The eCommerce conversion playbook: 15 levers ranked by evidence and impact

Prioritise commercial constraints, not cosmetic changes

Conversion optimisation should begin with the points where purchase intent is already being lost. A checkout error affecting every customer deserves attention before a new recommendation widget seen by a minority. The same principle applies to evidence: a controlled test is more decision-useful than a survey, and a survey is more useful than an unsupported industry claim.

The ranking below combines likely commercial impact with the quality and relevance of the available evidence. It is a starting point for diagnosis, not a universal implementation order. Your analytics, customer support data and checkout error logs should determine the final sequence.

1–5. Remove friction at the point of purchase

Checkout is usually the highest-leverage place to begin because it contains customers who have already expressed purchase intent. Baymard estimates that resolving the checkout issues it classifies as solvable could increase conversion by 35.26% for the average large US or EU site. This is a modelled opportunity, not a guaranteed result. Its US survey attributes abandonment to forced account creation for 18% of respondents, a long checkout for 17%, site errors for 17% and card declines for 10%. The average checkout contains 23.48 form elements, compared with an achievable 12–14.

Unexpected cost is the largest stated abandonment reason: 40% cite extra costs and 20% slow delivery. Sendcloud reports that 48% of European shoppers abandoned a basket because of delivery issues within the previous three months, while 76% prefer a specific delivery date at checkout. DHL found that 67% across 29 countries had abandoned because of the delivery offering. These are surveys rather than controlled tests, but the pattern is consistent: disclose the total landed cost and a credible delivery date before the final step.

Reduce the amount of typing and decision-making required. Stripe reports an average 22.3% conversion increase from Apple Pay, while displaying express checkout earlier roughly doubled conversion compared with presenting it only at the end. A Shopify-commissioned study places Shop Pay at up to 50% above guest checkout; that should be treated as a vendor-funded upper bound rather than a planning assumption.

Payment coverage should be local rather than merely extensive. Stripe found that adding at least one relevant non-card method increased conversion by 7.4% and revenue by 12% on average. Reported market effects included 39% for iDEAL in the Netherlands, 46% for BLIK in Poland and 12% for SEPA Direct Debit in the EU. In the Nordics, evaluate Swish, Vipps and MobilePay according to the countries you actually serve.

• Allow guest checkout and defer account creation until after purchase.
• Remove optional fields, preserve entered data after errors and explain declines clearly.
• Show shipping cost, taxes and a specific delivery expectation before checkout.
• Place relevant wallets where customers can use them without completing the full form.
• Monitor payment acceptance and conversion by country, device and payment method.

3, 10, 11 and 13. Strengthen the purchase decision

Reviews have their greatest effect when a product has none. Northwestern University's Spiegel Research Center found purchase likelihood was 270% higher with five reviews than with zero, increasing to 380% for higher-priced products. The incremental benefit then falls quickly, and purchase likelihood peaked at ratings between 4.0 and 4.7 rather than at a perfect 5.0. For an established catalogue, focus on coverage and credibility—not review volume for its own sake.

Product content must resolve the questions that block a decision. Baymard's 2026 benchmark rated 52% of desktop and 62% of mobile product-page experiences below acceptable, and 84% of research participants used size information when it was available. Prioritise accurate imagery, dimensions, fit, materials, compatibility and product-specific FAQs. Widely repeated claims that video alone raises conversion by 86–144% rely on weak sources and should not be used as a forecast.

Returns are part of the offer. Baymard found 13% abandoned because the returns policy was unsatisfactory. In the 2025 NRF and Happy Returns study, about 81% read returns policies before buying and 82% said free returns matter. Free returns may still be commercially wrong for a high-return category, so test the policy against margin and return behaviour. EU merchants should also confirm that their withdrawal journey meets current legal requirements.

Trust cannot be reduced to a row of badges. Baymard found 19% abandoned because they did not trust the site with their card details, but its research also suggests users struggle to distinguish security seals. A generic padlock sometimes outperformed lesser-known branded marks. Clear company details, recognised payment interfaces, transparent policies and a technically sound checkout are more credible than decorative badges.

