You ship a feature. Launch day adoption climbs from 15 percent on day one to 20 percent by day three, and everyone celebrates. A week later, adoption is still 20 percent. The curve peaked and went flat. Either your 20 percent absolutely love it, or you have a discovery problem, and you do not know which. That gap between knowing a metric and understanding what it means is what the Product Health Dashboard Agent closes. It runs every Tuesday at 9 AM and goes deeper than daily numbers.
Four dimensions, every Tuesday
The agent connects to your analytics platform and pulls three weeks of events, journeys, cohorts, and feature usage, then analyzes four things.
Feature adoption curves: for features released under 30 days, the day 1, 3, 7, 14, 30 trajectory against your historical baseline. Cohort retention: this week's new cohort compared to cohorts from 4 and 8 weeks ago, so you see whether new users are getting stickier or less sticky. Engagement depth: for each feature, three tiers, surface users who tried it once, regular users at 2 to 5 uses, and power users at 6-plus. Performance trends: API latency, error rates, page load, all trended week over week from DataDog, New Relic, or Sentry.
The output is a structured report: an executive summary, adoption metrics by feature, cohort retention analysis, engagement depth per feature, performance trends, A/B test results, a usage-anomaly deep dive, and 2 to 3 strategic insights for the roadmap.
The one insight that changes decisions
The move that matters is distinguishing not valuable from not discoverable. Instead of "adoption is low," you get "adoption is 20 percent and flat, but power users have 40 percent higher retention than the overall cohort, which suggests the feature is valuable but hard to discover." That flips the decision. Instead of killing the feature, you improve discoverability. Instead of building something new, you optimize what exists. The way you read it: if power-user retention runs well above overall, the feature is valuable to people who reach it and the problem is discovery. If power-user retention is flat or below average, you have a value problem.
What it changes
You make adoption calls on data, not gut. You stop killing valuable-but-under-discovered features and stop investing in features with zero power-user engagement. You see cohort retention degrading in the weekly digest instead of at quarterly review. You catch P95 latency creeping up on Monday instead of from customer complaints Friday. And your A/B test results start driving roadmap decisions instead of just closing out.
Complete your Claude MCP setup so Amplitude or Mixpanel, a monitoring tool, and Statsig or LaunchDarkly are connected. Then open Claude and ask it to build this week's dashboard: feature adoption, retention cohorts, support trends, and deployment frequency. Run it once this Tuesday and see which decision it changes first.