Customer Success
Health Score Builder
A weighted health score framework, tested against representative accounts.
The problem
Most health scores reuse a framework found in an article, shipped with the CRM or borrowed from a neighbouring company. The weights get set in a meeting, and what the combination produces is measured nowhere. Once the score ships, teams find every account green or the alerts warning of nothing, and the metric loses its credibility inside a quarter.
Nobody looks at the score before releasing it, even though weighting has to be judged on cases. What a framework does to an engaged but silent account, to a noisy support account that is nonetheless growing, to an account whose sponsor has just left, shows up on the first simulation.
Health Score Builder
How to read the result
Input
- Four signal families, each scored 0 to 100 for every profile: product usage, support, relationship, contractual
- Weight per family, rescaled to 100% when the score is computed
- Thresholds separating the green, amber and red bands, and configurable vetoes per family (relationship, contractual): at or below a rating threshold the green band is barred and a profile that would be green turns amber
- Representative account profiles to run the score against
How to read the result
- The simulation names the profiles that change band and the signal that moved them, vetoes included
- Every profile in one band: weights too flat, or thresholds misplaced
- Each family's contribution says what the framework measures
- One choice at the end: freeze this version, or rework the weighting before it ships
Limits
- What comes out here is a causal hypothesis. No renewal history has confirmed or contradicted it, and nothing guarantees its bands separate anything at all.
- Logins, sessions and active features are heavily correlated: adding them counts one piece of information three times and drowns out relationship and contractual signals.
- Thresholds are calibrated on the current shape of the book. A wave of small accounts or strong seasonality shifts them with no line of the framework having moved.
- The score rates a state and never a direction: an account sitting stably at the bottom of a band and one falling for three months show the same value.
A scoring framework is re-examined after each renewal cycle, on the accounts it classified wrongly. It is independent of the churn risk calculator's framework: inverted scale, its own families, weights and thresholds, and no score converts into the other. An unreadable entry leaves the last valid value in place with a message, and any value brought back to a limit is flagged. Numbers are read according to the language of the page: comma or point in French, point in English, comma in German and Spanish.
In the glossary
A first thirty-minute conversation
The tool gives the order of magnitude. The decision that follows can be discussed in thirty minutes.