How to build a CRO strategy

A CRO strategy is your plan for what to test, in what order, and how to judge each result. Without 1, testing turns into a pile of ideas that all feel urgent. With 1, every test answers a question the last test raised.

Most guides teach the same loop: research, hypothesis, test, learn. This guide shows that loop with real results from our published A/B tests, including a lift that cost revenue, a big swing visitors rejected, and a win built on earlier wins. Below are the 8 steps we use to build a CRO plan, plus a checklist you can use on your own site.

A CRO strategy is a ranked list of tests tied to 1 revenue goal

If you are new to the topic, start with What is CRO in marketing? for the full definition. In short, I describe CRO as the process of optimizing a website or app experience to increase the number of desired conversions (purchases, leads, revenue) based on the current traffic and audience.

A strategy puts that process in order. It has 4 parts:

• 1 goal. For a store, this is usually revenue per visitor. For a lead form, it is leads or lead value per visitor.
• The funnels you will work on. The paths visitors take from the first page they see to an order or a lead.
• A ranked list of test ideas. Each idea comes from research and has a score, so the team knows what goes next.
• A way to judge results. Rules for how long a test runs, which number decides it, and what happens to winners and losers.

When we start with a new client, we lay this out during the evaluation. We walk through the funnels and the approach for each one, and that becomes the strategy. It does not stay fixed. It changes as we learn more about the site and its users, both from the client’s team and from test results.

Step 1: Start with the 1 funnel step that loses the most visitors

My first tests come from research, not from a list of best practices. We look at the conversion rate of each step in the funnel and go after the step with the highest drop-off. If no step stands out, we start with the landing pages that get the most traffic.

Research has 2 halves, and you need both:

• Analytics shows what visitors did. Your funnel report shows where they left, on which device, and from which traffic source.
• Usability testing shows why. Watching real people use the page shows what confused them, what they missed, and what made them stop.

Analytics on its own tells you a step is leaking, but not what to change. Usability testing on its own gives you stories, but no sense of how many visitors they affect. Put together, they point to a problem worth testing. For example, analytics might show that most mobile visitors leave a long collection page without opening a product. Usability sessions might then show them scrolling past the filters and never finding their way back up. Now you have a clear problem, and a test idea that answers it. Our guide on How to do a CRO audit walks through this research step by step. If you would rather hand it off, we run a full website audit.

Step 2: Write every hypothesis in 1 sentence with 3 parts

Every test in our archive starts with the same sentence: “By [change X], we will [affect Y], resulting in [Z].”

• Change X is what you will build: the new headline, the moved button, the shorter form.
• Affect Y is the visitor behavior you expect to change, tied to the problem your research found.
• Result Z is the number you expect to move, such as conversion rate or revenue per visitor.

Here is a made-up example: “By showing shipping costs on the product page, we will remove the surprise at checkout, resulting in more completed orders.”

If you cannot fill in all 3 parts, the idea is not ready to test. The sentence also makes every result useful. A win supports the middle part. A loss tells you visitors did not see the problem the way you did, which is worth knowing before you spend more on it.

Step 3: Score every idea from 1 to 10 on 3 PIE factors

Research will give you more ideas than you can test. To put them in order, we use PIE: Potential, Importance and Effort. We rate each idea from 1 to 10 on each factor to get a score, and we generally go after the highest scores first.

• Potential: how much room the page has to improve. A step that loses a lot of visitors has more.
• Importance: how much traffic and revenue the page carries. A fix on a page few people see is worth less.
• Effort: how much work the test takes to build. In the classic version of PIE this factor is called Ease, so a lighter build scores higher.

Here is how 3 made-up ideas might score, adding the 3 numbers together:

1. Show shipping costs on the product page: 8 + 9 + 7 = 24
2. Shorten the homepage headline: 5 + 8 + 9 = 22
3. Rebuild the account sign-up flow: 7 + 4 + 3 = 14

The shipping test goes first. The sign-up rebuild waits, even though it has real potential, because few visitors reach that page and it takes the most work.

A scoring model is only as good as the evidence behind it. Any team can talk itself into a high Potential score for a pet idea.

That is why our ideas come out of research before they get a score. An idea backed by a funnel report and a usability session is much harder to inflate than a hunch.

Step 4: Judge on revenue per visitor, because a 6.0% conversion lift cut it 11.7%

Conversion rate counts how many visitors bought. Revenue per visitor also counts how much they spent. A test can lift the first and cut the second, and only the second pays the bills. So we judge every test on revenue per visitor, not conversion rate alone.

Here is a test that shows why. On 1 store, removing the large hero image at the top of collection pages had shown positive movement on mobile. An earlier desktop test of a denser grid, with 5 products per row, had lost: visitors preferred the standard grid. The next idea in that line was to remove the hero image on desktop collection pages.

Over 19 days, conversion rose 6.0%. Read on that number alone, it looks like a win. But revenue per visitor fell 11.7%, because average order value dropped about 17%. More shoppers bought, and each one spent less. The goal for this test was to keep revenue even, and it missed that goal.

The takeaway: decide which number decides the test before it starts. For most stores, that number should be revenue per visitor.

This test was still implemented, for a reason that has nothing to do with conversion. Step 6 explains why.

Step 5: Mix big swings with small tweaks, like a form visitors preferred by 60.5%

Not every test should be a small tweak. Some tests change 1 detail, like a line of copy. Others rebuild a whole page or flow. I use the terms evolutionary testing (small, steady changes) and revolutionary testing (big swings) to define the scope of a test. A healthy plan has both.

