What a health score is trying to do
A customer health score is an attempt to collapse multiple noisy signals about an account's wellbeing into a single number that a customer success team can act on. The idea is that instead of reading every ticket, every feedback submission, and every usage data point for every account, you get a score that surfaces who needs attention today.
That abstraction is useful, but it hides complexity. Every health score is a model — a set of assumptions about which signals matter, how much each one matters, and how they should combine. Two organizations can have very different retention dynamics, which means a health score tuned for one may tell the wrong story for the other. A health score is only as good as the signal selection and weighting behind it.
This is why transparency about what goes into a health score matters. A number without a rationale is just a guess dressed up in authority.
The signals that feed a feedback-driven health score
Rereflect builds its customer health score primarily from signals extracted from feedback. This is a deliberate choice: feedback is one of the richest sources of leading indicators of churn, because customers often express dissatisfaction in words before they express it in behavior (like reduced logins or canceled subscriptions).
The factors that contribute to the health score include sentiment trend, urgency rate, pain point frequency, and recency of engagement through feedback. Each factor is scored relative to the customer's own history as well as against the broader population, so a health score shift means something changed, not just that a particular customer tends to submit a lot of feedback.
- Sentiment trend — is the customer's average feedback sentiment improving, stable, or declining over the recent period?
- Urgency rate — what fraction of their recent feedback has been flagged as urgent or churn-risk?
- Pain point recurrence — are the same categories of complaints appearing repeatedly, or is each piece of feedback about a different issue?
- Feedback recency — has the customer been engaging with feedback channels recently, or has there been a long silence (which can itself be a signal)?
- Factor breakdown — Rereflect shows which factors are dragging the score down or lifting it up, so a health score is never just a black box.
How weights work — and why they are configurable
A health score is a weighted sum of its input signals. The weights determine which signals dominate and which contribute marginally. For some SaaS products, sentiment is the overwhelming predictor of churn — customers who start saying negative things leave quickly. For others, urgency rate matters more. For others still, pain point recurrence is the key variable.
Because retention dynamics differ, Rereflect makes the weights configurable per organization. If you know from experience that recurring pain points are a stronger churn signal in your product than overall sentiment, you can shift weight accordingly. This lets the health score model your business rather than a generic average of many businesses.
The honest caveat: if you are configuring weights without grounding them in observed outcomes, you are making educated guesses. They may be good guesses, but they are guesses. The best weights come from looking at customers who actually churned and asking which signals were elevated in the weeks before they left.
The churn probability and what it actually means
Rereflect surfaces a calibrated 30-day churn probability alongside the health score. Calibrated means the model attempts to express genuine probability rather than a raw score — a 70% probability should mean that, historically, accounts in that situation left roughly 70% of the time.
Out of the box, calibration is based on heuristics rather than your own retention history. The probability is most usefully read as a relative risk ranking: accounts with 80% probability are more at risk than accounts with 40%, and that ordering should be acted on accordingly. As you label churn events in Rereflect over time and trigger recalibration, the probability becomes grounded in your actual data.
Treat the 30-day probability as a prioritization tool. It tells you where to look and what to investigate. It does not tell you what will happen with certainty, and no heuristic model can.
Where to be appropriately skeptical
Health scores invite over-trust. A few places where skepticism is warranted:
- A good health score does not mean no risk — a customer who has not submitted feedback recently may look healthy simply because there is no signal, not because everything is fine.
- A bad health score does not always mean imminent churn — high-verbosity customers who express frequent frustration may have no intention of leaving; they may just be engaged and opinionated.
- Weights configured without outcome data are hypotheses — useful starting points, but not truths until validated against real churn events.
- Health scores derived purely from feedback miss signals in product usage, billing, and relationship data — they are one lens, not the full picture.
Used well, a health score is a triage tool — it helps you allocate your attention across accounts more intelligently than reading every ticket. Used poorly, it becomes a substitute for understanding customers rather than a prompt to understand them better.