CRO (conversion rate optimization)

CRO (conversion rate optimization) is a systematic process of increasing the share of site visitors who take a desired action: buy, leave a request, subscribe. It relies on research into user behavior, hypotheses and testing changes, including A/B tests.

In plain words

Traffic costs money: advertising, SEO, email campaigns. CRO is about getting more orders from the same traffic. Not "change the button to green", but understanding where and why people don't buy, fixing it and checking that the fix really worked.

Optimizely describes CRO as the process of increasing the percentage of conversions on a site or in an app, including generating ideas, testing them with A/B and multivariate tests, and improving the user experience (Optimizely). VWO (Wingify) puts it similarly: CRO improves a site's performance by increasing the share of visitors who take the desired action, and uses A/B tests, behavior analysis and UX improvements (Wingify/VWO).

What exactly you optimize depends on how you measure conversion: into an order, a request, a micro-conversion such as adding to cart.

The CRO process

Different sources divide the process into a different number of steps, but the logic is the same.

1. Research

First you find out where the problem is and why it occurs. VWO advises combining:

• quantitative data — what is happening: bounce rates, traffic sources, demographics, the funnel;
• qualitative data — why: session recordings, heatmaps, surveys (Wingify/VWO).

Optimizely adds funnel analysis, to find the points where people leave, and recommends starting with popular pages (results show faster) or with valuable pages that underperform the rest of the site (Optimizely).

For an online store, typical data sources:

• the funnel in Google Analytics: product view → cart → checkout → order;
• a split by device (for most Ukrainian stores the bulk of traffic is mobile — check your own analytics);
• session recordings and heatmaps (for example, Microsoft Clarity or Hotjar);
• usability testing: a few people try to buy a product, and you watch where they get stuck;
• on-site surveys and manager calls: what buyers ask, why they refuse cash on delivery;
• CRM data: which orders are not collected at Nova Poshta.

2. Hypotheses

Hypotheses are formed from the problems. VWO suggests three parts: a specific change based on research, the expected effect on the conversion metric, and an explanation of why that should happen (Wingify/VWO).

Example: "Session recordings show that on mobile people spend a long time looking for the delivery cost. If we show the approximate Nova Poshta delivery cost on the product page, more people will proceed to checkout, because the uncertainty about the final price will go away".

3. Prioritization

There are always more ideas than resources. They are scored on simple scales:

• PIE (WiderFunnel): Potential — how much room for improvement the page has; Importance — how much target traffic passes through it; Ease — how cheaply the change can be made. Each item from 1 to 10 (Mida). VWO recommends PIE specifically (Wingify/VWO).
• ICE (Sean Ellis / GrowthHackers): Impact — how much the change will matter if it works; Confidence — how sure you are; Ease — how easy it is to do (Mida). Scores are either multiplied or averaged — agree on one rule.
• PXL (CXL): instead of subjective scores, 10–12 questions, mostly yes/no (for example, is the change visible without scrolling, does it work on mobile) (Mida).

An illustrative PIE scoring example for a children's goods store (scores 1 to 10, the total is the average):

The first two ideas concern steps every buyer goes through, so they get high Importance. The banner is easy to make, but its effect on orders is doubtful.

Eppo warns that such scores are estimates that create an "illusion of precision", and they help rank ideas but don't replace testing by experiment (Eppo).

4. Test

The change is checked: part of the visitors see the new version, part see the old one, and the results are compared. It is important to wait for statistical significance and a sufficient test duration (Wingify/VWO). Details in the article A/B test.

If there isn't enough traffic for a test, changes that obviously fix a bug (a button doesn't work, the list of branches doesn't load) are rolled out right away, and the effect is tracked through the trend.

5. Analysis and learning

The result is analyzed regardless of whether the new version won, conclusions are recorded and shared with the team (Wingify/VWO). Optimizely calls the "only wins count" approach outdated: optimization fails only when the team learned nothing (Optimizely). Even well-designed experiments don't always win: in Optimizely's 2026 report on 173 thousand experiments (2018–2026), correctly set up ones won in 58% of cases, weak ones in 35% (Optimizely, 2026). This is data from a testing tool vendor.

