Funnel Drop-off
Funnel drop-off is the loss of prospects who leave between one stage of a conversion funnel and the next instead of advancing toward conversion.
Key takeaways
- Drop-off is a residual; it shows loss but never where people went.
- The last recorded step is not always the step that caused the exit.
- Convert loss percentages into recoverable head counts before ranking any fix.
- A step that filters unqualified visitors should drop people, by design.
- Segment drop-off by device and source before blaming the page design.
In depth
Drop-off is a residual: one minus the transition rate for a given step. That makes it a statement about what did not happen, which is why it never tells you where people went. Someone who leaves may have exited the site, switched to a competitor, gone to ask a colleague, or simply run out of time on a phone. The step where the last event was recorded is also not always the step that caused the loss; a demanding question often shows up as an exit one screen later.
Four things reliably raise drop-off at a step: the effort it demands, the sensitivity of what it asks for, how far the payoff still feels, and how long it takes to load. Effort and sensitivity can be traded against perceived value, which is why the same phone-number field is fatal on screen one and tolerable after a result has been shown. Lowering the ask lowers drop-off but also lowers what you learn, so each removed requirement should be weighed against what it was buying.
The working method is to convert a drop-off percentage into a count of lost people, then into an estimate of how many are recoverable. Only then does a fix earn a place in the backlog. Recovery tactics differ by cause: a saved-progress link or an email of the partial result addresses people who ran out of time, while a reworded question addresses people who did not understand it. In a scorecard funnel, per-question tracking narrows the search to a single screen before you start guessing.
Drop-off says nothing about who left, and that omission can invert its meaning. A step that removes unqualified visitors is doing its job even at a steep loss rate, and reducing that loss would hand sales more work for the same revenue. Comparing drop-off across steps of different kinds is also unsound: a question screen and a payment screen are not the same species of ask. And on very small volumes a single bad afternoon of bot traffic can create a drop-off spike that means nothing.
Example in practice
How to measure it
For each transition, keep the entrant count, the advance count, the loss count and the loss share, and sort the list by loss count rather than loss share. Add the time between the last event and the exit: a fast exit suggests refusal, a slow one suggests hesitation or distraction. Those two patterns need different fixes, and only the timestamp separates them.
Then test whether the loss is real or a measurement artefact. Compare the drop against a segment you trust, check that the tracking event fires on the step in question, and look for exits that arrive faster than a human could read the screen. Once the figure is trusted, track it as a weekly series, since a single sharp week usually reflects a release or a traffic shift rather than user behaviour.
Common mistakes
The usual mistake is treating every exit as a defect and stripping the funnel until almost nobody leaves. What follows is a larger list of people who were never going to buy, and a sales team that stops trusting the source. Before removing a step, look at what the people who passed it went on to do. If they convert at a much higher rate, the step is filtering, not leaking, and it should stay.
The second is diagnosing from aggregate numbers alone. A step that looks fine overall can be losing almost everyone on small screens or on one browser, and the average hides it completely. Split every drop-off figure by device, browser and traffic source before writing a hypothesis. Plenty of supposed copy problems turn out to be a form control that does not work under a mobile keyboard.
Frequently asked questions
Is funnel drop-off always bad?
No. Some drop-off is healthy because it filters out prospects who are a poor fit, sparing your sales team wasted effort. The concern is an unusually high loss at one step, which usually points to fixable friction rather than natural filtering.
How do I find where drop-off is happening?
Measure each transition separately by comparing entrants to advancers, rather than looking only at total funnel loss. The transition with the steepest, unexplained loss is where you should investigate and fix first.
What commonly causes drop-off in a quiz funnel?
Frequent culprits are overly long or confusing questions, slow loading, and asking for sensitive details like a phone number too early. Tracking each step lets you pinpoint the exact moment people leave and redesign it.
What is a normal drop-off rate between funnel steps?
It varies too much by step type to give a single figure. A low-commitment click loses far fewer people than a step asking for a phone number, and comparing the two is not meaningful. Use your own history as the baseline: the useful signal is a step whose loss changed, or a step that loses far more than similar steps in the same funnel.
How do I tell abandonment from rejection?
Look at the time on the step before the exit. People who reject an ask usually leave quickly and often without interacting with the field at all. People who abandon tend to linger, start filling something in, or come back later in a new session. The first calls for changing what you ask; the second calls for making it easier to resume.
Can I recover people who dropped off?
Only those you can still contact, which is why the order of your steps matters. If an email address is collected before the heaviest ask, a partial-progress reminder can bring some of them back. If contact details come last, everyone lost before that point is unreachable, and the only lever left is reducing the loss in the first place.
Why does my drop-off spike on mobile?
Usually because of input effort and layout rather than content. Long option lists, free-text fields, small tap targets and forms that jump when the keyboard opens all cost far more on a phone than on a desktop. Check the step on a real device rather than a browser emulator, and count taps required, not fields shown.
Is a high drop-off at the qualification step a problem?
Not necessarily, and often the opposite. If the step exists to separate fit from non-fit, a large loss there is the step working. Judge it by what happens after: if the people who pass convert and close at a much higher rate, keep it. If pass and fail groups behave the same downstream, the step is adding friction without adding information.
How do I find the exact question where a quiz loses people?
Record a step-reached event for every question rather than only quiz start and quiz completion. The counts then form their own mini-funnel, and the question with the largest gap between reached and answered is the one to look at. Without per-question events you can see that the middle leaks but not which screen is responsible.