Free tool
Health Score Test
Would your health score have warned you about last year's churn? Paste each account's score from 90 days before it churned or renewed, and the test checks whether the score told the two groups apart.
The test runs in this page. Your scores stay in your browser, and nothing is saved.
Sample data
Result
Your score misses churn.
Show the counts
| Score | Churned | Renewed |
|---|
What to fix first
Made with the free Health Score Test at retentionforge.io. Pass marks: 70% of churned accounts below 50 and 70% of renewed accounts above 65, scored 90 days out.
How to pull the data
Two lists from the last 12 months
-
Churned accounts
List every customer that churned in the last 12 months.
-
Their scores
For each one, find its health score 90 days before it churned. If your platform doesn't keep score history, rebuild the score from the inputs as they stood then.
-
Renewed accounts
List the customers that renewed or expanded over the same 12 months.
-
Their scores
Find each one's score 90 days before its renewal date.
Aim for at least 30 accounts in each list, though fewer still gives you a first read.
What passing looks like
- Churned
- At least 70% of churned accounts scored below 50, 90 days before they left.
- Renewed
- At least 70% of renewed accounts scored above 65, 90 days before renewal.
- Separation
- Churned accounts cluster around 30 to 50 and renewed accounts around 65 to 85, with a clear gap between them.
- Gray zone
- Few accounts land between 50 and 65, where nobody can tell whether a score is healthy or risky.
The test also reports how often a renewed account outscores a churned one when you pick one of each at random. Statisticians call it AUC. A common rule of thumb from Hosmer and Lemeshow's Applied Logistic Regression treats 70% to 80% as acceptable and 80% or more as excellent.
Why scores fail
- Too many inputsTwenty or thirty attributes bury the five to eight that track churn.
- Sentiment over behaviorNPS and CSM gut feel describe how customers feel. Churn follows what they do, such as usage and adoption.
- One model for every segmentEnterprise customers leave over ROI and lost sponsors. SMB customers leave when they stop using the product.
- Never back-testedA template with sensible-looking weights goes live before anyone checks it against past churn.
Send it to me
Want me to look at your results?
Only the summary numbers go with your note. Your scores don't.
Next step
Need a score that passes?
The Health score module of a Retention Build picks five to eight inputs from your own churn history and back-tests them before your team relies on the score.