How to Do Conversion Rate Optimization on a Business Website: A Practical CRO Process

Conversion rate optimization is the most disciplined way to get more quote requests, calls and demo bookings from a business website without raising ad spend. However, this guide is not about design principles. It covers the process: research, hypotheses, prioritization, A/B testing and the statistics behind them. I have run this workflow for clients since 2012, so I share a method, not a promise.

What is conversion rate optimization for a business website?

Conversion rate optimization (CRO) is the systematic process of raising the share of visitors who take a desired action, using research, testable hypotheses and controlled experiments. On a B2B or corporate site, that action is usually a quote form, a phone call, a demo request or a brochure download rather than a checkout.

The key word here is process. In other words, you are not doing a one-off makeover. Instead, you build a loop that teaches you something every round. You measure, research, write a hypothesis, prioritize, test and document. Then you start the next round.

Business websites usually get less traffic than online shops, and their sales cycles are longer. Therefore I set up CRO differently on them. For example, I do not push every change into an A/B test. I pick the method that fits the traffic, and the rest of this article explains how.

How does CRO differ from conversion-focused design?

Conversion-focused design is a set of principles that help you build a page well from the start: a clear headline, one main action, trust signals and a short form. I covered those in my article on conversion-focused web design. CRO is the method that checks whether that design actually works.

In other words, a design is an assumption. CRO then confirms or rejects it. Moreover, even a great designer cannot always predict visitor behavior. I have lost count of the variants I felt sure about that lost the test.

In short, the two are sequential rather than competing:

• First, you build a solid base with conversion-focused principles.
• Then, you use the CRO process to find the weak spots in that base.
• Finally, you make the proven improvements permanent.

Put simply, design principles answer "what should we do". CRO answers "did it work". This article deals with the second question.

What should you measure before starting conversion rate optimization?

You cannot optimize what you do not measure. The first step is to define which action counts as a conversion. I explained how I make that call in my guide on setting website conversion goals. In addition, here I only summarize the minimum setup CRO needs.

In GA4 you mark important actions as key events. Google renamed GA4 conversions to key events. The word conversion now refers to the actions you use for bidding in Google Ads. I also run these checks:

• Does the form event fire only on a successful submission?
• Do phone and WhatsApp clicks count as separate events?
• Does campaign traffic carry correct tags from a UTM builder?
• Does any submission count twice?

In practice, dirty tracking means dirty test results. So I collect at least two weeks of clean data before I start any test.

Is your traffic enough for A/B testing?

The most common mistake I see on business sites is launching an A/B test on a page with a few hundred visits a month. At low volume a test runs for weeks or months without a result. Moreover, even when it ends, the result is often unreliable.

Specifically, I use a rough split. It is a starting frame based on field experience, not a guarantee:

• Hundreds of conversions per page per week: classic A/B testing works well.
• A few dozen conversions per week: test only big, bold changes.
• A handful of conversions per week: use qualitative research, user tests and before and after comparisons instead.

That said, you can still do CRO on low traffic. However, the standard of evidence changes. If four out of five people in a user test get stuck at the same step, that is usually reason enough to fix it. On the other hand, trying to measure tiny copy tweaks on low traffic wastes time.

Which pages should you start with?

Not every page on a business website carries the same value. I rank pages on two axes: traffic volume and closeness to the conversion. Quote pages, contact pages and paid landing pages usually rise to the top.

The homepage is a weaker starting point than most people think. It also gets plenty of traffic, but intent is scattered. Some visitors look for jobs, some are suppliers, and some only want your address. As a result, the effect of a homepage change is hard to isolate.

In practice I follow this order:

1. Landing pages for paid ads, because you already pay for every visit.
2. Pages with quote, demo and contact forms.
3. Service pages with the most organic traffic.
4. Points where blog readers move to service pages.

That said, this order is a suggestion, not a rule. If your data points to another page, trust your data. Still, I recommend putting the pages you spend money on first. Every point you gain there lowers your cost per lead directly.

Which quantitative data should you look at?

