Conversion rate optimisation for e-commerce: where to start
A grounded CRO roadmap for e-commerce: how to audit, prioritise, and test the changes that actually move revenue. Strategy before tactics.
By NEXUS EditorialPublished
Most CRO projects fail before they begin. A team reads a few case studies, runs a button-colour test on a product page, sees no movement, and concludes that CRO does not work for their store. The test was not the problem. The starting point was.
Conversion rate optimisation is a research discipline that happens to use experiments. Skip the research and you are just guessing in public. This guide gives you the roadmap most teams wish they had been handed at the start: what to measure first, where the leverage actually lives in an e-commerce funnel, how to prioritise the long backlog of ideas you will end up with, and how to test without fooling yourself with noise.
What CRO actually is, and what it isn't
CRO is not a list of best practices. It is not making the buy button bigger. It is not copying what a brand on Twitter said worked for them. Those things might help. They might also do nothing. You will not know until you run the work properly.
At its core, CRO is a loop. You gather evidence about how your customers actually behave on your site. You form a hypothesis about a friction point. You design an experiment to test the hypothesis. You measure, learn, and feed the result back into your understanding of the customer. Then you do it again.
That loop has four stages:
Research. Quantitative and qualitative data about real user behaviour.
Hypothesis. A specific belief about why people are dropping off, and what change would fix it.
Test. A controlled experiment, usually A/B, that isolates the change.
Learn. A read of the result that updates your model of the customer, whether the test won, lost, or showed no clear movement.
If you only do step three, you are running a slot machine. You will get the occasional win, but you will never build a coherent understanding of your buyer. The teams that compound conversion lift over years are the ones treating CRO as research first.
The CRO audit: what to measure before touching anything
Before you change a pixel, you need a baseline. The audit is not glamorous. It is also the difference between testing on instinct and testing with evidence.
Funnel analysis
Map the steps from landing page to confirmed order. Pull the numbers from your analytics. For a typical Shopify store the stages are: sessions, product views, add-to-carts, checkouts initiated, checkouts completed. Calculate the conversion rate at each stage and the drop-off between them.
The biggest drop-off is your starting point. If 12% of product viewers add to cart but only 40% of those reach checkout, your cart and post-add-to-cart flow is the bottleneck. If product page to add-to-cart is the cliff, your product pages are the work. Numbers point at the room you should be in.
Session recordings
Hotjar and Microsoft Clarity both give you replays of real sessions. Clarity is free and surprisingly good. Watch fifty sessions. You will see things no spreadsheet can tell you: people rage-clicking a non-clickable element, hovering on the size selector for ten seconds, scrolling past the variant picker because it is below the fold, abandoning the checkout because a coupon field made them open another tab.
You are not looking for proof. You are looking for patterns. After thirty sessions you will start hearing the same complaints in your head. Those become your hypotheses.
Exit surveys
A single-question survey triggered on exit-intent or after a thirty-second idle on the checkout page does more for your roadmap than a year of think-aloud user testing you will never get around to commissioning. Two questions worth asking:
"What stopped you from buying today?"
"If you had bought, what would have convinced you?"
Free-text answers are gold. You will see the same words appear: "shipping," "size," "price," "trust." Each of those is a testable hypothesis waiting for a treatment.
Search query data
If your site has internal search, the queries are a direct read of unmet demand. Customers typing "linen shirts" into the search bar of a tailored shirts store are telling you exactly what to merchandise. Empty result pages tell you what to stock or what to rename. Customers typing your own product names misspelt tell you to add synonyms. Search query reports also surface category navigation failures — if people are searching for things that exist in your menu, your information architecture is hiding the goods.
Cart and checkout are where most CRO wins still hide. Source: Pexels
Highest-leverage areas in e-commerce, ranked by impact
Once the audit is done, you will have a backlog of fifty ideas. Some areas matter more than others. The order below is the rough impact ranking for most stores doing under £5m a year. Larger operators with established product pages may find the order flips — cart and checkout often becomes the top opportunity for high-traffic stores.
Product page
The product page is where the buying decision is made. Everything before it is a setup. Five elements drive the conversion:
Hero image. A single dominant image that loads instantly, shows the product in context, and is large enough to see detail. Lifestyle imagery generally outperforms pure white-background photography for considered purchases, but white-background still wins for grocery-style commodity buying. Test, do not assume.
Description. Read it aloud. If it sounds like a manufacturer's spec sheet, rewrite it. The job of the description is to handle the objection a buyer brings to the page. What are they worried about? Sizing, durability, returns, fit, ingredients, compatibility. Address those head-on.
Social proof. Reviews, star ratings, photo reviews, count of customers who bought this in the last month, expert endorsements. Social proof works because buying alone is uncomfortable. Reviews are not optional in 2026; what varies is how aggressively you surface them and which voices you elevate.
CTA. Above the fold, single primary action, no competing colour. "Add to bag" outperforms "Buy now" for considered purchases because it preserves optionality. The opposite holds for low-ticket impulse items.
Trust signals. Free returns, secure checkout, dispatch time. These are not decoration. For first-time buyers they are the difference between adding to cart and leaving the tab open for three weeks.
Cart and checkout flow
The cart is where qualified buyers walk away. Industry abandonment rates sit in the 65–75% range, which means most stores leak two-thirds of their qualified demand at the moment of payment. A few targeted fixes here often beat months of product-page experiments.
Audit your checkout against this list:
Guest checkout available, prominent, and not gated behind "create an account."
Single-page or accordion checkout, not a four-step linear funnel.
Shipping cost visible before the payment step.
Coupon code field collapsed by default. Visible coupon fields cause buyers to leave to hunt for a code.
