What it is.
Every day it checks key revenue and usage metrics, including churn MRR and feature adoption, against a statistical baseline instead of a hand-set threshold. An unusual reading becomes a structured incident.
The incident is logged to a central database, the product team gets a Slack message and an email, and an AI step attaches context and a root-cause suggestion. A daily health report goes to leadership.
Metrics live in dashboards nobody opens, so this comes to you instead.
What you get.
- Daily checks on revenue and usage metrics, including churn MRR and feature adoption.
- Anomaly detection against a statistical baseline instead of a hand-set threshold.
- A structured incident record per anomaly, in a central database.
- Slack and email alerts to the product team, with AI-generated root-cause context.
- A daily health report for leadership.
How to set it up.
Pick the four or five metrics that would genuinely change a decision, since monitoring everything produces noise.
Point the Postgres nodes at the tables holding those metrics, and check the queries by hand first.
Let the baseline build before you trust an alert, because a statistical baseline needs history.
Set the incident database up with an owner field, so every incident has somebody attached.
Read the root-cause suggestions skeptically for the first month; they are a hypothesis and not a finding.
Tune sensitivity until you get a small number of alerts you act on.
Use cases
Treat revenue like a monitored system
Check churn revenue and adoption daily against the recent baseline.
Catch the anomaly, not the trend
Alert on the deviation instead of waiting for a monthly review.
Get a first guess at the cause
Read the analysis attached to the alert before you open the dashboard.
Best for
Metrics nobody checks
Churn MRR moves quietly and gets noticed at the monthly review, which is weeks late.
Product and growth working together
An incident record with an owner is the format both teams already understand.
Leadership reporting
A daily health report replaces the ad hoc request for numbers.