6, 8 and 9. Improve speed, relevance and findability

Performance improvements affect every stage of the journey. A Deloitte and Google analysis of 37 European and American brand sites covering 30 million sessions associated a 0.1-second mobile speed improvement with 8.4% higher retail conversion and 9.2% higher order value. Because that study was observational, Vodafone Italy's controlled result is useful supporting evidence: a 31% improvement in Largest Contentful Paint produced 8% more sales.

Personalisation can improve commercial relevance, but execution quality determines the outcome. McKinsey reports a typical revenue effect of 10–15%, with a broad 5–25% range by sector and implementation. Begin with useful, explainable applications—recently viewed products, complementary items and category affinity—before complex individualisation.

Site search serves a disproportionately valuable audience. Across 609 million searches on 113 retail sites, search users represented about 24% of visits and 44% of revenue. Econsultancy reports conversion of 4.63% for search users versus 2.77% for non-search users. This is correlation: high-intent customers are more likely to search. The correct conclusion is to remove search failure, not to assume that adding search automatically creates the observed lift.

• Measure speed at the template level and on real mobile devices, not only on the homepage.
• Track zero-result searches, misspellings, filter exits and search-to-purchase rate.
• Require every recommendation treatment to improve revenue or margin per session in a controlled test.

7 and 12. Expand affordability and recover lost demand

Buy now, pay later can unlock orders where affordability is the constraint. Stripe's test across more than 150,000 sessions found revenue increased by up to 14%, with more than two-thirds of BNPL volume attributed to net-new sales. The largest conversion gains appeared between $500 and $1,500, although effects were visible across order values. Compare that incremental contribution with merchant fees, which an older RBC estimate places at 3–5%, and with returns and default-related customer outcomes.

Cart recovery addresses demand already created rather than improving the on-site conversion rate itself. Klaviyo reports that abandoned-cart flows are its highest-earning automation, averaging 3.33% conversion and $3.65 revenue per recipient, with the top decile reaching $28.89. Treat those values as platform benchmarks. A restrained sequence tied to actual basket state is usually more defensible than repeated discounting.

• Offer BNPL where order value and customer preference justify its cost.
• Measure approval rate, contribution margin, refunds and repeat purchase—not only checkout completion.
• Exclude customers who completed elsewhere and stop the recovery flow immediately after purchase.

14 and 15. Prepare for AI discovery without manufacturing urgency

AI-assisted product discovery is becoming commercially relevant, although it is not yet a mature on-site conversion lever. Adobe reported that AI-referred traffic converted 42% better than non-AI traffic in March 2026. That reflects traffic quality, not proof that technical changes create a 42% lift. European research also indicates that 43% of shoppers have used AI to decide where to buy and 19% say it influenced their final decision.

Make products machine-readable through accurate structured data, consistent availability and pricing, descriptive product copy, clear policies and indexable comparison information. This improves the quality of information available to assistants without designing the store around an immature channel.

Urgency belongs last because credible effect-size evidence is limited and regulatory risk is increasing. The European Commission has identified fake countdown timers as a dark pattern in its Digital Fairness work. Use urgency only when the underlying constraint is real—for example, a documented order cut-off, genuine stock level or fixed campaign deadline.

A disciplined 90-day testing sequence

Start with instrumentation. Confirm that checkout steps, error states, payment outcomes, delivery selections, refunds and completed orders are measured consistently. Segment by device and country; mobile cart abandonment has historically been reported at 80.02% versus 66.41% on desktop, although more recent data suggests the conversion gap is narrowing.

Use the first month to remove defects and obvious policy friction. In the second, test decision support and payment relevance. In the third, improve discovery and recovery. Run tests long enough to capture the purchase cycle, avoid changing multiple elements in the same journey, and make the primary decision on refund-adjusted contribution—not platform conversion alone.

How to interpret the evidence

The figures in this guide come from Baymard Institute, Northwestern University's Spiegel Research Center, Stripe, Shopify-commissioned research, Sendcloud, DHL, Deloitte and Google, Vodafone Italy, McKinsey, Econsultancy, NRF and Happy Returns, Klaviyo, Adobe and other cited industry research supplied for this analysis.

Several findings are survey-based, observational or published by vendors with a commercial interest. They are useful for prioritising investigation, but they are not transferable promises. Store category, market, traffic quality, device mix, order value and the quality of the existing experience will determine the result. A Shopify benchmark of roughly 1.4% average conversion and more than 4.7% for the top decile illustrates the opportunity range; it does not establish the right target for every business.