Funnel conversion rate: count each step

A funnel is a chain of counts, so an error early in the chain carries forward. A thank-you page that fires its tracking event on refresh counts the same person twice, and the next step looks like it converts better than the one before it. A checkout event that also fires on the cart page makes checkout look busier than it is. A sign-up counted on a second domain can go missing altogether. Each of these produces a rate that looks like a finding, and none of them reflects a visitor's choice. Treat any step with more visitors than its parent as a counting problem first.

• Each step's count, for the same dates, from the first page to the goal
• A later step with more visitors than the step before it, which points to a counting fault
• The event behind each count: does it fire on refresh, or on a page it should not reach?
• Domains: does the step after a sign-up or payment stay on the same domain?
• Filters: are internal and team visits excluded the same way at every step?

Two counts make a rate only when they cover the same visitors in the same window. A step counted by session and the next counted by user, or a step counted on the day of the visit and the next counted on the day of the purchase, produces a rate that moves for reasons that have nothing to do with buyers. Pick one date range, one unit and one definition for each event, and write them next to the rate. Then the rate can be compared with the same rate the following month, and a change can be traced to the page rather than to the counting.

Once the counts are clean, find the largest loss between two neighboring steps. That pair, not the overall total, is where to look. Take the stop point for that pair from a few simulated visitors, then check the screen they left on against the real counts. If the simulated visitors stop where the counts drop, you have a candidate cause. If they get through, the loss may come from a reason a test cannot show, such as price or timing, and the page is not the place to start.

Before you show a funnel rate to anyone, check that its counts agree with each other and with another source. A sign-up count should roughly match the new accounts in your own database for the same days. A payment count should match the orders. If they disagree, the funnel is measuring the tracking, and fixing the page will not change the rate. Keep a note of each disagreement and its cause, so the next person who reads the rate knows which counts were checked.

Tag managers, consent banners and checkout updates change which events fire, and a new one can make a step look better or worse without any change in buyers. Check the counts for each step after every release that touches tracking. Run the same path through a test account behind a login, and count the events it fires. Send each traffic source through its own run before you compare rates across them.