Conversion Rate Optimization Guide (2026): The Process, the Traffic Math, and a Real GA4 Audit

Most conversion rate optimization guides are written by companies that sell testing software, so they assume you have the traffic to test. Most sites do not. This refreshed conversion rate optimization guide keeps the process. Additionally, it adds the arithmetic that decides whether A/B testing is even available to you, and it shows what to do when it is not. It also runs the first step on Anobee's own Google Analytics account. The previous version of this page recommended a measurement setup that, it turned out, this site had not got right either.

A note on what changed. The earlier version carried five internal-link placeholders that were never filled in and an "advanced tools" section with no tools in it. It also carried an invented framework and a checkout case study that cannot be sourced. All of that is gone. Instead, every figure below is from Anobee's own analytics, from a named study, or a calculation you can repeat.

What conversion rate optimization is

A conversion is any action you have decided matters: a purchase, a trial signup, a form submission, a newsletter subscription, a click on an affiliate link. The conversion rate is the number of conversions divided by the number of visitors, times 100. Sessions work too, as long as you pick one denominator and stick to it. A page with 5,000 visitors and 150 form submissions converts at 3%.

The number means nothing on its own. It moves with traffic source, device, season and how narrowly you define the conversion. Therefore a rate from another site in another industry is not a target. The useful comparison is your own rate for the same page and the same source last month or last quarter. Conversion rate optimization is the discipline of moving that number up on purpose: measuring where visitors drop out, finding out why, changing one thing, and checking whether the change worked.

It sits between acquisition and revenue. SEO, ads and social bring visitors; CRO decides what those visitors do. The bounce rate guide covers the closely related question of why visitors leave immediately. Meanwhile, this guide covers the visitors who stay and still do not act.

Step 0: check that your conversion is actually being counted

Every CRO guide starts with "collect data". However, almost none say to check that the data is real. This is the step Anobee got wrong, and it is worth showing because the failure is quiet.

On 22 September 2026, the GA4 property for anobee.com had four events marked as key events, which is GA4's term for conversions [2]: contact_success, newsletter_success, form_submit and purchase. The admin page showed "No stream data detected" beside three of the four. newsletter_success and contact_success had never fired in the last 28 days; only form_submit had. Meanwhile, the Events report for the 90 days to 21 September showed the events the site actually sends. Those were form_start (51), form_submit (39), cta_click (8), newsletter_submit (5, from 2 users), contact (5) and submit_lead_form (5). The newsletter conversion this site cares about most fires as newsletter_submit. However, that event had never been marked as key. The key-event report was counting a name that does not exist.

The consequence is not dramatic. Nothing broke. The Home dashboard simply showed "Key events: 0" for weeks, and a person glancing at it would conclude the newsletter form had never converted. It had, five times. The fix takes a minute: star the event that fires, unstar the ones that do not. However, finding it took a side-by-side read of two admin screens that nobody normally opens together.

The general rule: before optimising anything, open the event list. Confirm the conversion you plan to improve appears there with a non-zero count, under the exact name you are reporting on. GA4 will let you mark any event as key, including one that never occurs [2]. The Google Analytics setup guide and the Google Tag Manager setup guide cover getting the events in; this step is about checking they arrived.

The conversion rate optimization process in six steps

CRO is a loop, not a project. Each pass produces the hypothesis for the next.

1. Collect data. Quantitative first: a GA4 funnel or path report showing where visitors drop between steps. Then qualitative: session recordings, heatmaps, a short on-page survey, and, if at all possible, five real people trying to complete the action while you watch.
2. Form a hypothesis. Not "make the button bigger". A hypothesis names the observation, the change and the expected effect: "Recordings show mobile visitors abandoning at the shipping field; a single-column layout should cut abandonment because they are currently zooming to read it."
3. Prioritise. Score each hypothesis on impact, confidence and ease. Run the high-impact, high-confidence, easy ones first, even when a riskier idea is more interesting.
4. Test. A/B if you have the traffic (see the next section). If not, a structured before-and-after with guardrails.
5. Analyse and implement. A result that reached its planned sample size is a result, whether it won, lost or was flat. Ship winners. Also record losers, so nobody re-tests them next year.
6. Iterate. The biggest leak has usually moved. Therefore, go back to step 1.