Research tells you where the problem sits. Quantitative data answers "what is happening". My first stop in GA4 is a funnel report for the pages on the conversion path. At which step do visitors drop off?

Next, I look at these breakdowns:

1. Device: if mobile converts far below desktop, the problem likely lives in the mobile experience.
2. Source: do paid and organic visitors behave differently on the same page?
3. Landing page: which pages get lots of traffic but few leads?
4. Form fields: at which field does abandonment spike?

Pages with high traffic and low conversion are the best starting point, because the same relative lift creates more extra leads there. For engagement problems, see my article on reducing bounce rate on a business website.

However, numbers alone are not enough. In short, they show you where the problem is, not why. For example, you may see drop-off rise at step two of a form. You still do not know if a field confuses people or if an unexpected price appears there. Qualitative research fills that gap.

How do you listen to visitors with qualitative research?

Qualitative research answers "why is it happening". My four main methods are session recordings, heatmaps, short on-site surveys and interviews with the sales team.

For recordings, Microsoft Clarity is a free and capable starting point. While watching, I look for patterns: rage clicks on elements that do nothing, back and forth movement in forms, long pauses at the pricing block. That said, recordings can eat hours. So I filter to non-converting sessions on the problem page only.

For an on-site survey, one question is enough: "Is anything stopping you from requesting a quote today?" The answers often reveal things no analytics tool can show.

Above all, the sales team is a gold mine. The five questions prospects ask most on the phone are the five questions your site fails to answer. For instance, if everyone asks about delivery time, that information is either missing or hard to find.

How do you write a good CRO hypothesis?

A research finding is not a test yet. You need to turn it into a testable hypothesis. I use this template: "Because we saw [data], we expect that [change] will raise [metric] for [audience], because [reason]."

For example: "Because recordings show mobile visitors stalling at the tax ID field, we expect that moving this field to a later stage will raise mobile form completion. The reason is simple: visitors rarely have that number at hand on first contact."

The template has three benefits:

• It protects you from random ideas like "the green button looks nicer".
• It fixes the metric you will track before the test starts.
• Even if the test loses, the reason tells you what you learned.

I do not add a hypothesis without a reason to the list. Because when it loses, it teaches you nothing. It only burns traffic and time.

How do you prioritize tests with ICE and PIE?

After research you usually have far more ideas than you can test. Traffic is limited, so you need an order. The two most common frameworks are ICE and PIE.

In practice I start with ICE, but I never leave the confidence score open. I base it only on the strength of evidence. Both quantitative and qualitative support means a high score. Likewise, one person's opinion means a low score. That way the loudest voice in the meeting does not jump the queue.

How do you set up an A/B test step by step?

Once you pick the top hypothesis, you set up the test. The order matters, and I never skip these steps:

1. Choose one primary metric, such as quote form submissions.
2. Write down secondary metrics, such as phone clicks and form starts.
3. Calculate the sample size and duration before launch.
4. Split traffic randomly and evenly between the versions.
5. Check both versions by hand on several devices and browsers.
6. Do not change anything else on the page during the test.

On the tooling side, Google Optimize no longer exists. Google shut it down on 30 September 2023. Instead, today I use a third party testing tool or a simple server side split. Whichever you choose, also send the variant name to GA4 as a separate parameter.

How do you calculate sample size?

Sample size is the number of visitors each version needs before the test can give a reliable answer. You calculate it before the test, not during it. You need four inputs: the current conversion rate, the smallest effect you care about, the significance level and statistical power.

Here is a worked example. Say your current rate is 3 percent and you want to detect a 20 percent relative lift, from 3 to 3.6 percent. For 95 percent confidence and 80 percent power, a common rule of thumb gives about 16 × p × (1 − p) / difference² per variant. So 16 × 0.03 × 0.97 / 0.006² comes to roughly 12,900 visitors per variant.

For many business pages that means months of traffic. Consequently, the math tells you something useful. Either you test a bolder change aiming at a bigger effect, or you choose another method for this page. Also, skipping the calculation makes the result impossible to interpret later.

What does statistical significance mean, and how do people misread it?

