In short
Learning goal: by the end of this lesson you can turn a client's funnel data, heatmaps and recordings into a ranked list of tests, with the traffic maths that says which ones can actually be run.
Who this is for: agency strategists running the first audit for a new client, or re-auditing an existing one before planning the next quarter.
Most audit decks fail in the same way. They list forty observations, score them on gut feel, and recommend tests the client's traffic can't settle in under a year. A good audit does three things: it finds where money leaks, it explains why with evidence, and it checks that each proposed test can reach a result in a reasonable time.
What to measure at this step
| Metric | Definition | Primary goal or diagnostic |
|---|---|---|
| Revenue per visitor (RPV) | Total revenue ÷ all visitors in the period (why it beats conversion rate) | The north star for most tests; profit per visitor where product costs are known |
| Step rate | Share of visitors at one step who reach the next (product view → add to cart → checkout → purchase) | Diagnostic: tells you where to look |
| Step traffic | Visitors per day reaching each step | Decides which tests are feasible |
| Click and scroll patterns | Where visitors click, how far they scroll, repeated and dead clicks | Diagnostic: suggests why a step leaks |
| Recordings | Individual visits played back | Diagnostic: shows a broken journey; never a sample to count |
The rule that keeps an audit honest: diagnostics tell you where to look, and only a test on the primary goal tells you whether a change helps.
Step 1: Map the funnel and find the leak
Start from 60 to 90 days of the client's store data, excluding sale periods. Write down the visitors per day and the step rate at each step: landing, product view, add to cart, checkout start, purchase.
Compare each step with the client's own history rather than with an industry average. Benchmarks hide too much variation between categories and price points to tell you whether one store is underperforming. The one widely cited reference point is Baymard Institute's average documented cart abandonment rate of 70.22%, from 50 studies (list updated September 22, 2025). It is useful for showing a client that most carts are abandoned everywhere, not for grading their store.
Then look for the step where the drop is large and the traffic is high. A 10-point leak on a step with 4,000 visitors a day is worth more attention than a 30-point leak on a step with 50.
Step 2: Explain the leak with evidence
For each leaking step, gather evidence before writing a hypothesis:
- Heatmaps on the pages at that step, split by device. Look for clicks on elements that aren't links, important content below the point most visitors scroll to, and repeated clicks.
- Recordings filtered to visitors who added to cart but didn't buy, or who viewed a product and left. Watch ten or twenty, take notes, and stop. Recordings explain; they don't measure.
- The checkout itself. Place a real order on mobile. Note every surprise: a shipping cost that appears late, an account prompt, a slow payment step. Baymard's survey of US shoppers lists extra costs that were too high as the most common reason for abandoning a checkout, given by 40% of those who abandoned for a reason other than browsing (same source).
- Support tickets and reviews. Questions shoppers ask before buying are often the clearest hypotheses you'll get.
Step 3: Write hypotheses with a mechanism
Each idea in the audit should say what changes, for whom, and why it should change behavior. "Add trust badges" is an observation. "Mobile shoppers who reach payment hesitate because the returns policy isn't visible there; showing it under the payment methods should raise orders per shopper who reached checkout" is a hypothesis you can test and learn from either way.
Step 4: Score honestly, then check feasibility
Score each hypothesis for impact, confidence and effort. Be strict about confidence: evidence from the client's own data scores higher than a pattern you've seen work elsewhere. Then do the step most audits skip: check whether the client's traffic can settle the test.
(three candidate tests for one illustrative client, not a real store)
All three plan for a 10% relative change, at 95% confidence and 80% power, with two arms splitting the traffic evenly.
| Test location | Daily visitors | Baseline rate | Visitors per arm | Days |
|---|---|---|---|---|
| Product page, judged on orders | 4,000 | 2.5% | 64,199 | 33 days |
| Collection page, judged on orders | 2,400 | 3.0% | 53,211 | 45 days |
| Checkout, judged on orders per shopper who reached it | 400 | 45% | 1,931 | 10 days |
The product page idea scored highest on impact, but it needs over a month. The checkout idea, on a Plus store, can be read in two weeks. The collection page idea needs 45 days at a 10% change; either test a bolder change there or move it down the list.
Sizes like these come from the same formula Benchmyrk uses to judge results; the sample size calculator runs them for any client, and sample size and test duration explains the four inputs.
Common problems
- Auditing pages, not the funnel. A beautiful product page review is wasted if the leak is at shipping costs.
- Scoring by enthusiasm. Write the evidence next to each confidence score so a colleague could check it.
- Ignoring traffic. An idea the client can't test in under eight weeks is a fix-or-skip decision, not a test.
- Treating recordings as data. Twenty recordings show what can happen, not how often.
- Benchmarks as targets. Clients remember the benchmark you quoted, not the caveat.
Test ideas, with the mechanism behind each
| Idea | Mechanism | Judge on |
|---|---|---|
| Delivery date on the product page | Removes uncertainty about arrival | RPV |
| Shipping cost shown before checkout | Removes a late surprise, the most common checkout complaint in Baymard's survey | RPV; checkout start as a diagnostic |
| Returns policy near the payment step (Plus) | Reassurance at the point of commitment | Orders per shopper who reached checkout |
| Fewer, clearer variant choices | Less effort to pick | RPV; add to cart as a diagnostic |
A 10-minute exercise on your own client
- Open one client's store analytics for the last 60 days.
- Write down visitors per day and the rate at each step.
- Mark the step with the largest drop that still has at least 1,000 visitors a day.
- Write one hypothesis for that step, with its mechanism.
- Put the step's daily traffic and baseline into the sample size calculator at a 10% change. If it needs more than 6 weeks, write down the bolder version of the idea.
You can do this with any analytics tool. Benchmyrk is optional.
Do it in Benchmyrk
Auditing ahead of a sale? The BFCM version for agencies starts with planning tests across a client portfolio.
Check your understanding
- A step has a 40% drop and 60 visitors a day. Another has a 15% drop and 3,000 a day. Which do you test first? Usually the second. The first is a fix-or-skip decision: at 60 visitors a day, a test would take far too long to settle.
- Recordings show five shoppers rage-clicking a size chart. Is that a result? No. It's evidence for a hypothesis. Test a fix and judge it on revenue per visitor.
- The product page idea needs 33 days. What are your options? Run it for the full time, test a bolder change that needs fewer visitors, or put a smaller, faster test first.