Conversion rate for website: what to compare

A conversion rate is one count divided by another. Change either count and the rate changes, even when the site does not. A visit can mean a session, a person, or a page load that a bot made. A conversion can mean a click on the button, a signup, or a purchase, and a setup can count the thank-you page again each time someone reloads it. Your own team's visits, staging traffic and repeat visits from the same person can all sit in the total. The visitor just sees a page, so the mismatch is invisible to them and to anyone who only reads the final rate.

• Denominator: does your analytics count sessions or people, and does it leave out your own team's visits?
• Numerator: does a conversion fire once per completed signup or order, or on every click of the button?
• Traffic mix: do paid, search and social visitors sit in one rate, or in separate rows?
• Thank-you page: does the conversion fire again when someone reloads it or returns to it?
• Test visits: do staging or preview visits land in the same property as your live traffic?

Write down the two numbers behind the rate before you look at it. Write what a visit is in your tool, whether it is a session or a person, and what a conversion is, in the tool's own event name. Then place a test order and see which events fire. If the event fires on the click but the sale is not finished, your rate counts attempts, not sales. A wrong version is a sign-up event attached to the Continue button. A right version is a single event fired after the account exists. Fixing the event changes the rate without the site changing at all, so do it before any redesign.

A site rate blends visitors who came for different reasons. A visitor from a search for the exact product converts differently from a visitor who clicked a general ad, and both are in one number. If the ad traffic grew this month, the rate can fall with no change to the page. Split the rate by traffic source first, then by device, and compare each row with the same row last month. Only then ask whether the page is the problem. A page that converts search visitors well can still lose the ad visitors who arrived with a different promise. Write down which source you are judging, every time.

Some rate problems that look like page problems are counting problems. A thank-you page that reloads adds a conversion each time. Your own team and the staging site add visits that never had a reason to buy, so they pull the rate down. A bot that loads the page adds a visit and no sale. Check the thank-you page by reloading it once, and see whether the count moves. Filter your team's visits in the analytics tool itself, and keep staging out of the live property. A rate you cannot trust is not a rate to optimize.

The rate moves whenever the traffic, the price or the form changes, so one comparison is not enough. After each change, rerun the same kind of visitors through the same task and compare who got further. Run the task on a phone as well as a laptop, because a checkout that works on one can stall on the other. With paid credits, visitors can also sign in with a test account you provide.