What it does
Feature Adoption Reporter measures what happens after a feature ships, which is the part almost nobody schedules. For every released feature it computes reach across your whole base, penetration within the segment it was actually built for, repeat-use as a stickiness signal, and attributed business impact. Then it does the thing that makes the report useful: it separates the two problems a low number could mean. Few people found it but those who did came back is a discovery and distribution gap — a marketing and in-product placement problem. Many tried it and few returned is a value and retention gap — a product problem. Those go to different teams, and conflating them is how a feature gets relaunched when it should have been fixed. Three guards keep the verdict honest, and all three matter. A feature released fewer than thirty days before the cut-off is marked too early rather than judged. A low-usage feature carrying a material share of portfolio business impact is classed strategic-low-usage and is never an automatic sunset — which is exactly the compliance or enterprise feature that a naive usage report would recommend killing. And a rate above 100% is clamped, flagged in warnings and excluded from the portfolio averages rather than quietly skewing them. The input handling is unusually careful too. Every value you supply passes validation once, and nothing it changes is silent: a dropped row names its exact field and expectation, an empty portfolio is treated as absence of data rather than health, and the bundled sample only runs on an explicit request and says so in the report. Each step is callable alone — `list_sunset_candidates` for the kill list, `summarize_adoption` for the totals, `build_recommendations` for the actions, `build_adoption_report` for the whole dashboard with no model involved at all. No connectors and no credentials: you supply the portfolio. Built for product managers, product marketers and customer success.
Example prompts
- Report adoption for the features we shipped in the last two quarters
- Which features should we sunset?
- Is this a discovery problem or a value problem?