Error messages inline and human ("Card number is missing a digit") not modal and cryptic.
Order summary persistent on the right at all stages.
Site search
A buyer who uses site search converts at two to three times the rate of a browser. Treat search like a feature, not an afterthought. Synonyms, typo tolerance, merchandising rules, and clean empty-state pages will all move revenue. Most Shopify stores ship with a search experience that is barely better than a SQL LIKE query. Switching to Algolia, Searchspring, or Klevu is often a clean win for stores past a few hundred SKUs.
Mobile experience
Sixty to seventy per cent of your sessions are mobile. Forty to fifty per cent of your revenue is mobile. The gap between those numbers is your mobile conversion opportunity.
Common mobile-specific killers:
Sticky add-to-cart bar missing on product pages.
Variant pickers that require zooming.
Checkout form fields without correct input types (no inputmode="numeric" on card fields, no autocomplete on address fields).
Image galleries that hijack vertical scroll.
Pop-ups that cover the CTA and have a 6px close button.
Pull your phone out and try to buy something on your own store. You will find five problems before you finish a coffee.
Page speed
Speed is a conversion lever, not just an SEO lever. Every additional second of load time on a mobile product page knocks roughly 7–10% off the conversion rate of that page, depending on category. Get your Largest Contentful Paint under 2.5 seconds on a real 4G connection, get your Interaction to Next Paint under 200ms, and you have removed a meaningful drag on the whole funnel before you have changed a single piece of copy.
Speed compounds with everything else. A faster product page makes a copy test cleaner, because you are not measuring the noise of users abandoning during load.
Email capture and nurture
You will not convert most first-time visitors. CRO is partly the art of failing gracefully. A well-timed email capture — not a pop-up two seconds after landing, but a contextual offer after a real engagement signal — turns lost sessions into a second chance. The nurture sequence that follows is where the long-tail conversion lives. Welcome series, abandoned cart, browse abandonment, post-purchase. None of this is product-page work, but all of it sits inside the conversion rate you should be tracking.
Prioritising what to test
You will end up with more ideas than you can run. ICE scoring is the simplest framework that survives contact with reality.
Score each idea from 1 to 10 on three axes:
Impact. If this works, how much does it move the needle?
Confidence. How sure are you, based on the research, that it will work?
Ease. How much effort, in design, build, and risk, will it take to ship?
Average the three. Sort. Test the top of the list first.
ICE is not a precision instrument. The point is to force you to articulate why you think a test will work. An idea you cannot confidently score is an idea you have not researched. Send it back.
A/B testing basics
A test that runs against insufficient traffic is not a test. It is a coin flip with a dashboard. Three numbers matter.
Sample size. Use a calculator (Optimizely, VWO, and CXL all offer free ones) to work out how many sessions per variant you need. The inputs are your baseline conversion rate, the minimum detectable effect you care about (usually 5–10% relative lift), your desired statistical power (80%), and your significance threshold (95%). For a typical store with a 2% conversion rate trying to detect a 10% relative lift, you are looking at roughly 30,000 sessions per variant. That is two to four weeks of testing on most mid-sized stores.
Statistical significance. Stop your test when you hit the pre-calculated sample size, not when the dashboard shows a green tick. Peeking at tests and stopping early is the single most common reason CRO programmes generate false wins that never replicate.
Duration. Run for at least two full weekly cycles, ideally three. Weekday and weekend buyers behave differently. Payday buyers behave differently. A test that ran from a Wednesday to a Sunday is reading a slice of customer behaviour, not the average.
If you cannot hit statistical significance in four weeks, the test is too small for your traffic. Either pick a higher-leverage area or batch changes into a more meaningful treatment.
Quick wins vs structural fixes
CRO programmes need both. Quick wins keep momentum and pay for the structural work. Structural fixes are where the compounding gains live.
Quick wins look like: fixing a broken trust badge, removing a dead coupon field, switching a CTA from outline to filled, adding express checkout buttons, fixing the mobile variant picker. Days of work, single-digit percentage gains, but consistent.
Structural fixes look like: rebuilding the product page template, redesigning the navigation, replacing the search engine, rewriting every product description in a category, migrating to a faster theme. Weeks of work, double-digit percentage gains, and the foundation that makes future testing cleaner.
Run them in parallel. Use quick wins to maintain reporting and prove the discipline. Use structural fixes to actually move the business.
ICE is just a way of forcing yourself to stop guessing. Source: Pexels
The research → hypothesis → test → learn loop, in practice.
How we approach CRO within a project
Our CRO engagements start with two weeks of research before we ship anything. Funnel pulls, fifty session recordings, exit survey deployment, search query review, heuristic walkthrough of the site against the ranked levers above. We produce a research deck that says, in plain terms, where the leaks are and which we believe matter most. That deck is the artefact the rest of the work hangs from.
We then build a prioritised test backlog using ICE, with the top five hypotheses written out as full briefs: hypothesis, treatment, metric, sample size, expected duration. We ship one or two tests at a time, never enough variants to overlap or interfere. We document every test, win or lose, in a repository the team can search a year later. The losers teach you as much as the winners; the only wasted test is the undocumented one.
The structural work — product page rebuilds, checkout reworks, performance work — runs alongside the testing programme rather than blocking it. By the end of a typical six-month engagement you have a research-grounded view of your buyer, ten to fifteen tests in the log, three to five structural changes shipped, and a process the in-house team can take over.
Book a CRO strategy session
If you have a store doing more than £50k a month and a sense that the conversion rate could be better but no clear plan, a CRO strategy session is where to start. We will run a heuristic audit against the levers above, point at the research you should be commissioning, and sketch the first three tests we would prioritise.