A 95 percent significance level caps at 5 percent the chance of seeing a difference when no real difference exists. It does not mean "the new version has a 95 percent chance of winning". In practice, that is the misreading I hear most often.

The second big mistake is checking results daily and stopping the test on the first day it crosses the threshold. Results swing during a test, because traffic mixes change daily. If you stop early, you crown false winners. Therefore I run every test until it reaches the sample size I calculated upfront.

The third mistake is scanning many metrics and claiming a win on whichever one moved. Look at enough metrics and one will show a significant difference by pure chance. In short, you pick the primary metric first and decide on that basis.

Finally, significance says nothing about size. A significant but tiny lift may not cover the cost of building it.

How long should an A/B test run?

The sample size sets the duration. On top of that, I add two rules. First, a test should cover at least one full week, and ideally two. Business visitors on Monday morning behave very differently from visitors on Saturday night.

Second, I do not let tests run forever. As weeks pass, cookie deletion, campaign changes and seasonality pollute the data. In my experience, a test that needs more than four to six weeks usually sits on the wrong page or tests too small a change. Still, this is a warning sign, not a rule.

During the test I also watch for these points:

• If a big ad campaign starts, I finish the test before it or postpone it.
• I keep public holidays out of the test window.
• I check the traffic split between versions every week.

If the split drifts clearly away from the plan, the setup has a bug. In that case I then discard the result.

Can A/B testing hurt your SEO?

Not if you set it up correctly. Google's guidance on website testing says A/B tests are fine as long as you follow a few rules.

Here is how I apply them:

• I never show Googlebot different content than users see. Otherwise it would count as cloaking.
• For tests on separate URLs, I add a canonical tag on the variant pointing to the original.
• If I need a redirect, I use a temporary 302 instead of a permanent 301.
• When the test ends, I remove the variant and move the winning content to the live page.

I also watch speed. Client side testing tools can delay the first render and cause a brief flicker. That hurts both the experience and the measurement. Checking the page with Lighthouse during the test stops a variant from losing because of speed alone.

How do you read test results?

When the test ends, I look at the primary metric and its confidence interval first. For example, a single "12 percent lift" figure can mislead. If the interval runs from 1 to 23 percent, the real effect could sit anywhere in that range.

Next, I check segments, but carefully. A variant that wins on mobile and loses on desktop is interesting. However, segment results rest on smaller samples. They are a source of new hypotheses, not a final verdict.

On business sites I take one more step: I check lead quality. A shorter form can bring more submissions while fewer of them turn into deals. So, where possible, I tag the leads from the test period in the CRM. Then I compare their quality with the sales team a few weeks later.

Finally, I document every test on one page: hypothesis, duration, sample, result, decision and lesson. Over time this archive becomes some of the most valuable marketing knowledge a company owns.

What do you do when a test does not win?

Many tests do not win, and that is normal. Still, a losing or flat test is not a failure. It is information: you learned cheaply that an assumption was wrong.

After such a result I ask four questions:

1. Was the change big enough for visitors to notice?
2. Was the reasoning right, or did we misdiagnose the problem?
3. Did the right audience see the test?
4. Did the tracking work as expected?

Most of the time the answer sits in the first question: the change was too small. A button color or a one word headline tweak rarely moves the needle at business site traffic levels. In those cases I test the same problem again with a bolder solution. For instance, instead of editing the headline, I rebuild the value proposition and the page order.

When is a before and after comparison enough?

If your traffic cannot support an A/B test, the method left is often a before and after comparison. You ship the change and compare the period before with the period after. It is easy, but the evidence is weak.

The weakness comes from everything else that changes between the two periods. Seasons shift, a competitor launches a campaign, the ad budget grows. For example, if leads rise after a form change in September, you cannot tell whether the form did it or the end of summer holidays did.

I reduce that risk with a few safeguards:

• I compare two equal periods of at least four weeks each.
• Next, I look at the same period last year to control for seasonality.
• I track an untouched control page over the same time.
• Finally, I record the change date as an annotation.

In short, before and after works for big, obvious changes. It cannot prove small differences, so I report such results as a signal, not as proof.

How do you involve sales and marketing in the CRO process?

