# Your Customer Health Score Is Probably Astrology: Here's How to Test It

DevFeed: [Your Customer Health Score Is Probably Astrology: Here's How to Test It](<https://devfeed.tech/articles/your-customer-health-score-is-probably-astrology-here-s-how-to-test-it-102743.md>)

Original publisher: [Read original article](<https://hackernoon.com/your-customer-health-score-is-probably-astrology-heres-how-to-test-it?source=rss>)

Author: Malek Damiri

Published: 2026-10-27T21:19:28Z

Content type: tutorial

Language: en

Sources: [HackerNoon](<https://devfeed.tech/sources/hackernoon.md>)

Topics: [Data Quality](<https://devfeed.tech/topics/data-quality.md>)

Tags: [analytics](<https://devfeed.tech/tags/analytics.md>), [backtest](<https://devfeed.tech/tags/backtest.md>), [churn](<https://devfeed.tech/tags/churn.md>), [customer-health-score](<https://devfeed.tech/tags/customer-health-score.md>), [customer-retention](<https://devfeed.tech/tags/customer-retention.md>), [customer-success](<https://devfeed.tech/tags/customer-success.md>), [data-science](<https://devfeed.tech/tags/data-science.md>), [how-to](<https://devfeed.tech/tags/how-to.md>), [saas](<https://devfeed.tech/tags/saas.md>), [snapshots](<https://devfeed.tech/tags/snapshots.md>), [sql](<https://devfeed.tech/tags/sql.md>), [startups](<https://devfeed.tech/tags/startups.md>)

## AI overview

Test a customer health score against point-in-time snapshots and subsequent churn, since comparing today's scores with past churn can make the score look predictive after the fact. The article's SQL labels whether accounts churn within 90 days, compares churn rates by score band, and calculates AUC against individual signals. In its simulated data, red accounts churn 1.5 times as often as green ones, but the score's AUC is 0.556; a usage-drop rule scores 0.643, while support tickets score 0.434. These results illustrate the method, not expected outcomes for other customers. The article recommends testing inputs individually, comparing the score with a simple baseline, calibrating bands to observed churn rates, and publishing quarterly backtests. It cautions that few churn events, renewal timing, and CSM interventions can distort results.

## Source excerpt

Malek Damiri shows how to backtest a customer health score in SQL: point-in-time snapshots, churn by band, and AUC against a simple baseline.