A health score that predicts something, because it was tested against who actually churned.
Red, yellow, green driven by CSM sentiment was never designed to forecast anything. We design a numeric model from your own renewal and churn history, weight it by what actually correlated with the outcome, validate it on accounts it has not seen, and put it where the CSM plans their Monday.
Most health scores are a mood ring with a legend.
They make the dashboard look managed. They do not tell you which green account is going to churn in ninety days, which is the only question that matters.
Sentiment dominates the inputs
A pulse column filled in by the CSM carries most of the weight. That is recency and rapport, not data, and it is systematically kind to the accounts the CSM likes.
Lagging indicators only
By the time usage drops and tickets spike the decision to leave has already been made internally. The score reads the autopsy, not the vital signs.
No backtest, ever
Nobody has checked whether green accounts renewed at a higher rate than yellow. Until that check exists the score is decoration.
Trajectory is ignored
A 65 that was 80 last quarter and a 65 that was 50 last quarter are different accounts. A single snapshot number hides the only signal that matters.
Nobody acts on it
The score lives in a dashboard tab. It is not ranked, not routed, and not in the view the CSM opens on Monday morning.
Built from your outcomes, not from a template.
A method that produces a model your team can explain to a sceptical CSM and a sceptical CFO, in the same meeting.
Outcome dataset
Week 1Twelve or more months of renewal, churn, contraction and expansion outcomes, joined to product usage, support history, engagement records and commercial data. Thirty outcomes is the floor; a hundred is comfortable.
Dimension design
Week 1 to 2Candidate signals grouped into dimensions that fit your business: value realisation, technical health, relationship depth, commercial posture and trajectory. Product-led and enterprise businesses weight these differently, so we do not assume.
Retroactive scoring and correlation
Weeks 2 to 3Every historical account scored on each dimension as it stood before the outcome. Correlation and lift analysis to find which dimensions actually predicted what happened, and which were noise.
Weights, thresholds, holdout
Weeks 3 to 4Weights set from the analysis, thresholds tuned for the alert volume your team can act on, then validated on a holdout set the model has not seen. If it does not beat the current score on the holdout, it does not ship.
Parallel run and rollout
Weeks 4 to 6The new score runs beside the existing one for two to four weeks, surfaced as a ranked list with the driving signals per account. Alert routing, trajectory flags and the CSM's weekly view built in Planhat. Cutover when the team trusts it.
Deliverables you can point to.
Every item is handed over in your tools, walked through on a call, and accepted against written criteria before the final invoice.
The joined history, with the definition of every input and where it came from.
Dimensions, signals, weights, thresholds and the analysis that produced them.
How the model performed on accounts it had not seen, against the score you use today.
Calculated metrics, numeric score properties, trajectory flags, alert routing and a ranked CSM view.
Side by side with the legacy score for the agreed period, with the cutover decision documented.
When and how to re-run the analysis as the business changes, so the score does not decay quietly.
This is the right engagement if
- You have at least a year of renewal and churn outcomes
- Product usage data exists somewhere, even if it is not yet in Planhat
- CSMs have stopped trusting the current score, or never did
- Leadership wants a number that survives a CFO's questions
Look elsewhere if
- Fewer than thirty historical outcomes to learn from
- No access to usage or support data at all
- You want the vendor's default score turned on and left alone
Specifics.
Do we need a data science team?
No. The analysis is rigorous but it is correlation and lift work on a few hundred rows, done by someone with a mathematics degree and a data engineering background. The output is a model your ops team can maintain.
Will CSMs accept a score that disagrees with their gut?
They accept it when they can see the driving signals per account and when the parallel run shows it catching things they missed. Explainability is a design requirement, not a nice to have.
How often should it be recalibrated?
Whenever the business changes shape: new segment, new pricing, new product surface. Otherwise annually. The recalibration playbook is part of the handover.
Can this run on a platform other than Planhat?
The model and the analysis are platform-independent. The build phase assumes Planhat's calculated metrics and views; other platforms are scoped case by case.
Often paired with
Automation and data architecture
Audit, catalogue and consolidate the automation estate; build rollups that reconcile.
Read more Revenue visibilityTeam cockpits and NRR reporting
Gap to target and week-over-week movement at every level, plus prioritisation.
Read more ImplementationPlanhat implementation
New instance or rescue, sequenced so dashboards come last and stay right.
Read moreBring twelve months of outcomes. We will tell you whether a model is possible.
Thirty minutes on your data and your current score. You will leave knowing whether the history supports a predictive model and what it would take to build one.