Most CRO programmes fail because they start with a list of best practices instead of finding out where intent is actually lost.
Ask a room what to test and you will get opinions immediately: the button copy, the hero image, the number of form fields. Ask the same room where visitors are actually giving up and the answers get vaguer. That gap is why so many conversion programmes produce a lot of tests and very little revenue.
Conversion work is research first and experimentation second. The testing is how you confirm what the research suggested — it is not how you find out what is wrong.
Measure before you test
A surprising share of conversion work is built on top of broken tracking. Duplicate events, a thank-you page firing twice, a form success that was never instrumented at all, a consent banner suppressing analytics for a third of visitors. If the baseline is wrong, every result afterwards is noise dressed as evidence.
The first week of any serious programme should be spent proving the numbers can be trusted: submit every form yourself and confirm exactly one event arrives, reconcile the analytics figure against the CRM or the inbox, and check what happens to measurement when someone declines cookies.
It is unglamorous and it is the highest-leverage thing you will do. A programme that discovers in month four that its baseline was inflated by duplicate events has to discard everything it concluded in months one to three.
Find the real drop-off
Funnel analysis tells you where. Session recordings and form analytics tell you why. Customer interviews tell you what they were worried about that your page never addressed.
Use all three, in that order. The quantitative view narrows the search to a step; the qualitative view explains it. Neither works alone — analytics will tell you 60% abandon at the quote form without telling you it is because the first field asks for a policy number nobody has to hand.
In our experience the loss is rarely on the page everyone is arguing about. It is more often a mismatch — an ad that promised one thing and a landing page that talks about another — or an unanswered objection about price, timescale or risk that the page simply never mentions.
Write hypotheses, not ideas
A test worth running has a stated belief behind it: because [evidence], we expect [change] to produce [effect] on [metric]. "Because 60% of exits happen on the pricing step and interviews cite uncertainty about what is included, we expect an itemised inclusion list to raise step completion" is a hypothesis. "Make pricing clearer" is a preference.
The difference matters after the test rather than before it. A hypothesis that loses still teaches you something about your customers, because you learn the belief was wrong. A preference that loses teaches you only that this particular execution did not work.
Test fewer things, properly
A test needs enough traffic behind it to mean something. Running six underpowered tests a month feels productive and teaches you nothing; running two properly powered ones tells you something you can build on. Work out the sample size you need before launching, and if the answer is that the test would take five months, that is a finding — test somewhere with more traffic, or make a bolder change.
Change one thing at a time. If you rewrite the headline, reorder the page and add a testimonial in the same variant, a win tells you the page got better but not what to do next.
Record the losers as carefully as the winners. Most organisations accumulate a folder of successful tests and no memory at all of what has already been tried and failed, which is how the same idea gets tested three times in four years.
When not to test at all
Below a few hundred conversions a month, A/B testing will rarely reach significance in a useful timeframe. That does not mean conversion work stops — it means the method changes to qualitative research, obvious-fault removal and sequential before-and-after comparison, with the honest acknowledgement that the last of those is weaker evidence.
Fixing an obviously broken mobile checkout does not need a test. Some changes are large enough and self-evident enough that measuring them is a formality; spending six weeks proving them costs more than the certainty is worth.
The compounding argument
Doubling conversion rate has the same effect on enquiries as doubling ad spend, except you pay for it once and it keeps paying. It also improves every channel simultaneously, because organic, paid, social and direct traffic all arrive at the same pages.
That is the whole case for doing this work before buying more traffic — and the reason conversion work usually belongs earlier in a growth plan than it is placed.
Common questions
- What is a good conversion rate?
- There is no useful universal benchmark, because the number depends on traffic source, price point, purchase complexity and how a conversion is defined. A better test is relative: how does this page compare with the same page last quarter, and how does it compare with your other traffic sources? Chasing an industry average is how businesses conclude a well-performing page is failing.
- How much traffic do you need for A/B testing?
- Enough to detect the size of change you care about, which for most sites means several hundred conversions per variant rather than several hundred visitors. Below roughly a few hundred conversions a month, tests rarely reach significance in a useful timeframe, and qualitative research plus fixing obvious faults will produce more improvement than experimentation.
- What should you fix before running conversion tests?
- Tracking accuracy first — duplicate events and uninstrumented forms invalidate everything measured afterwards. Then obvious faults: broken mobile layouts, forms that fail validation silently, pages that take several seconds to become usable. Testing around a broken foundation measures the breakage rather than the change.