Conversion rate optimisation for eCommerce: building a CRO programme

Conversion rate optimisation helps an eCommerce business turn more visits into purchases, but its value extends beyond the conversion rate. A CRO programme identifies where purchasing journeys break down, which obstacles matter and which changes deserve resources. Without that discipline, a website can change constantly without anyone knowing what worked.

For decision makers, the question is not simply how to increase online sales. It is how to do so while protecting margin, customer experience and operational sustainability. That starts with measurement, not button colours.

Conversion rate optimisation starts with objectives and measurement

Before looking for opportunities, define what improvement means. More orders are not necessarily a good outcome if discounts absorb the margin or create operational problems.

Choose a primary metric aligned with the commercial objective, supported by metrics that flag unwanted effects. Purchase conversion rate might lead; average order value, margin, cancellations and returns provide context. Some signals arrive later, so allow for that when evaluating results.

The denominator matters too. Purchases per session and purchasers per user are not interchangeable. Keep definitions consistent across reporting and decisions.

A measurement checklist

• Check that purchases are recorded without duplication.
• Reconcile analytics with commerce platform records where possible.
• Validate product views, basket additions, checkout starts and purchase events.
• Check tracking across the website and payment services.
• Document measurement gaps related to consent, devices and tools.
• Record promotions, technical incidents, stock shortages and campaign changes.

Perfect measurement is not a prerequisite. Knowing which data is dependable, and where its limits lie, is. An apparent decline may reflect a missing event rather than a worse shopping experience.

Read the funnel without confusing symptoms with causes

The funnel maps a commercial journey: arrival, browsing, product evaluation, basket, checkout and purchase. People do not always follow that sequence, but it helps locate points of friction.

Aggregate data rarely tells the whole story. Examine stages by device, traffic source, new and returning visitors, product category and market. Segmentation should reveal useful differences, not produce ever-smaller groups that are difficult to interpret.

Hypothetical example: a falling conversion rate could reflect more visitors who are browsing rather than ready to buy. A decline confined to mobile could suggest an interface or technical problem. Both are possibilities to investigate, not automatic diagnoses.

Basket abandonment needs the same care. Baymard Institute’s research on cart abandonment distinguishes several motivations, including browsing and comparison. Not every abandoned basket can be recovered by changing the checkout.

For each weak point, ask a concrete question. Is delivery information missing? Do unexpected costs appear? Can people complete payment? The funnel tells you where to look; understanding why requires further research.

Use qualitative research to understand obstacles

Behavioural data shows actions and interruptions. Qualitative research helps explain the expectations, doubts and difficulties behind them.

Start with support enquiries, pre-purchase questions and reasons for returns. Many businesses already hold this material, but keep it separate from website optimisation.

Add usability testing with people who reflect the shop’s audience. Give them realistic tasks without suggesting the correct route. Watching someone find a product or interpret delivery conditions is different from asking whether they like a page.

Nielsen Norman Group’s introduction to usability provides a framework for considering ease of learning, efficiency, errors and satisfaction. An attractive page is not necessarily easy to use.

Session recordings and interaction maps can add context when configured in line with consent and data protection requirements. They do not explain intent on their own.

Hypothetical example: participants hesitate to add an item because they cannot establish whether it works with something they already own. The issue to investigate is product information, not the prominence of the purchase button.

Turn evidence into prioritised hypotheses

A list of problems is not yet a CRO programme. Each opportunity needs a testable hypothesis connecting an observation, a proposed change and an expected effect.

A useful format is: “We observed this obstacle in this audience segment. We propose this change to reduce that difficulty. We will assess its effect through this metric while monitoring these possible side effects.”

Hypothetical example: support enquiries reveal uncertainty about delivery dates. The proposal is to show an estimate on the product page, helping shoppers make an informed decision. Monitor purchases and delivery enquiries, while checking that logistics can support the information displayed.

