What a Shopify CRO specialist does, and how to check the work

A CRO specialist's output is a decision log, not a redesign

The easiest way to understand what a Shopify CRO specialist does is to look at what they leave behind. A good one leaves a written record: what was measured, what was changed, what happened, and what was decided next. A weak one leaves a new look and a deck.

The reason is mechanical. Conversion rate is a ratio of two noisy counts, sessions and orders. Most stores see that ratio move for reasons unrelated to anything the site did: a different ad set, a sale email, a holiday, a bad week of weather in your biggest market. The only way to attribute a movement to a change is to isolate the change, measure it against a baseline, and write down the result whether it flattered the work or not. That discipline is the job. The design skill is the smaller half.

This page describes what a specialist does week to week, what evidence you should ask for before hiring one, and a short test for telling a tested change from an opinion.

A specialist's week is mostly measurement and sequencing

The visible work, the layout tweaks and the new section on the product page, is perhaps a third of it. The rest is invisible, which is why it is the part worth asking about.

Reading the funnel. In Shopify admin, Analytics shows an online store conversion funnel: sessions, sessions with add to cart, sessions that reached checkout, sessions that completed it. A specialist reads each step ratio by device, because a store can pass on desktop and fail on the phone that most of its traffic uses. The step with the widest gap against its siblings is where the week starts.

Watching real sessions. Numbers say where people stop. Session recordings and heatmaps, from a tool such as Microsoft Clarity, suggest why: a tap target that does nothing, a size selector hidden below the fold, a banner that covers the add to cart on a small screen. These are hypotheses, not findings. A recording shows one person; it justifies a test, not a verdict.

Checking the page is not broken first. Before testing a headline, a specialist checks that the page loads and behaves. Google publishes thresholds for this. On web.dev, Web Vitals defines a good Largest Contentful Paint as 2.5 seconds or less, assessed at the 75th percentile of page loads, segmented across mobile and desktop. If the product page misses that on a phone, no copy change will recover what the delay costs. The fix is upstream, and we cover it in why your Shopify store is slow.

Writing hypotheses. A hypothesis is one sentence with three parts: the observation, the change, the metric expected to move. "Buyers on mobile do not see delivery cost until checkout; showing it on the product page will raise the share of add-to-cart sessions that complete checkout." If it cannot be written that way, it is a preference.

Running and closing tests. One change per test where traffic allows, a primary metric chosen in advance, a stop rule chosen in advance. Then the result goes in the log, including the ones that lose. A store with a record of ten tests, four of them flat or negative, is in better shape than one with a record of ten wins, because the second record was not written honestly.

Cleaning up. Ending a test means switching off everything it touched: the variant code, the discounts, the embeds, the tracking events. Leftovers are how a test from March is still quietly changing prices in October. Ask any candidate how they decommission.

Traffic decides what kind of testing is honest

This is where most CRO claims quietly fail, and the thing a store owner can check without any statistics training.

A split test needs enough orders in each variant to separate a real difference from chance. A store with a few hundred orders a month cannot detect small differences in any reasonable time. If a specialist promises a conversion lift of a precise size from a test on a low-traffic store, ask how many orders each variant received and how long it ran.

For smaller stores the honest work looks different:

• Fix known defects first: speed, broken elements, a missing sticky add to cart, shipping cost revealed too late. These are removals of friction with well-documented mechanisms, so they do not need a test to justify them. They need a before and after on the funnel.
• Make larger, rarer changes and compare periods, stating plainly that the comparison is weaker than a split test and what else changed in that window.
• Save split tests for the pages with the traffic to carry them, usually the best-selling product page and the cart.

A specialist who runs the same testing programme on every store regardless of volume is selling a process, not reading your numbers.

Fewer form fields is evidence; "best practice" is not

Not every change needs your own test. Some have independent, published evidence, and a good specialist cites it rather than inventing a rationale.

Baymard Institute's research on checkout form fields reports that the average checkout in 2024 contained 11.3 form fields, that 17 percent of users had abandoned an order because of checkout complexity, and that most checkouts need only 8 fields. That gives you a count you can check yourself in two minutes: put one product in a cart on your phone, open checkout, and count every input between you and the card number. Everything above 8 is a candidate for removal, and the claim behind it comes from a source you can open.

This is the pattern to look for. A claim with a named source and a method you can repeat is evidence. "Trust badges increase conversion" with no source is an opinion that has been repeated enough times to sound like evidence. We go through the checkout in Shopify checkout optimization and the wider list in the store audit checklist.

What to ask for before you hire one

Six questions, each of which has a concrete answer or does not.

1. Show me the log from a test that lost. Anyone doing real testing has one. Names can be removed; the hypothesis, the sample and the decision cannot be.
2. What was the primary metric, and who chose it before the test started? The answer should be revenue per session or conversion rate, set in advance. If a metric was picked after the result, the result is a story.
3. How many orders did each variant get? A number, not "enough".
4. What else changed during the test window? Sales, campaigns, theme updates, app installs. If the answer is "nothing", be sceptical.
5. What do you do on a store with too little traffic to test? The answer should be a different method, not a refusal and not the same method.
6. How do you shut a test down? See the cleanup point above.

Also ask what the work is measured by. For an agency relationship, the useful measure is not the number of tests shipped. It is whether the funnel moved and whether you can see why. For your own selection process, the measure we would use is qualified enquiries: how many of the conversations you start turn out to be with a person who can actually describe their store and what they want changed.

A tested change versus an opinion: five checks

You can apply these to anything a specialist shows you, including ours.

A before and after across different weeks is not worthless. It is a weaker kind of evidence, and the report should say so. The failure is presenting it as the stronger kind.

Costs sit in a different conversation. If you are comparing retainers and project fees across providers, what a Shopify agency costs covers how engagements are structured, and the services page describes how we run ours.

What to fix first, and how long it takes

Start with a day of your own measurement, before you talk to anyone. Pull the funnel by device, count your checkout fields, run the product page through PageSpeed Insights on a phone, and write down the weakest step. That one afternoon gives you a baseline, and a baseline is what lets you tell afterwards whether any work paid off.

Then fix the defects that need no test: the speed problems, the missing mobile add to cart, the late shipping cost, the surplus fields. Expect one to three weeks for a store with a competent theme, depending on how many apps are in the way.

Only then consider split testing, and only on pages with the orders to carry it. If you would rather have the baseline, the ranking and the first fixes done by someone who has seen it across many stores, that is what our Store Teardown is for.