Everybody has an opinion about where visitors leave. The chart takes an afternoon and ends the argument, which is why it so often does not get built.
Every term used above is defined in our experimentation glossary, and three experiments taken apart step by step sit in the case studies. Longer pieces with the workings attached are in the field notes.
A funnel built on page views counts one person visiting checkout four times as four people. Build it on unique visitors per step or every conclusion after it is wrong.
Do this: Check whether your funnel counts sessions, events or people. Rebuild it on people before reading anything into it.
A step that loses 60% of ten visitors loses six people. A step that loses 8% of twenty thousand loses sixteen hundred. Percentages point at the wrong step surprisingly often.
Do this: Sort your steps by people lost, not by rate. The order usually changes.
Send traffic and baseline rate. We reply with the smallest effect your test can actually detect. One email, no call.
Mobile and desktop funnels often behave so differently that the combined chart describes neither. A step that looks mildly bad can be catastrophic on one device and fine on the other.
Do this: Rebuild the funnel twice, once per device. If the shapes differ, treat them as two funnels from now on.
People leaving a pricing page after seeing the price are not a leak, they are qualification. Chasing that number down usually means hiding the price, which moves the problem later and makes it worse.
Do this: For each drop, ask whether you want those people to continue. If the answer is no, stop measuring it as a loss.
This is teaching material and our own reading of standard practice, not advice for your specific site. Check anything important with your own specialist before you act on it.
Traffic, baseline rate and what you changed. We reply with the minimum detectable effect and whether the result meant anything. No cost, no call.