How to improve ecommerce conversion rate without guessing

"Lots of visits, no sales" has two very different causes. Either the wrong people arrive, or the right people arrive and cannot buy.

Threads on Shopify's community forum show small sellers asking this exact question. In one thread, replies point at missing trust signals, unclear product pages, poor-quality traffic and costs that appear at checkout¹. These are forum posts from small shops, not data. They do show how many different things one symptom can hide.

Split the problem with your own numbers:

• By source. If one source converts and another does not, the site is probably fine. The traffic is the problem.
• By device. If every source converts badly on one device, suspect the site.
• By entry page. A landing page is the page people reach after clicking your ad. When the ad promises one thing and the page opens with another, buyers leave. That is a message problem, not a button problem.

Also keep your expectations honest. Baymard Institute, a UX research firm, averages published cart abandonment studies and finds that most carts are abandoned². An abandoned cart is the normal state of a shop, not a fault. A fault is a step where cart abandonment rate jumps for one device, one source or one payment method.

A funnel report shows how many people reach each step. GA4 funnel exploration lets you define the steps from events and choose whether the funnel is open or closed³. For shops, Google's ecommerce event set includes begin_checkout, add_shipping_info, add_payment_info and purchase, which give you the checkout steps⁴. Google's help pages are the platform describing its own tool: good for what a report does, silent on whether your data is right.

Before you trust any of it, run one check. Compare GA4 purchases with orders in your shop for the same dates. If they disagree, your tracking is wrong somewhere. Set-up errors, consent settings and ad blockers can all stop events firing. Fix measurement first. A diagnosis built on a broken funnel is a confident guess.

Then read the funnel for the biggest gap you can act on, not the biggest gap you can see. A drop on a step nobody can change is trivia.

GA4 also has a checkout journey report: a closed funnel from begin checkout through shipping and payment to purchase, which only shows data once those ecommerce events are set up⁵. Use it to find the step where people stall. Then sort what you find into two piles.

Fixable:

• Payment errors and failed card attempts.
• Forced account creation before purchase.
• Shipping costs that first appear on the last step.
• A payment method your customers expect, missing from your shop.

Not fixable: people who were comparing prices, saving a basket for later or never intended to buy. They were never a sale. Chasing them with discounts buys orders you would have had anyway, at a worse margin.

Payment errors are worth checking first because they are cheap to find and every failed attempt was a buyer who wanted to pay. For the full treatment, see checkout optimisation ecommerce.

Analytics says where people leave. It does not say why. For that you need qualitative research: watching real sessions, talking to customers, short usability sessions. Session replay and heatmap tools help with the first.

Know what a small usability study can and cannot do. Nielsen Norman Group, a usability research firm, separates qualitative studies, where a handful of participants is enough to find problems, from quantitative studies, which need far more people to measure anything⁶. So if a few testers get lost on your size guide, you have found a problem. You have not measured how many customers it affects.

Page speed is another candidate. Core Web Vitals are Google's metrics for real-user page experience. Read Core Web Vitals conversion case studies before you assume a speed fix will lift sales.

To turn all of this into a to-do list, use a conversion rate optimisation audit before you run a single test.

One agency that sells CRO argues, in its guide to choosing a provider, that reading about conversion optimisation is not the same as doing it, and that the work rests on a learning strategy⁷. That is a vendor talking about its own trade. Our view is simpler: a test without a diagnosis behind it tests whatever came to mind first.

Not every change needs an experiment.

• Fix directly when the problem is clear and the fix has no plausible downside: a broken button, a payment error, a form that rejects valid postcodes.
• Test when you cannot predict the effect: a new page layout, a different way to show delivery costs, a restructured product page.

An A/B test randomly splits visitors between two versions and compares them on a metric you set in advance. Your traffic limits what you can test. The minimum detectable effect (MDE) is the smallest lift your test design can reliably pick up, and lower traffic means a larger minimum. A guide from Shopify, a platform vendor, says A/B testing is probably not the best optimisation effort for low-traffic stores, and that talking to customers may pay back better⁸.

That is the honest position for many brands: some pages have enough traffic to test, and many do not. The detail is in A/B testing low traffic ecommerce. If you are weighing CRO against buying more traffic or doing it in-house, see whether CRO is worth it for ecommerce.

A higher conversion rate is not automatically good news. A discount can lift the rate and still lower what you keep.

Example (illustrative). A shop adds a free-delivery offer. Orders per hundred sessions rise from two to three. But the offer is paid for out of each order's margin, and the new orders are smaller. Whether the shop is better off depends on contribution margin per session, not on the rate.

Our convention is to judge CRO work on orders and contribution margin per session, with the conversion rate as a diagnostic. If a change lifts the rate and lowers margin per session, you have moved revenue, not profit.

Record the baseline before you change anything, and give each change a pre-agreed way to be judged. A result you can only explain afterwards is a story.

For ZFD: we begin with the funnel and the orders, not with a list of tactics. We fix clear faults on evidence and keep tests for changes your traffic can actually measure.