Conversion Rate Optimisation: Where Should You Start?

Optimisation is a measurement job; if the measurement is wrong, every decision rests on a wrong foundation. The first task is to write down what a conversion is: a sale, a form submission, a phone tap, a booking. If there are several goals, one is set as the main conversion and the others as supporting ones.

Next, the tracking setup is tested. Fill in the form yourself, try the purchase step, and confirm that the action appears in your analytics tool as a single conversion. The same conversion being counted twice, or not at all, is surprisingly common.

Finally, split the data: device, traffic source, new and returning visitors. The overall conversion rate is usually an average and hides the real information. A low rate on mobile and a high one on desktop says the problem lies in the mobile experience, not in the design as a whole.

Divide the conversion path into steps: landing page, product or service page, basket or form, completion. At each step look at how many people move on to the next. The step with the biggest loss is where the work begins.

One small distinction matters here: a big loss is not always a big opportunity. A loss on the landing page may come from an ad that brings the wrong audience, and then the answer is sought in traffic, not on the page. You need to separate a loss caused by an obstacle on the path from one caused by the wrong visitor.

Numbers tell you where, not why. For the why you need behaviour tools such as heatmaps and session recordings or, in the simplest form, five real people using the site while you watch. A button nobody taps, a price nobody reads or a form field nobody understands becomes visible this way.

A hypothesis is a one-sentence prediction built on an observation and an expectation: 'Mobile visitors leave because the form is long; if we cut the fields, completion will rise.' A change without a hypothesis teaches nothing, whatever the result.

A good hypothesis has three parts: what we saw, what we will change, and what we expect to move. This structure also makes the result easier to read when the test ends. If the expectation did not hold, the hypothesis was wrong, and that is information too.

• Observation: what did you see in the data or the recordings?
• Cause: what is the most likely reason for this behaviour?
• Change: how will you alter a single element?
• Measure: which indicator do you expect to move, and in which direction?

Not every idea is worth the same. Rank ideas with three questions: how large could the effect be, how many people does it touch, and how easy is it to carry out? An easy fix at the busiest, most leaky step of the path comes before a bold redesign of a corner of the home page.

The first gain usually comes not from a new idea but from removing an obstacle: an unnecessary form field, a vague button label, a slow page, an element that overflows on mobile, an unexpected extra charge at payment. It makes sense to remove the obstacle first and persuade afterwards. You can measure page speed with our free site speed test.

In an A/B test, half the visitors see the current page and half see the changed one. Only one element should be changed; otherwise you cannot know which change made the difference. The test should run long enough to cover every day of the week, because weekday and weekend behaviour differ.

Stopping early is the commonest trap. In the first days one variant looks ahead and the test is halted, but a short-term gap is often chance. Decide before you start how long the test will run and how many visitors it needs, and do not judge the result until you reach that point.

Without enough traffic, an A/B test gives no meaningful result. For such sites the route is different: make the change, compare equal-length periods before and after, and back it up with behaviour recordings. This method is weaker than a test, but it is better than false certainty.

The commonest mistake is entering optimisation without verifying measurement. If conversion is counted wrongly, every test chases a wrong number. The second is changing several elements at once; the third is applying another site's working tip without asking your own audience.

There are also cases where optimisation is not the starting point. If very few visitors come to the site, the problem is traffic, not conversion, and visibility comes first. If the offer itself is unattractive, or the product or price does not suit the audience, the result stays limited however well the page is fixed.

• Starting without testing conversion tracking
• Changing several elements in the same test
• Stopping the test on the first day that looks good
• Copying another site's tip without adapting it to your audience