Conversion Rate Optimisation

Where the wins usually are

After enough audits the same handful of problems keep appearing, and they are rarely the ones people expect. Checkout friction, unanswered product questions and slow mobile pages account for most of the lost revenue I find — not button colours or headline wording.

• Checkout and cart. Unexpected shipping cost is still the single biggest cause of abandonment. Showing it earlier costs nothing.
• Product pages. Missing sizing, delivery timing or returns information — the questions that make people leave to go and search for the answer.
• Forms. Every non-essential field costs completions. Most enquiry forms ask for two or three things nobody needs until later.
• Mobile. Where most of the traffic is and most of the testing is not. Tap targets, sticky elements, keyboard behaviour on inputs.
• Landing page to ad match. If the ad promised something the page does not immediately confirm, the visitor leaves before reading anything.
• Trust. Reviews, returns policy, real contact details, delivery expectations — visible at the moment of hesitation, not buried in the footer.

How do I know a test result is real?

Because it reached statistical significance against a sample size worked out beforehand — not because the chart looked encouraging on day four. Most reported “wins” are noise that was called early, and a false positive is worse than no test at all: it gets rolled out and quietly costs money for months.

What that means in practice:

• Work out the sample size first. Your current conversion rate and the smallest improvement worth having decide how much traffic the test needs. If that is more than you get in two months, do not run it — fix something obvious instead.
• Agree the stopping rule before starting. Checking daily and stopping at the first good-looking moment is how you manufacture results that do not survive contact with reality.
• Run whole weeks. Traffic behaves differently on a Tuesday than a Sunday, and part-weeks bias the outcome.
• One change at a time, unless traffic is genuinely high. Multivariate testing sounds thorough and needs far more traffic than most sites have. Below that, it produces confident nonsense.

I would rather tell you a test was inconclusive than dress up a 3% lift that will not replicate. Inconclusive is a real answer, and a great deal cheaper than acting on a wrong one.

Speed and trust move more than layout

Two things shift conversion rate more reliably than most design changes, and both can be measured before any test runs.

The first is page speed. Core Web Vitals are a ranking input, but the commercial case is simpler: a product page taking four seconds on a mid-range phone loses buyers before it can persuade anyone. I check real-world mobile performance early, because there is no sense testing headlines on a page people leave before it renders.

The second is trust. Cart abandonment and checkout abandonment are usually caused by an unanswered question rather than a design flaw — what shipping costs, when it arrives, what happens if it does not fit. Visible returns terms, delivery timing, real reviews and recognisable payment marks are trust signals, and on anything unfamiliar or expensive, social proof consistently outperforms another round of button tests.

How an engagement runs

1. Research fortnight. Analytics, recordings, heuristic review and customer voice, pulled into one prioritised list scored on likely impact against effort.
2. Quick fixes first. The friction that needs no test — a broken mobile layout does not require a control group.
3. Testing cycle. Where traffic supports it, one structured experiment at a time, each with a written hypothesis and a pre-agreed success metric.
4. Documented results. Every test recorded, winners rolled out, losers explained. Over time this becomes a genuine map of how your customers decide.

How much traffic do I need before CRO is worth doing?

For statistically sound A/B testing, roughly 1,000 conversions a month gives you results in a sensible timeframe. Below that, testing takes months to reach significance — so the work becomes structured diagnosis and fixing obvious friction rather than split testing, which on low-traffic sites usually produces bigger gains anyway.

Is this design work?

Partly. But most conversion problems are not aesthetic. They are a form asking for information people are not ready to give, a price that appears three steps too late, a mobile layout that buries the button, or a page answering a question the visitor never asked. Diagnosis comes before design.