CRO stalls quickly if it stays a job for designers or analysts alone. Many of my best hypotheses came from people who talk to customers every day. That is why I put sales and customer service at the start of the process.

My routine is simple. Once a month I hold a thirty minute call and ask three questions. What do customers ask most? Which objection comes up most? What did callers misunderstand after seeing the site? The answers become raw material for the next test round.

I also share results with the team. Sales notices a drop in lead quality long before analytics does. Meanwhile, marketing checks that the ad message matches the page message. As a result, CRO stops being one person's project and becomes a shared learning habit.

What should you watch for with mobile visitors?

On business websites, mobile and desktop visitors often do different jobs. Mobile visitors usually want a quick fact or want to call right away. Desktop visitors compare, download files and fill in longer forms. Therefore I never merge mobile and desktop results into one rate.

These are the mobile test areas I find most productive:

• A sticky call and WhatsApp button at the bottom of the screen.
• The right keyboard type in form fields, such as a number pad for phone numbers.
• Moving long text blocks into expandable sections.
• Showing the value proposition and the action together on the first screen.

I covered the technical side in my mobile friendly test guide. From a CRO angle, I add one note. On mobile, a phone click can be worth as much as a desktop form. So define the primary metric in mobile tests as "form or call". Otherwise you may call a change that lifts calls a loser.

Why is CRO different for B2B quote forms?

In B2B, a conversion is often the start of a sales conversation rather than a sale. More forms do not always mean more revenue. This shapes how you run conversion rate optimization.

So on B2B sites I extend the metric chain: form submission, qualified lead, proposal and won deal. I make the test call on form volume, then check it a few weeks later against the qualified lead rate. That is also why feeding offline conversion data back into Google Ads pays off.

I covered form design itself in my article on booking, quote and demo forms. From a CRO view, my most productive tests usually involve fewer fields, multi step forms and a note next to the form on what happens next. For example, "We call you within one business day" reduces uncertainty. Only write it if you can keep that promise.

Which tools do you need for conversion rate optimization?

You do not need an expensive software stack to start CRO on a business site. My minimum kit looks like this:

• GA4, for funnels, segments and key event reports.
• Microsoft Clarity, for recordings and heatmaps.
• An A/B testing tool or a server side split, if traffic allows.
• A simple spreadsheet, for the hypothesis list, ICE scores and test archive.
• A CRM, to track lead quality and closed deals.

On the paid side, checking value with a ROAS calculator shows what a CRO gain adds to ad efficiency. When the conversion rate rises, the same budget brings more leads, and bidding strategies learn faster.

Discipline matters more than the tool. Irregular testing with an expensive tool teaches less than regular testing with a spreadsheet.

The CRO mistakes I see most on business websites

Over the years I have seen the same mistakes at many companies. These come up most often:

1. Copying a competitor's site without any research.
2. Running several overlapping tests on the same page.
3. Stopping a test the first day it hits significance.
4. Counting form volume and ignoring lead quality.
5. Forgetting to ship the winning variant to the live page.
6. Not documenting results and retesting the same idea a year later.

I would add the best practice list trap. A change that worked on someone else's site may not work for your audience. That said, such lists are a useful source of hypotheses, as long as you do not treat them as proof.

Most of these mistakes share one root: impatience. CRO can frustrate a team that wants weekly wins. Its real value lies in the learning that builds up over months.

How do you turn conversion rate optimization into an ongoing process?

The real return of CRO comes from consistency, not from any single test. I build it with a simple calendar: one research day a month, a hypothesis list review every two weeks and a learning summary every quarter.

I also make ownership clear. Who proposes a test, who approves it, who builds it and who documents it? If those questions stay open, the process quietly dies within a few months. For reporting, I suggest fixing your indicators with the framework from my article on digital marketing KPIs.

To sum up, conversion rate optimization runs in this order: clean tracking, research, reasoned hypotheses, prioritization, properly sized tests and documented decisions. If you want to build it yourself, these steps give you a solid start. If you want to handle it together with design and infrastructure, see my web design service. For the paid traffic side, see Google Ads management.