Funnel Cohort Analyst

Funnel Cohort Analyst answers the two questions every product and growth team asks about a funnel: where are people dropping out, and do the ones who make it through actually stay. You hand it your event data and it computes both — step-by-step conversion with the drop-off at each stage, and a cohort retention matrix showing how each signup cohort behaves over time. Every number is derived deterministically. A language model describes what the numbers show, but it cannot invent one or change one, and without a model available the output falls back to a plain template — so the figures are the same either way. On a metric that will end up in a board deck, that distinction is the whole point. The privacy handling is deliberate. User identifiers are used for counting and aggregation only and are never echoed back, and only aggregated numbers ever reach the language model. Your raw event stream does not leave the deterministic core. Event data is treated as data rather than instruction, and runs are idempotent — the same events always produce the same funnel. `analyze_funnel` returns the maths alone with no narration, which is what you want when the output is feeding a dashboard rather than a human. `run_full` adds the written read of what changed and where the loss is concentrated. There is no warehouse to connect and no credentials to configure — you pass the events in. That also means you can run it against an export from any source, not just the ones a connector would have supported. Built for product managers, growth teams and analysts who need a funnel read without waiting on a data team.