Pages

You cannot test everything. The right pages to focus on are the ones where small lifts produce the biggest revenue or lead impact. Three categories almost always make the cut:

• Top entry pages. If 30% of all sessions land on your homepage, a 10% conversion lift on the homepage moves more revenue than a 50% lift on a deep blog post.
• Conversion-critical pages. Cart, checkout, contact, booking, quote request. These pages have lower traffic but every visitor is high-intent. Friction here is expensive.
• High-traffic low-CVR pages. Identified in Step 1. These are the pages where existing demand is being wasted.

For most ecommerce stores I prioritise: homepage → top 3 collection pages → cart → checkout → top 5 product pages. For most service businesses: homepage → top 3 service pages → contact form → about page.

Step 3: Form a Hypothesis (Not a Guess)

A real hypothesis has three parts: a specific change, a predicted outcome, and a reason grounded in data.

Weak: “Make the CTA button bigger to increase clicks.”

Strong: “Session recordings show 40% of visitors scroll past the primary CTA without engaging. Adding a sticky CTA in the header should lift click-through to /contact by 15-25% based on similar implementations across our portfolio.”

The strong version forces you to be specific about what you expect and why. When the test runs, you learn something either way: either the hypothesis was right (now you know that pattern works on this audience), or it was wrong (now you know your assumption about the friction was wrong, which is also valuable).

Step 4: Prioritise With ICE or PIE Scoring

By Step 3 you’ll have more hypotheses than you can test in a quarter. You need to rank them. The two common frameworks are ICE and PIE.

ICE scoring:

• Impact (1-10) — how much will this move the needle if it works?
• Confidence (1-10) — how confident are you in the hypothesis based on the data?
• Ease (1-10) — how easy is it to implement?

Score each hypothesis, sum the three, sort descending. The highest scores run first.

PIE scoring (an alternative used by CXL and others) replaces Impact with Potential and adds Importance:

• Potential — how much improvement is possible on this page?
• Importance — how important is this page to the business?
• Ease — how easy is the change to implement?

Either framework is fine. The point is forcing structured comparison rather than picking what feels exciting.

Step 5: Design and Build the Test

For each hypothesis, decide the test type:

• A/B test. Two variants run simultaneously, traffic split 50/50. Best for isolating the impact of a single change.
• Multivariate test. Multiple elements tested in combination. Requires significantly more traffic to reach statistical significance. Use sparingly.
• Sequential test. Variant replaces control entirely for a defined period, then results compared to the prior period. Less rigorous statistically but useful when traffic volume is too low for true A/B testing.
• Painted door test. Add the new feature as a non-functional element (button, link, banner) and measure click-through to gauge demand before building the real thing.

Tools vary by platform. VWO and Optimizely are the established platforms. Shopify stores can use native split testing in the Shopify admin or third-party apps. Google Optimize was the free option but was deprecated in 2023.

Step 6: Run the Test With Statistical Rigour

Two rules that protect you from drawing wrong conclusions:

• Decide your sample size in advance. Use a sample size calculator (most testing tools include one) to determine how many sessions or conversions each variant needs to reach 95% confidence. Don’t peek and call the test early when you see an early lead.
• Run for at least one full business cycle. Minimum two weeks for most businesses, often four weeks. Conversion behaviour varies by day of week and time of month. A test run for three days during a promotion will mislead you.

Statistical significance matters but it is not the whole story. A test that hits 95% confidence with a 1% lift on a small sample may still be noise. Look at both the p-value and the magnitude of the effect.

Step 7: Analyse, Document, Iterate

After the test ends, three things happen:

• Decision. Implement the winner permanently, kill the loser, or run a follow-up if the result was inconclusive.
• Documentation. Record what was tested, why, what the result was, and what you learned. A simple shared spreadsheet works. After 12-24 tests you have a documented body of evidence about what works on your site.
• Next hypothesis. Each test informs the next. A win on a homepage CTA hierarchy test suggests the same pattern might lift other pages. A loss suggests your hypothesis about the audience was off and the next test should investigate why.

The compounding gain from a documented CRO program over 12 months is significant. The compounding loss from a series of un-documented experiments is also significant: the same wrong test gets run twice because nobody remembered the first attempt.

Jobs-To-Be-Done Customer Interviews

The most under-used CRO research method is the Jobs-To-Be-Done interview. JTBD asks not “what do customers want” but “what job did they hire your product to do, and what was happening in their life when they decided they needed it”. The framework comes from Clayton Christensen’s Harvard Business Review work and reframes customer research around the moment of switching.

