Conversion Rate Optimisation
Conversion rate optimisation means finding out why visitors leave without buying or enquiring, changing the page, and checking whether the change helped. The checking is the part most often skipped.
A test needs enough visitors and enough conversions to separate a real effect from chance. Many sites do not have them. On those sites an A/B test produces a confident-looking number that means nothing.
We will tell you which situation you are in. Most tests, run honestly, show no clear winner, so we cannot promise an uplift in conversion rate, revenue or enquiries. What you get is evidence you can rely on.
Finding out where and why people leave
Opinions about what to change are cheap. Research narrows them to the few worth testing.
Analytics review
Funnel reports show the step where most visitors drop out, split by device, browser and traffic source. A fault confined to one browser is a bug, not a design question.
Session recordings and heatmaps
We watch how real visitors move: where they hesitate, what they click that is not a link, which form field makes them give up.
Form analysis
Field by field: which are left blank, which trigger errors, which could be removed without harming the business.
Customer questions
Short on-page surveys and interviews with recent buyers. We also read support tickets and sales call notes for objections the page never answers.
Technical checks
Page speed, layout shifts, broken payment methods and error messages, tested on actual phones as well as desktop.
Walk-through
A reviewer completes each main task as a first-time visitor and notes every point of confusion.
Being honest about sample sizes
These four points decide whether a test result deserves belief.
Size is set in advance
Before launch we calculate how many visitors each version needs, from your current conversion rate and the smallest change worth detecting.
No stopping early
Checking daily and halting at the first significant reading produces false winners. The test runs to its planned size and for whole weeks.
Small effects need large samples
Detecting a modest improvement can take tens of thousands of visitors per version. If that means a year of traffic, the test is not worth running.
Inconclusive is a result
When the versions cannot be told apart we report exactly that, and do not search through segments until something looks significant.
How a test is run
One written record per test, kept whether it wins, loses or shows nothing.
Hypothesis
A sentence naming the change, the expected effect and the research finding behind it.
Design and build
The variation is built and checked on the main browsers and devices. Flicker on load and tracking errors are fixed before any visitor sees it.
Run
Visitors are split at random. We check in the first days only that data is arriving and nothing is broken.
Read
Results are analysed at the planned size against the primary measure chosen beforehand, with secondary measures as context.
Decide
A winner is coded into the site properly. A loser or a draw is recorded with what was learned.
Does your site have enough traffic to test?
A/B testing is sensible when
- The page under test records hundreds of conversions a month.
- Conversion tracking has been verified against real orders or enquiries.
- Developers can put a winning version live within a reasonable time.
- You will accept a result that contradicts the preferred design.
Skip testing and fix directly when
- Conversions number a few dozen a month or fewer.
- Research has found plain faults, such as a broken form on mobile.
- The site is about to be rebuilt anyway.
- Traffic swings so much by season that no two weeks are comparable.
Software used
Testing and recording tools are set up under your account and with your consent banner in control.
- Behaviour
- Microsoft ClarityHotjar
- A/B testing
- VWOOptimizelyConvertAB Tasty
- Analytics
- Google Analytics 4Google Tag ManagerLooker Studio
- Speed
- PageSpeed InsightsWebPageTestChrome DevTools
What clients ask about testing
How much improvement should we expect?
We do not forecast one. Published figures from other sites tell you nothing about yours, and a large share of well-run tests end without a winner.
Our traffic is low. Is there any point?
Yes, in research and direct fixes. Removing an obvious obstacle does not need a test to justify it. We simply do not dress the outcome up as a measured uplift.
Do recordings capture personal data?
Tools are configured to mask form input and to respect the visitor's consent choice. Your privacy adviser should confirm the settings meet your obligations.
Will tests slow the site or harm SEO?
A badly installed script can cause flicker and delay. We load it as the vendor recommends, keep tests short-lived and avoid showing crawlers different content from visitors.
What do we keep afterwards?
The research findings, the log of every test, the code for winning versions and the accounts themselves.
Share two numbers
Monthly visitors and monthly conversions on the page that matters most. From those we can say whether testing is realistic. We get back to you within one business day.