A CRO audit includes a numbers-based snapshot of the funnel, a friction analysis, a ranked list of hypotheses and a testing plan.
It doesn't deliver gains. It delivers hypotheses: once tested, about two in ten will produce a measurable effect. Allow about a month.
The benchmark
A solid audit hands over six deliverables, and each one leaves proof you can check without being an expert.
| Deliverable | The proof to look for |
|---|---|
| Measurement audit | A quantified gap between GA4 purchases and Shopify orders ("tracking OK" isn't enough) |
| Quantified funnel analysis | Volumes alongside percentages, by device, source and visit type |
| Friction analysis | The number of sessions watched and customer messages read |
| Ranked hypotheses | Each one points to a dated, sourced observation in the audit |
| Testing plan | The duration of each test, calculated on your traffic, and the list of what can be applied without a test |
| Presentation | Your tech and content teams in the room |
In the authors' practice, the whole audit takes four weeks. The first goes to measurement, the second to the numbers and frictions, the third to the conversion audit itself, and the last to the roadmap.
An independent consultant I interviewed this summer also quotes about a month for a full audit. That timeline assumes access arrives quickly and the presentation date is already on the calendar.
The rule
Expect an audit to give you a testing calendar, dated and sized to your traffic. That format surprised a fashion brand when the authors presented theirs in July.
The brand's contact, who expected three to five big projects such as a product page redesign, received about twenty small tests spread over six months based on each season's traffic. Her reaction:
"I thought I'd walk out thinking, okay, that's project 1 and we start with that, then project 2. […] So actually, you shouldn't think in projects. You should think in tests."
Then she asked for time to think it over and a scaled-down version.
The authors reason this way because they once treated an improvement as 100% certain, only to find it didn't deliver the expected result. In their view, a big project ships dozens of changes at once, and no one can untangle their effects.
So judge the audit on its traceability.
Each hypothesis should point to a dated observation, and each test to a duration calculated on your traffic, using the method in How Many Sessions Do You Need to Launch an A/B Test?. Pricing and what moves the quote are covered in How Much Does a CRO Audit Cost?.
The steps
- Measurement. Check that the first deliverable quantifies the gap between your tools. As long as that gap is unknown, everything that follows rests on numbers nobody has verified.
- Hypothesis format. Require one sentence per hypothesis that states the observation, the proposed change, the target metric, the expected effect and the effort.
- Triage. Ask which hypotheses will go through a test and which will be applied directly. A broken filter, for example, gets fixed without waiting.
- The presentation. Set its date at signing, with the people who will carry it out.
- The announcement. Before delivery, ask your provider for its winning-test rate and the definition behind it, then brief your leadership.
For your leadership, copy this sentence and rewrite it with your numbers:
"Of every ten hypotheses we test, about two will win, and each only modestly. Most of the others will be neutral. The value will come from the accumulation of gains, and so from the number of tests our traffic lets us run."
The trap
The success rate you're quoted depends on the definition used. In June 2026, a specialized US agency reviewed 2,288 of its own A/B tests, run across 71 engagements, and published the result under each definition.
It counts 50.5% winners when a variant only needs to be deployed or clearly ahead, and 19.1% when it requires 95% significance. The same page lists figures from the major testing platforms, which range from 12 to 20% significant winners.
The authors themselves cite about one winning test in five, the historical average for the brands they work with.
That figure raises the same question as the others: the projection they presented to another brand this summer states it without specifying what they count as a winning test.
If your provider promises a much higher rate, ask for their definition, and check that inconclusive tests are included in the total.
For an audit before a redesign, the purpose changes, since you take stock of what earns money before touching anything. What Should You Audit Before a Shopify Redesign? covers it in detail.