The mistake that stalls most programmes is not stopping the tests. It is stopping the locating. Teams keep testing the page they first looked at, instead of finding where the largest drop is now.

The traffic math: can you A/B test at all?

An A/B test can only detect a difference if enough visitors pass through each variant. How many depends on the baseline conversion rate and on the smallest lift you care about. The table uses the standard two-proportion test at 95% significance and 80% power. Evan Miller's calculator [5] uses a more conservative method and returns somewhat larger numbers. Accordingly, treat these as the floor, not the ceiling.

Now apply it to a real small site. Anobee's GA4 recorded 1,051 users in the 90 days to 21 September 2026 and five newsletter submissions, a rate around 0.5%. To detect even a 50% lift in that rate, an A/B test would need about 15,600 users per variant, or 31,000 in total. At the current pace that is roughly seven years of traffic for one test. A 20% lift would take about forty. Consequently, this site cannot A/B test its newsletter form, and no amount of tooling changes that.

Two things follow. First, most "A/B test your CTA colour" advice is not written for sites of this size. Following it produces noise that looks like results. Second, the baseline matters as much as the traffic. A page converting at 10% needs a twentieth of the visitors that a page converting at 0.5% does. That is one reason to test high-intent pages (pricing, checkout, a services page) before low-intent ones.

Conversion rate optimization without A/B tests

If the arithmetic rules out testing, conversion rate optimization does not stop. Instead, it changes method.

Watch sessions instead of counting them. Microsoft Clarity records sessions and builds heatmaps, and states that it is free with no traffic limits [4]. Ten recordings of visitors who reached a form and left tell you more than a hundred conversions' worth of aggregate data. You see the hesitation, the rage clicks and the field that gets retyped. The Microsoft Clarity guide covers setup; the Clarity vs Hotjar comparison covers when the paid tool earns its cost.

Test with five people. Ask five people who match your audience to complete the action while thinking aloud. This is the oldest usability method there is. On a small site it finds the confusing label, the hidden price and the button that looks disabled faster than any dashboard will.

Fix the obvious before measuring anything. Some changes do not need a test because the alternative is plainly worse: a form that errors without saying which field is wrong, a checkout that hides the total, a CTA below three screens of introduction. Ship them, note the date, and move on.

Use before-and-after with guardrails. Change one thing, then compare four weeks after with four weeks before, on the same page and the same traffic source. Only trust the comparison if traffic and its mix stayed roughly stable. This is weaker than a controlled test, and it will occasionally mislead you. Nevertheless, it is better than changing five things at once and remembering none of them.

Borrow other people's research where it is documented. Baymard's checkout research is the clearest example. Across 50 studies the average documented cart abandonment rate is 70.22% [1]. The leading reasons visitors give for abandoning during checkout are extra costs being too high (40%), slow delivery (20%), not trusting the site with card details (19%), being forced to create an account (18%) and a long or complicated checkout (17%) [1]. That list is a prioritised hypothesis backlog for any ecommerce site too small to generate its own.

How to run a CRO audit with free tools

This is the diagnostic pass, in the order that surfaces the biggest problem fastest. Moreover, everything here is free.

1. Confirm events. Open GA4 → Admin → Events. Check that each funnel step has a non-zero count and is marked key where it should be. For ecommerce that is view product, add to cart, begin checkout, purchase; for SaaS, view pricing, start trial, complete signup. Step 0 above.
2. Find the biggest drop. In GA4, build a funnel exploration across those steps for the last 30 to 90 days. The largest percentage drop between two adjacent steps is where you start, regardless of which page anyone on the team dislikes.
3. Check the traffic mix. Conversion rate differs sharply by source. In Anobee's 28 days to 21 September, chatgpt.com / ai-assistant was the second-largest first-user source (40 of 357 users), ahead of Google organic (34). Visitors arriving from an AI assistant have usually read a summary of the page already. Therefore a long introduction serves them badly. Segment before you judge.
4. Record the problem step. Install Clarity on that page and wait for 100 to 200 sessions. Then watch 10 to 15 recordings of visitors who reached the step and did not continue.
5. Read the heatmap. Compare where attention goes with where the design assumed it would go. A CTA nobody hovers over is a placement problem, not a copy problem.
6. Write one hypothesis. Specific observation, specific change, expected effect.
7. Decide the method. Use the traffic table. Test if you can; before-and-after with guardrails if you cannot.

