Building a customer health score your team believes
A health score earns its keep when the retention desk acts on it without arguing. That takes a short list of inputs, an honest weighting, and a feedback loop from every save.
Priya Raman//7 min read/Analytics
Every health score starts the same way. Someone builds it, the desk works it for two weeks, half the flagged accounts turn out to be fine, and the score quietly stops being opened. The fix is smaller inputs and faster feedback.
Start with four inputs
Service reliability, payment health, sentiment, and tenure. Four inputs are enough to sort a book of 12,000 accounts into a workable order, and each one is defensible in a conversation with a service manager.
| Input | Definition | Weight |
|---|---|---|
| Service reliability | Missed visits, skips, reservices in 180 days | 35% |
| Payment health | Declines, days past due, autopay status | 30% |
| Sentiment | Latest survey score and review text | 20% |
| Tenure | Months on the plan, with month 14 flagged | 15% |
A score is a queue, not a verdict. Its only job is deciding which ten accounts get a call today.
Close the loop on every save
When the desk works an account, record what happened and whether the account stayed 90 days. Without that record, the score never learns which signals matter in your market and which ones were noise.
Software that holds the account data, the risk score, and the save outcomes together closes that loop by default. The advantage over a homemade model is that outcomes get captured automatically rather than by discipline.
A score nobody revisits after a quarter is a spreadsheet with better formatting.
Review the score every quarter
- Compare flagged accounts against accounts that actually cancelled.
- Drop any input that adds nothing to the prediction.
- Raise the weight on the input with the longest lead time.
- Ask the desk which flags felt useful, since they hear the reasons.
Ship a rough score this month. A score at 70 percent accuracy directs attention far better than a perfect one that arrives in the spring.
