how to see where users drop off

users drop off at the step in your flow where the largest share of them leaves and does not come back, and you find it by laying your steps in order and watching how many people make it from each one to the next. that ordered view is a funnel, and the step with the steepest fall is your biggest leak. the fast way to see it in smolanalytics is to ask, in plain english, "where do people drop off?" and get a real computed answer: the verdict names the exact step and the percentage lost, off your live events, never a number a language model made up. it is open source, a single go binary you can self-host free, or hosted from $9/mo, and the answer is guaranteed by a CI test that fails the build if the ai's number ever differs from the dashboard's.

what a drop-off actually is

a drop-off is the gap between two steps that a user should have crossed but did not. someone lands on the pricing page but never clicks checkout. someone starts signup but never confirms their email. someone adds to cart but never pays. each of those is a drop-off, and every product has a handful of them stacked back to back.

the trap is treating drop-off as one big vague number. "our conversion is 2%" tells you almost nothing, because it hides where the 98% actually left. drop-off is only useful when it is attached to a specific step, because that is the only thing you can go and fix. see funnel analysis for the full definition of the ordered-step view this rests on.

how funnels find it

a funnel is your flow written down as an ordered list of steps, for example: visit → signup started → email confirmed → first action → paid. the funnel counts how many people reached each step, and how many of those went on to the next one. the number that matters is the conversion between adjacent steps, because a low step-to-step rate is a leak and a high one is fine.

read a funnel top to bottom and it looks like a staircase where each stair is shorter than the one above it. that shrinking is expected, some fall-off is normal. what you are hunting for is the one stair that is dramatically shorter than its neighbours. that disproportionate fall is where users drop off, and it is almost never where you guessed.

reading the biggest leak

do not eyeball the whole funnel and pick the step with the fewest people at the bottom. the bottom step always has the fewest people, that is just gravity. the biggest leak is the step with the worst step-to-step conversion, the steepest single fall relative to the step right before it.

an example. say 1,000 people start, 800 finish signup, 120 confirm email, and 100 take a first action. the raw counts drop hardest at the top, but the conversions tell the real story: signup holds 80%, email confirmation holds 15%, first action holds 83%. the leak is email confirmation, not signup, even though signup lost more absolute people. the steepest rate, not the smallest count, is the thing to fix.

this is exactly the read smolanalytics hands you for free. ask "where do people drop off?" from the dashboard bar, or from your own cursor or claude code over mcp, and the verdict names the step and the percentage lost, so you do not have to do the arithmetic or second-guess which stair is the real one. the answer is computed from the same deterministic funnel report the dashboard renders, never generated by the model, so it is the real figure or nothing.

fixing it

once the funnel names the step, the fix stops being a guess. the leak is a specific transition, so you go look at that one transition. if email confirmation is the leak, the questions are narrow: is the email arriving, is it landing in spam, is the link obvious, is the copy asking for something confusing. you change one thing at that step, then watch the same funnel to see if that stair got taller.

that loop is the whole job: find the steepest fall, fix that one step, re-read the funnel, move to the next steepest fall. you do not need session replay or heatmaps or a survey to start, you need to know which step is bleeding and then go fix that step. smolanalytics deliberately does not ship replay, flags, experiments, or surveys; it gives you the funnel, the retention curve, and a plain-english verdict on what to fix next, from one snippet. the features page lays out exactly what it computes.

and because the morning brief runs on the same reports, a new leak that opens up overnight shows up in tomorrow's brief instead of waiting for you to go looking. that is the point: the drop-off finds you.

try it

drop one <script src=".../sdk.js"> and call smolanalytics.init(key, {host}), or run the demo with docker run -p 8080:8080 ghcr.io/arjun0606/smolanalytics, then ask "where do people drop off?" and read the leak it names. it is open source (MIT), a single go binary with no kafka or clickhouse beside it, self-host free forever, or hosted from $9/mo. code is on github.

the answer used to need a data analyst, some sql, and a heavy tool. now you just ask, and the verdict points at the step that is losing you the most users.