Then the cycle repeats.

How CRO differs from "improve the design"

A redesign usually begins with the thought "the site looks outdated". CRO begins with the question "where and why are we losing buyers".

• Starting point. In CRO it is behavior data. In a redesign it is often taste or someone else's example. VWO explicitly advises not to assume that a cleaner design or another site's successful test will work for you, and to always rely on data (Wingify/VWO).
• Scale of changes. CRO more often changes one thing at a time, to understand what exactly worked. A redesign changes everything at once, and if sales fell, it's unclear why.
• Verification. In CRO a change is considered successful when the numbers showed it. A redesign is often judged by the team's feedback.
• Experts' intuition is often wrong. Even a correctly set up test doesn't always win (58% in Optimizely's 2026 report), so an "obvious" improvement without testing remains a bet (Optimizely, 2026).

Design still matters: among typical CRO mistakes VWO lists neglecting basics such as distracting sliders and weak stock photos. The difference is that in CRO design decisions get tested.

What has changed: AI and traffic from AI answers

• AI in the CRO process. In Optimizely's 2026 report, customers using the platform's AI features ran about twice as many experiments; the authors call this an observation, not a proven cause (Optimizely, 2026). Analytics, hypotheses and prioritization stay with people: AI speeds up preparing variants.
• A new channel. According to Contentsquare, traffic coming from AI assistants is starting to convert at a level closer to search (Contentsquare, Digital Experience Benchmark, 25.02.2026). So it is worth looking at conversion split by source for this channel too.

How different sources see it

• VWO emphasizes combining quantitative and qualitative data and PIE prioritization.
• Optimizely emphasizes goals, the funnel and continuous learning; it doesn't impose a specific prioritization model.
• NN/g reminds us that an A/B test shows what changed but not why, so it should be complemented by qualitative research (NN/g, A/B Testing 101).
• CXL proposes PXL to reduce the subjectivity of scores.

Common mistakes

• Starting with tests without research. Testing button colors at random is an expensive way to learn nothing.
• Copying other sites' "best practices" without checking them on your own audience.
• Optimizing the wrong metric. Conversion rose because of discounts, and margin fell.
• Ignoring the mobile version. Desktop improvements won't help if people buy from phones.
• Ignoring the offline part. For stores with cash on delivery, collection matters, not only checkout.
• Not recording results. Six months later the team repeats ideas that were already tested.

FAQ

What is CRO? Conversion optimization: a process in which you research why visitors don't buy, change the site and check whether there are more orders.

How do I increase an online store's conversion rate? Find the funnel step where the most people leave; watch session recordings and talk to buyers to understand the cause; change one thing and check the result with a test or a comparison of periods.

Where do I start with site conversion optimization? With analytics that is set up properly: without correctly collected events (product view, cart, checkout, order) it is impossible to find where the problem is.

Do I need an A/B test for CRO? Preferably, when there is enough traffic. With little traffic, rely on usability testing, qualitative data and comparison of periods.

Related terms

• Conversion rate
• A/B test
• Micro-conversion
• Cart abandonment
• Usability

Sources

• Optimizely. Conversion rate optimization (Optimization Glossary) — optimizely.com/optimization-glossary/conversion-rate-optimization
• Wingify (VWO). Conversion Rate Optimization guide — wingify.com/conversion-rate-optimization
• Optimizely. Lessons learned from 173K experiments, 2026 — optimizely.com/field-notes/guides/173000-experiments
• Contentsquare. Digital Experience Benchmark Report, 25.02.2026 — contentsquare.com/blog/digital-experience-benchmark-report
• Tim Neusesser. A/B Testing 101. Nielsen Norman Group, 30.08.2024 — nngroup.com/articles/ab-testing
• Mida. Test prioritization frameworks: ICE, PIE, PXL — mida.so/blog/test-prioritization-frameworks-ice-pie-pxl
• Eppo. ICE scoring method — geteppo.com/blog/ice-scoring-method