Landing pages and forms: the changes that keep earning their place

These are not laws. They are the changes that repeatedly survive testing on sites with enough traffic to test. Consequently, they are reasonable defaults for sites without.

• Message match. The headline should restate the promise of the ad, the search result or the AI summary that brought the visitor. A page that greets a specific query with a generic brand line loses the visitor before the first scroll.
• One primary action. Secondary links can exist, but there should be no doubt about the one thing the page wants.
• Proof at the point of doubt. Testimonials in the hero are seen by everyone and persuade nobody, because no question has formed yet. Put proof next to the price, the form or the guarantee.
• Total cost early. Baymard's top abandonment reason is unexpected extra costs [1]. Show shipping, tax and fees before the final step, not on it.
• Fewer fields, inline errors. Every field is a filter. Keep the ones that qualify; drop the ones that merely collect. Show errors beside the field, not in a summary at the top.
• Guest checkout. Forced account creation is a documented abandonment reason on its own [1].
• Specific button copy. "Start my free trial" tells the visitor what happens next; "Submit" does not.
• Mobile tap targets and loading states. Small targets cause mis-taps; a button with no feedback gets pressed twice.

Anobee's own services page guide applies these to the one page type most small businesses actually need to convert.

A/B testing done properly

For sites that clear the traffic bar, a few rules separate real results from noise. Similarly, they separate a testing programme from a guessing one.

Calculate the sample size before starting, using the baseline rate and the smallest lift worth acting on [5]. Then run until you reach it, and for at least one full business cycle so weekdays and weekends are both represented.

Do not peek. Early leads reverse. A test that looks decided on day three is often flat on day fourteen. Therefore, fix the stopping rule in advance and keep to it.

Do not change the test mid-run. A copy tweak to one variant resets the comparison.

Read segments after, not instead. A variant can win on desktop and lose on mobile. Look, but do not go hunting for the one segment where a losing test "really" won. Otherwise every losing test becomes a winner somewhere.

Choose a tool that integrates with your analytics. Google Optimize shut down on 30 September 2023 [3]. Google now points users to AB Tasty, Optimizely and VWO, which integrate with GA4, and has opened its APIs so other tools can as well [3]. Also prefer server-side or flicker-free implementations. A testing script that briefly shows the original page before swapping in the variant is both a data problem and a page-speed problem.

Conversion rate optimization by business type

The last row is Anobee's own category, and it is where the Step 0 problem lives. On a content site the "conversion" is small and easy to forget to measure. Consequently, it goes unmeasured. The email list building guide covers the signup side in detail.

Tools: what is free and what is worth paying for

The free column covers the whole diagnostic loop of conversion rate optimization. However, the paid column only starts to matter once a page has thousands of conversions a year, not visitors, to test against.

Common conversion rate optimization mistakes

• Optimising an unmeasured conversion. See Step 0. If the key event is zero, nothing you change can be evaluated.
• Testing without the traffic. Running A/B tests on a page that will never reach sample size. Then acting on whichever variant happened to lead when the deadline came.
• Copying benchmarks. Treating another industry's average as your target.
• Adding urgency before removing friction. Countdown timers on a checkout that hides the shipping cost.
• Judging on clicks instead of the conversion. A variant that lifts CTA clicks and lowers purchases is a loss.
• Redesigning instead of testing. A full redesign discards everything that was working along with everything that was not. Afterwards, there is no way to tell which was which.
• Never writing it down. Without a test log, the same idea gets re-tested every eighteen months by whoever joined most recently.

Bottom line on conversion rate optimization

Conversion rate optimization is measurement first, research second, testing third, and the order matters more than the tools. Confirm the conversion is being counted; Anobee's was not. Find the largest drop in the funnel and watch real people hit it. Also do the arithmetic before promising an A/B test, because at typical small-site volumes the test cannot see the answer. Then change one specific thing, note the date, and compare honestly. The sites that improve are rarely the ones with the most tests. Ultimately, they are the ones that stopped guessing.