Building a health score: factors, weighting, pitfalls
A health score aggregates several factors (usage, support, relationship, finance) into one health indicator per account. For it to guide action, three conditions: factors from crossed sources, a weighting confronted with the reality of departures, and an explanation attached to every score. A score without a cause produces alerts that end up ignored.
In short
- A single-source health score (logins, for example) generates too many false signals to be followed.
- The initial weighting matters less than confronting it regularly with real departures and real saves.
- The score must arrive with its cause: an unexplained red dot does not trigger an action, it triggers a sigh.
What a health score must do, and not do
The role of a health score is not to predict the future with certainty, it is to sort the portfolio: name the accounts that deserve attention today, and free up time on those doing fine. Any ambition beyond that, a "predictive" score to the decimal point, produces false precision.
It must not replace judgment either: a score is a conversation starter, not a verdict. Teams that use it well rely on it to decide where to look, then actually look.
The families of factors
Robust factors fall into four families, each living in a different tool. That is precisely why a single-source score fails: it only sees a quarter of the picture.
Product usage
Frequency, depth (key features), breadth (active users). The trend matters more than the absolute level.
Support
Volume and severity of tickets, resolution times, recurring pain points on critical capabilities.
Relationship
Rhythm and tone of exchanges, meetings held or postponed, engagement of the sponsor and key users.
Finance
Late payments, downgrade requests, disputed invoices. Often the latest signal, but the surest.
Weighting: start simple, confront with reality
The classic trap is spending weeks calibrating theoretical weights. The opposite method works better: start with a simple weighting, then confront it with the facts. Every departure the score did not flag and every alert that did not materialize is a calibration data point.
This confrontation must be a ritual, not a yearly project: a score whose weighting has not moved since its creation describes yesterday's portfolio.
The classic pitfalls
Four mistakes come up in most abandoned health scores.
The single-factor score
Based on logins alone, it confuses activity with satisfaction: a very active account can be at its limit, a quiet account can be perfectly settled.
The frozen weighting
Calibrated once at launch, never confronted with real departures. The score drifts silently and the team stops believing in it.
The score without explanation
An account turns red with no visible cause. Nobody knows what to do, so nobody does anything.
Automatic thresholds without context
A mechanical alert at every threshold crossing drowns the team in false positives and discredits the real alerts.
Static health score vs living health score
The difference is not in the formula, but in what surrounds the score: its sources, its maintenance and its explanation.
How Phano helps you
Phano replaces the health score you have to maintain with an explained diagnostic: every night, six analysis techniques cross your connected sources and confront their results. The Customer Success Manager receives the accounts that deserve attention with the cause and the proposed action; the Account Manager sees the same diagnostic weighted by account value. No weighting to calibrate, no formula to maintain.
Go further
Your data stays yours
Security, isolation and compliance by default. Not an add-on.
Per-organization isolation
Every organization is partitioned by Row Level Security at the database level, with a double membership check server-side.
AES-256 encryption
All data is encrypted at rest across the entire database, and in transit.
Anonymization before AI
Emails and phone numbers are masked before any model call. The original data never leaves our European servers.
GDPR compliance
Export and deletion of your data on demand. Transfers outside the EU governed by Standard Contractual Clauses.
Frequently asked questions
Which factors should a health score include?
Four families: product usage (frequency, depth, breadth), support (volume, severity, delays), relationship (rhythm and tone of exchanges, sponsor engagement) and finance (late payments, downgrades, disputes). A reliable score crosses several families at least: each one only sees part of the account.
How do you weight a health score?
Start with a simple weighting rather than a theoretical calibration, then confront it regularly with the facts: departures the score did not flag, alerts that did not materialize. The score's errors are what calibrate it. A weighting never revisited produces a score that drifts silently.
Why is my health score unreliable?
The most frequent causes: a single source (often usage), a weighting frozen since creation, scores without explanation and mechanical thresholds that multiply false positives. The symptom is always the same: the team stops looking at the score.
Is a health score enough to predict churn?
No. A score sorts the portfolio and names where to look; predicting churn also requires the probable cause and the action window, account by account. A score without an attached cause flags without enabling action, which amounts to observing the risk rather than treating it.
A Customer Success tool that tells you what to do.
Connect your CRM. The first diagnostic arrives the same day, in your tools.