Five to eight JTBD interviews with recent customers (within the last 30-60 days, while the decision is fresh) usually surfaces the real motivations behind your conversions and the friction that almost stopped them. The output reshapes hypotheses for the rest of the CRO process, because you’re optimising the funnel around what customers were actually trying to accomplish, not what you assumed.

Expert / Heuristic Review

Before user testing, run an expert heuristic review against established usability principles. Nielsen Norman Group’s 10 usability heuristics are the standard reference: visibility of system status, match between system and real world, user control and freedom, consistency, error prevention, recognition rather than recall, flexibility, aesthetic and minimalist design, help users recognise errors, and help and documentation.

An expert review by someone trained in these heuristics produces a list of specific friction points faster than waiting for quantitative data to surface them. For most businesses this is a 4-8 hour engagement that informs the first quarter of CRO testing.

Competitive Analysis as a CRO Input

Competitive analysis in CRO is different from SEO competitor analysis. The question isn’t “what keywords are they ranking for” but “what does their conversion funnel look like and where are they reducing friction better than we are”. Audit the top 3-5 competitor sites by walking through them as a customer: how fast is their site, how clear is their value proposition, how short is their checkout, what trust signals do they lead with, what payment methods do they offer. Document patterns that show up across multiple competitors — those are usually genuine best practices, not noise.

Usability Testing

Separate from heatmaps and session recordings, formal usability testing puts a real person in front of your site and asks them to complete a specific task (e.g. “find an engagement ring under $5,000 and add it to your cart”). You watch where they get stuck, what confuses them, and what they say out loud.

Five users is enough to find the majority of usability issues, per Nielsen’s well-known research. Moderated testing (you guide the session) produces deeper insight; unmoderated platforms like UserTesting are faster and cheaper. One round of usability testing on a new design before launch saves more rework cost than it spends.

The CRO Formulas Worth Tracking

CRO is measured in more than just conversion rate. The full set of metrics that matter:

• Conversion rate = conversions / sessions
• Revenue per visitor (RPV) = total revenue / total visitors (a lifted CVR with a lower AOV can still hurt RPV)
• Cost per lead (CPL) = total ad spend / total leads (CRO lifts directly lower CPL)
• Cost per acquisition (CPA) = total ad spend / total customers
• Lead-to-customer close rate = customers / leads (a lead-CVR lift means nothing if those leads don’t close)
• Return on test investment (ROTI) = (revenue lift from the test) / (cost of running the test)

Track all of these per quarter. Pure conversion-rate-only reporting hides whether your CRO program is making money.

Personalisation: When to Move Beyond A/B Testing

Personalisation goes beyond A/B testing by serving different experiences to different audience segments based on behavioural, demographic or contextual data. It’s the right move once you’ve exhausted the obvious A/B test wins and you have enough traffic to support segment-level experiences.

Common personalisation moves: returning-customer messaging different to first-time visitors, geo-specific shipping messaging, audience-specific landing pages for paid campaigns, behavioural triggers (cart abandoners, repeat browsers). Tools like Dynamic Yield, Mutiny and Optimizely Personalize handle this at scale. Personalisation is usually 6-12 months into a mature CRO program, not month one.

Post-Conversion Retention (The Overlooked Lever)

Most CRO programs stop at the conversion event. The highest-leverage extension is the post-conversion experience: thank-you page upsells, abandoned-cart recovery, onboarding sequences, repeat-purchase nudges. For ecommerce, post-purchase upsells regularly add 5-15% to average order value with minimal effort. For service businesses, the equivalent is the lead-to-customer nurture sequence after the form submission.

How AI Overviews Are Changing CRO Math

Since Google rolled out AI Overviews more widely in 2025-2026, the click-through rate on organic results has shifted. Sites whose content gets cited inside an AI Overview often see higher-quality (lower-volume) clicks because the user has already read a summary and clicked specifically to dig deeper. Sites whose content doesn’t appear in the AI Overview at all see reduced impression-to-click ratios.

The CRO implication: the conversion rate of organic traffic in 2026 should be measured against a lower-but-better-qualified visitor pool. Average conversion rate per session may rise even as raw click volume falls. See our guide to schema markup for AI search for how to position your content for AI Overview citations.