Small tweaks take less effort and still win. In 1 of our tests, cutting the homepage hero copy and dropping its heading lifted conversion 7.6% over 21 days, and the change was implemented.

Big swings take more work and can move the number much further, in either direction. On 1 lead form, visitors moved through several short steps to request a price. We tested putting every question on a single page, so visitors could see the whole form at once.

Visitors preferred the multi-step form by 60.5% over 17 days, and we paused the test. Seeing every question at once likely made the form look longer than it was. 1 easy answer at a time got visitors started, and once they started, they kept going.

A result this large is not a wasted test. It is a strong signal. After a test, I look for how much the tested element moved the number. If the movement is high, we have a signal that the element is important to the user. Generally, we take 2 different directions for the next test. Applied here, that means the form stays on the plan, and the next tests explore 2 new ways to build on the step-by-step format visitors preferred.

Step 6: Every test ends in 1 of 3 calls: implement, pause or do no harm

Once a test has enough data, it gets 1 of 3 calls:

• Implement: the variation won on the number that decides the test, so it becomes part of the site.
• Pause: visitors preferred the original. The variation lives in the testing tool, so pausing the test puts every visitor back on the original page.
• Do no harm: the result is not a clear win, but there is a reason to keep the change anyway.

When do I keep a do-no-harm result? Generally when a client is passionate about an approach that aligns with branding or business direction more than with conversion. These are bigger-picture goals we optimize toward.

The hero image test in step 4 is an example. It missed its revenue goal. But keeping the collection page layout the same on mobile and desktop mattered to the business, so the change was implemented for that reason, with the revenue tradeoff on the record.

The key is to write the reason down. A do-no-harm call should be a clear business choice, made in the open, not a quiet way to keep a change that did not win.

Step 7: Build each test on the last, like stacked wins that lifted revenue per visitor 21.0%

A strategy is not a list of stand-alone tests. Each result should feed the next idea. On 1 store, we had already found several collection page wins on mobile. The next test ran with those earlier wins in place and added a back to top button, so shoppers could get back to the top of a long page in 1 tap.

Conversion rose 15.4% and revenue per visitor rose 21.0% over 19 days. The variation led every day from day 3, and it was implemented. Because the button ran together with the earlier wins, we cannot credit the lift to the button alone.

The takeaway: stacking changes teaches you less about each one, but it locks in gains faster. Decide which you need before you build the test, and say so in the hypothesis.

The hero image test in step 4 was built the same way. Its idea came from 2 earlier results: a positive move on mobile, and a denser grid that visitors turned down. Neither result alone would have pointed to that test. Together, they made it the logical next step. This is what a strategy gives you that a list of ideas cannot: each test starts from what the last ones taught you, win or lose.

A winning signal does not stay on 1 page, either. If we see a winning signal in a funnel, we check whether it holds further up and down the funnel. It often shapes the language and images in paid ads too.

Step 8: Update the plan after every result, with 2 weeks of data per test

Your strategy should change as results come in. We monitor every test several times a week, and we like 2 weeks of data before calling a test. That covers weekday and weekend visitors, who often behave differently.

Peak weeks need extra care. During a big sales event, the audience shifts toward people who are ready to buy, and different factors may matter more. During those weeks we watch results several times a day, but never make a call on less than a full day of data. Any result from a peak week gets retested once the audience returns to normal.

After each call, go back to your list:

• Rescore ideas that the result made more or less likely to work.
• Add 2 new directions for any element that showed a strong signal.
• Check winning signals further up and down the funnel.
• Log the result, the call and the reason, win or lose.

Running this loop test after test is the core of our experimentation and testing work.

A CRO strategy is not a list of ideas. It is a ranked plan that changes every time a test teaches you something.

Your CRO strategy checklist has 10 items

Use this list to check your own plan:

1. You have 1 goal, and for a store it is revenue per visitor.
2. You know the conversion rate of each step in your main funnels.
3. You started with the step that loses the most visitors, or your top landing pages.
4. Your ideas come from analytics (what visitors did) and usability testing (why).
5. Every idea has a 1-sentence hypothesis: “By [change X], we will [affect Y], resulting in [Z].”
6. Every idea has a PIE score, and the highest scores go first.
7. Your plan has both small tweaks and big swings.
8. Every test runs about 2 weeks before a call, and peak-week results get retested.
9. Every test ends with implement, pause or do no harm, and the reason is written down.
10. Each result changes the list: new directions for strong signals, new scores for the rest.

Frequently asked questions

What is a CRO strategy?

A CRO strategy is a plan for what to test on your site, in what order, and how to judge each result. It ties a ranked list of test ideas to 1 goal, usually revenue per visitor, and it changes as test results come in.

Where do you start on conversion rate optimization?

Start with research. Look at the conversion rate of each step in your funnel and go after the step with the highest drop-off. If no step stands out, start with the landing pages that get the most traffic.

How do you create a CRO strategy?

Research your funnels with analytics and usability testing, write each idea as a 1-sentence hypothesis, score the ideas with a model like PIE (Potential, Importance, Effort), and test the highest scores first. Judge each test on revenue per visitor, and update the list after every result.

How do you optimize your conversion rate?

Find the step where visitors drop off most, learn why through usability testing, and A/B test a change that answers that problem. Implement winners, pause tests where visitors preferred the original, and build the next test on what you learned.

How do you improve a conversion rate optimization program that has stalled?

Go back to research and look for the signal in past tests. When a tested element moved the number a lot, in either direction, it matters to visitors, so take 2 different directions for the next test. Mix a few bigger swings in with small tweaks, and check winning signals further up and down the funnel.