Use shared criteria to order the work:

Criterion	Practical question
Evidence	Does the problem appear in data, observations or recurring enquiries?
Relevance	How closely does it relate to purchasing decisions and commercial objectives?
Reach	Which shoppers actually encounter the obstacle?
Effort	What technical, content and operational resources are required?
Risk	Could the change damage margin, accessibility or functionality?
Measurability	Can available volumes support a useful evaluation?

Prioritisation should not imply precision that does not exist. A reasoned assessment is more useful than an elaborate score built on assumptions. Bug fixes, evidence-backed improvements and exploratory ideas also need different treatment.

A/B testing: when to experiment and when not to

An A/B test compares alternatives randomly assigned to comparable groups. Optimizely’s explanation of A/B testing sets out the principle: evaluate alternatives against a defined objective rather than internal preferences.

Before launch, specify the primary metric, monitoring metrics, eligible audience, assignment method and stopping rule. Check that the experiment records exposure correctly and gives the same person a consistent experience.

Sufficient traffic depends on the question

There is no universal visit threshold that makes every test viable. Sample requirements depend on the starting conversion rate, the smallest effect worth detecting and the chosen statistical parameters. What matters is traffic eligible for the experiment, not total website traffic.

If collecting the required sample takes longer than the commercial context can remain reasonably stable, testing may be the wrong approach. Running through promotions, assortment changes or substantial site updates can make results harder to interpret and apply.

Do not stop an experiment simply because an apparent advantage emerges. Choose the analysis method and stopping rules beforehand, rather than interpreting results opportunistically.

What to do when volumes are too low

Use qualitative research, technical checks, usability testing and changes supported by converging evidence. A reproducible fault that prevents payment should be fixed, not preserved to create a control group.

After release, monitor commercial and operational signals. Remember that a before-and-after comparison remains observational: seasonality, campaigns and product availability may affect it. It does not offer the same causal evidence as a controlled experiment.

Respect inconclusive results too. They do not establish that alternatives are equivalent; the data may simply be insufficient to distinguish a useful effect.

Checkout and performance: common priorities, not universal recipes

Checkout brings together sensitive decisions about personal details, delivery, costs and payment. Baymard Institute’s checkout usability research offers a reference for examining these steps, but does not replace research into your own shop.

Practical checks include guest checkout, clear costs, helpful form errors and retaining entered details after an unsuccessful payment. Payment methods should suit the audience and remain commercially sustainable.

For businesses selling across European markets, check currency clarity, supported destinations, delivery conditions and any applicable cross-border charges. Coordinate these with finance and logistics. They are operational commitments, not simply interface copy.

Performance concerns loading, responsiveness and visual stability. The Core Web Vitals guidance from web.dev explains how to measure these dimensions. Combine real-world usage data with diagnostic checks, taking account of the devices and pages involved.

A faster page does not guarantee more orders, but it can remove a genuine obstacle. That is why CRO and eCommerce development need to work together: components, integrations and platform constraints shape what can be improved.

Make CRO an ongoing programme

A programme works when each activity has an owner, a rationale and a subsequent decision. Marketing, development, support and operations need to share relevant evidence without turning every change into a debate about taste.

Keep a working record of the observed problem, hypothesis, intervention, evaluation method, result and limits of interpretation. This also prevents previously rejected ideas from resurfacing simply because the team has changed.

An ecommerce consultant should help connect those stages, rather than deliver recommendations detached from implementation. In our conversion rate optimisation work with partners, the focus is on linking analysis, priorities and delivery. An audit without the ability to act remains a document; releases without learning remain activity.

The pace depends on available resources and volumes. A few well-supported initiatives brought to a clear decision are more useful than numerous experiments left unresolved. Continuity comes from the quality of the process, not the number of tests launched.

Sources

• Baymard Institute — Cart abandonment data and reasons
• Baymard Institute — Checkout usability research
• Nielsen Norman Group — Introduction to usability
• Optimizely — A/B testing definition and method
• web.dev — Core Web Vitals guidance

Need to establish where to start? Talk to us about your measurement, purchasing obstacles and available resources, so we can scope a CRO programme proportionate to your business.