Sharing Learnings Across the Organisation

The compounding gain from CRO programs comes not from the tests themselves but from the documented learnings being absorbed by the whole team. The patterns that emerge after 12-24 documented tests inform product decisions, marketing copy, paid campaign landing pages, email design, and even sales conversations. Most CRO programs lose 60-70% of their potential value because the learnings stay locked in the CRO person’s head.

The practical fix: a shared CRO log accessible to product, marketing and sales; a monthly summary of completed tests; explicit knowledge-transfer sessions when a test result has implications outside CRO. The agencies and teams that run the most effective CRO programs treat the documentation as more valuable than any individual test result.

Common Mistakes That Sabotage CRO Programs

The patterns I see most often on businesses with stalled or directionless CRO efforts:

• Testing too small. Two new button colours on a low-traffic page. The test will never reach significance. Save that energy for a meaningful change on a high-traffic page.
• Testing without a hypothesis. “Let’s try a different layout and see what happens.” If you can’t articulate what you expect and why, you’ll find a way to read the result however you want.
• Calling tests early. Three days in, the variant is +12%. Declare victory, ship it, watch the lift evaporate over the next month. Pre-decide your sample size and respect it.
• Ignoring qualitative data. The numbers tell you what is happening. Recordings and surveys tell you why. Both inputs together produce sharper hypotheses than either alone.
• Treating CRO as a one-off project. The compounding gain comes from sustained iteration. A single round of CRO is usually less valuable than a year of small, documented improvements.
• Optimising for vanity metrics. Click-through to the next step is not the same as conversion. Lifting CTA clicks but tanking checkout completion is a loss disguised as a win. Always measure to the actual revenue or lead outcome.

The Tools I Actually Use

Listed in order of how often I install them on new client engagements:

• Microsoft Clarity — free, generous, good enough for heatmaps and session recordings on most businesses.
• Google Analytics 4 — for funnel analysis, landing page CVR, conversion event tracking.
• Google Search Console — for query-to-page intent comparison.
• Hotjar — when Clarity isn’t enough or the client already has it.
• VWO or Optimizely — when running formal A/B tests at scale.
• Shopify native split testing or theme-level variant testing for ecom stores.

I cover these in more depth in CRO Tools: What to Use to Run Conversion Tests and Heatmaps.

Want Help Running CRO on Your Business?

If you have an ecommerce store or service business and want a fresh set of eyes on your conversion funnel, we offer a free first-pass CRO audit. We look at your analytics, run a quick heatmap, identify the highest-leverage opportunities, and send back a prioritised list of recommended tests.

Request your free CRO audit →

Frequently Asked Questions

What is a CRO process?

A CRO process is a repeatable framework for identifying, prioritising, designing and measuring conversion rate optimisation experiments. The seven-step version I use covers baseline audit, page prioritisation, hypothesis formation, scoring, test design, running with statistical rigour, and documented iteration. The process exists to replace ad-hoc tactic-led experimentation with a structured approach that compounds over time.

How long does CRO take to show results?

Most well-designed tests need two to four weeks to reach statistical significance, depending on traffic volume and the size of the effect. Significant compounding gains from a documented CRO program typically appear over three to six months as multiple tests stack up. Expect 12 months for a mature program to produce its largest cumulative lifts.

What is the ICE framework for CRO?

ICE is a prioritisation framework that scores each test idea on Impact (how much it could move the needle), Confidence (how confident you are in the hypothesis based on data), and Ease (how easy it is to implement). Each is scored 1-10, summed, and sorted descending. Highest scores test first. ICE forces structured comparison rather than picking ideas based on excitement.

How much traffic do I need for A/B testing?

Most testing tools include a sample size calculator. For a typical A/B test detecting a 10% lift with 95% confidence on a baseline conversion rate of 3%, you need roughly 10,000-30,000 sessions per variant. Lower-traffic sites can still run CRO via sequential tests, painted door tests, and qualitative research, but formal A/B testing usually needs more traffic than small businesses have.

Should I test button colours?

Probably not as your first test. Button colour changes occasionally produce measurable lifts but they rarely have outsized impact compared to higher-leverage changes (CTA placement, headline clarity, form length, social proof position, page load speed). Save button colour experiments for after you’ve optimised the higher-impact elements.

What is the difference between A/B testing and multivariate testing?

A/B testing compares two variants of a single change (e.g. headline A vs headline B). Multivariate testing tests multiple changes in combination (e.g. two headlines × two CTAs × two images = 8 variants). Multivariate requires significantly more traffic to reach significance because the sample is split across more variants. For most businesses, A/B testing produces faster, more reliable results.