Fix a leaking signup flow
Counting tells you where people leave. Watching tells you why. Doing one without the other is how a team spends a quarter redesigning the wrong step.
The situation
You know the flow loses people, because the number who finish is much smaller than the number who start. What you do not know is which step, or what happens there.
Guessing is expensive: every redesign is a sprint, and a redesign of a step that was working makes things worse in a way nobody notices for a month.
How it is done
1. Make each step a goal. A goal is a page load or an event you have named. Three or four of them across the flow is enough. More than that and you are measuring the instrumentation rather than the flow.
2. Put them in a funnel. Now you have how many reached each step and how many did not. The gap between two steps is the number worth caring about; the total at the end is not.
3. Turn on recordings, sampled. You do not need every session. A sample of the step that leaks is enough to see the pattern, and sampling is the difference between a useful feature and an expensive one.
4. Heatmap the step. Clicks on things that are not links, and a scroll depth that stops above your button, are the two findings that come up most often and are invisible to any count.
5. Annotate the fix. A dated note on the chart is what turns "conversion improved in March" into "conversion improved when we changed the address form".
Where this stops
It shows what happened, not what somebody thought. A recording of a person hesitating tells you where the hesitation was, and nothing about why. For that you have to ask them. See hear from visitors before they leave.
It cannot see your server. A step that fails because a request timed out looks, in a replay, like somebody who stopped. If the flow is broken rather than confusing, your error logs will find it faster than this will.
Recording has to be on before the problem. There is no way to replay a session that was not recorded, so a flow you care about is one to sample continuously rather than switch on after a bad week.
Other jobs
Prove which marketing brought revenue
Tie the money to the campaign, without the referrer report saying Direct.
Understand what app users do
Screens, sessions, releases and crash-free rates for software people install.
Hear from visitors before they leave
A comment attached to the page it is about, rather than a survey nobody starts.
Report without handing out logins
A read-only link for a client or a board, and a digest that arrives on its own.
Watch a launch day
See traffic arriving as it happens, and know within the hour if it is going wrong.