Thresholds come from churn history
Round numbers carry no information. Derive the cut points instead.
1. Pull every account from the last two years with its outcome, renewed or churned. 2. Recompute the score as it stood 90 days before each renewal date, not as it stands today. 3. Group accounts by score decile and calculate the observed churn rate in each decile. 4. Find the decile where churn rises sharply. That boundary is the red threshold. 5. Set the green threshold where churn falls to the base rate and treat the middle as amber.
The output is a band with a number attached. Instead of "red means unhealthy," the team can say that accounts below the cut point churned at a specific observed rate, which is what makes a save play worth funding.
The trade the threshold makes
Every cut point balances two errors. A low threshold flags few accounts and misses churn. A high threshold flags many accounts and buries the team in false alarms.
| Threshold set too low | Threshold set too high |
|---|---|
| Short list, high precision | Long list, low precision |
| Real churn arrives unflagged | Team stops trusting the flag |
| Looks efficient in a report | Consumes capacity on healthy accounts |
Bands are routing rules, not labels
A band should map to a specific action with an owner. Red routes to a named save play with an executive touch. Amber routes to an adoption intervention. Green routes to expansion review. If two bands trigger the same email, they are one band.
Watch the direction of travel as well as the position. An account crossing from green into amber is more urgent than an account that has sat at the bottom of amber for a year, because the crossing means something changed recently and is still changing. The same logic governs deal slippage, where the strongest signal is a close date moving rather than a close date sitting far out.
Recalibration
Thresholds drift because customer bases drift. New segments, a pricing change, or a product release all alter the relationship between score and outcome, and a threshold frozen at launch slowly describes a customer base that no longer exists. Rerun the decile analysis every two quarters, and treat any change in the scoring inputs as an automatic trigger to rerun it early. Thresholds validated this way also make the score usable in a forecast, since a band with a measured churn rate can be applied to net revenue retention projections rather than sitting on a dashboard as a color.
Frequently Asked Questions
What health score counts as at risk?
The cut point should come from your own data, not from a convention. Pull two years of accounts, plot the actual churn rate at each score decile, and find the point where the curve bends upward. That bend is the threshold. Setting red at 40 because 40 looks low produces a band that has no relationship to cancellation.
How many bands should a health score have?
Three is enough for routing and two is enough for triage. More bands create meetings about whether an account moved from amber to light amber. The bands exist to assign work, so their number should match the number of distinct plays a team can actually run.
How wide should the at-risk band be?
Wide enough to fill the intervention capacity you have and no wider. If a CSM team can work 30 accounts a month, a red band that flags 200 is a list nobody works. Size the band to capacity first, then improve the score until the same capacity catches more of the churn.
How often should thresholds be recalibrated?
Every two quarters, and immediately after any change to the scoring inputs. A threshold is a statement about the relationship between a score and churn in a specific customer base. When the product, the pricing, or the segment mix moves, that relationship moves with it.
Put these metrics to work
ORM builds custom revenue forecast models that turn concepts like health score thresholds into prescriptive action for your team.
Schedule a Demo