Conversion rate, computed properly

Conversion rate is the most quoted number in online business and the one most often computed wrongly, usually by dividing the wrong two things and then comparing the result with somebody else's.

Which two numbers

Outcomes divided by the people who could plausibly have produced one. The second half is where it goes wrong: dividing orders by all site visits mixes people who came to read an article with people who came to buy, and the resulting number means nothing and moves for reasons you cannot trace.

So compute it per path. Visitors to a product page who ordered. People who started the checkout who finished. People who landed from one campaign who enquired. Each of those is a number you can act on; the site-wide figure is an average of unrelated things.

Why comparing with industry averages is a mistake

Because the denominator differs. A site with a large blog has a low site-wide rate and may convert its product-page visitors well. A site that buys narrow, high-intent traffic has a high rate and may be losing money. The published averages aggregate across both, and across sectors, prices and definitions of conversion, so a comparison tells you nothing about your own position.

Your own previous figure, for the same path, over a long enough period, is the only comparison that carries information.

The order to look for problems

1. 01Check the path step by step rather than the total. Product page to basket, basket to checkout, checkout to payment, payment to confirmation. A large drop at one step is a defect you can go and look at.
2. 02Check it by device. A rate that is fine on a computer and poor on a phone is a layout or a form problem, and most of your traffic is on the phone.
3. 03Check it by source. A poor rate from one channel usually means the promise in the advert does not match the page.
4. 04Only then consider the page itself. Most conversion problems are found in the first three checks, and most effort goes straight to the fourth.

What moves it, in rough order of effect

• Matching the page to the promise that brought somebody there. Larger than any design change.
• Removing surprises: a cost, a step, or a requirement that appears late.
• Answering the question that stops the purchase, which is usually about fit, timing or what happens if it is wrong.
• Fewer form fields and no forced account.
• Speed, particularly on a phone and particularly on the step where somebody is paying.

Notice that none of those is a button colour. The things with the largest effect are about what the page says and what it asks for, not how it looks.

A low rate is sometimes correct

Not every low conversion rate is a problem, and treating it as one leads to changing pages that were working. Three cases where a low figure is the right answer.

• The purchase is considered. Somebody buying something expensive visits several times over weeks. A first-visit conversion rate for that product is measuring the wrong moment, and the useful figure is enquiries per visitor or returning visitors who convert.
• The traffic is informational by design. An article answering a general question attracts people who are not buying anything. That is how search traffic works, and the article's job is to be found and to lead onward, not to convert.
• The audience is wrong rather than the page. A channel sending people who were never going to buy produces a low rate that no page change will fix. The decision there is about the channel.

Which is why the rate has to be read per path. A site-wide figure mixes all three of these with the genuinely broken cases, and the average tells you to improve something without saying what.

The rate that should always be high, and is the one worth attacking first, is the last step: of the people who started paying, how many finished. A large drop there is never correct and is almost always mechanical.