What it does
Onboarding Funnel Optimizer answers the question every product manager eventually has to answer out loud: activation is stuck at 18%, so where exactly are we losing people, and what do we do about it? You give it the activation funnel — users per step, in order, pasted straight from your analytics — and optionally the session-recording notes your team has collected. It computes per-step conversion and drop-off, identifies the single highest-loss transition rather than a vague weak area, corroborates it against the session signals for that step, and produces three ranked A/B test hypotheses aimed at that specific transition. The three hypotheses are the deliverable. Most funnel tools stop at "step 3 is bad", which leaves the hardest part — deciding what to actually test — with you. These come ordered by expected lift and tied to the loss point the numbers identified. The privacy handling deserves a mention because session recordings are personal data. Free-text observations are masked before they are stored, shown, or sent to the model: emails, phone numbers, long digit runs that look like card or ID numbers, and IP addresses are all redacted. `mask_session_signals` exposes that masking as its own tool, so you can clean a batch of notes even if you are not running the analysis. Each step is callable alone: `compute_conversions` for the per-step numbers and the overall activation rate, `find_biggest_dropoff` for the single answer, `suggest_ab_tests` for the plan. Every metric is derived from the funnel you provide. The language model only narrates, session text is scanned for prompt injection before it reaches the prompt, and a template fallback keeps the agent working with no model configured. No connectors and no credentials — nothing is fetched and no request leaves. Built for product managers, growth and UX teams.
Example prompts
- Our activation is stuck at 18% — where are we losing people?
- Which step in this funnel loses the most users?
- Give me three A/B tests to fix our worst drop-off
