What is funnel analysis?
The technique for finding the exact step where a flow leaks the most people, so you know what to fix next.
Steps, conversion, and drop-off
A funnel is an ordered list of steps a user is supposed to pass through, top to bottom. For a product it is often something like visit → signup → activate → pay. Funnel analysis puts a number on the narrowing between each pair of steps:
- ·Steps are the events in order. Each user either reaches a step or does not, and reaching step 3 means they also reached steps 1 and 2.
- ·Step conversion is the count who reached a step divided by the count who reached the step before it. It is always between 0% and 100%.
- ·Drop-off is one minus the step conversion, the share who left at that step. The step with the biggest drop-off is the leak.
- ·Overall conversion is the count who finished the last step divided by the count who entered the first. Every step after the leak is capped by it.
A useful funnel report adds two things on top: time-to-convert (how long a step takes, so a slow step shows up even when few people abandon it) and a breakdown by a property like source, plan, or device (so you can see that the drop-off is much worse for mobile than desktop).
A concrete example
Say 1,000 people land on your pricing page this week. Here is the funnel to a paid plan:
| step | reached | step conversion | drop-off |
|---|---|---|---|
| 1 · visit pricing | 1,000 | n/a | n/a |
| 2 · start signup | 400 | 40% | 60% |
| 3 · activate | 120 | 30% | 70% |
| 4 · pay | 84 | 70% | 30% |
Overall conversion is 84 / 1,000 = 8.4%. But the headline number is not the point. The biggest drop-off is step 2 to step 3, activate, where you lose 70% of the people who started signing up. That is the leak. Doubling activation from 30% to 60% would roughly double paid customers, because every step below it is fed by that step. Spending the same effort on the pay step (already 70% conversion) would move far fewer users. Funnel analysis is what tells you to work on activation, not checkout.
Just ask where people drop off
In smolanalytics you track the steps as events (one snippet on the web, one POST /v1/events call from a server), then you do not build a funnel chart by hand. You ask, in plain English, from a dashboard bar or from your own Cursor or Claude over MCP:
The answer is computed from your events by the same deterministic report the dashboard renders, never generated by a language model, so the leak is named exactly, with its real drop-off number, not a plausible guess. A CI agreement test fails the build if the editor's answer ever differs from the dashboard, so the number you are told to act on is the number that is true.
You do not even have to ask. The verdict, on the dashboard and in a morning brief, does this unprompted: it looks across your funnels, finds the step where you are losing the most people, and names that leak as the thing to fix. That is the whole idea, the tool tells you which step of the funnel to work on instead of leaving you to read a chart and guess.