Most customer health scores are vibes wrapped in a spreadsheet. Green, yellow, red. Nobody can tell you why an account is yellow. Nobody can tell you whether yellow accounts actually churn more than green ones.
This post covers what a real customer health score looks like, how to build one, how to validate it against actual churn, and how to make it drive action instead of decoration.
What a customer health score actually is
A customer health score is a single number that summarizes how likely an account is to renew, expand, or churn. The score rolls up multiple inputs into one indicator.
The point is not the number. The point is what the number triggers. A score that does not change anyone's behavior is a dashboard ornament.
Health scores work when three things are true. The inputs predict outcomes. The score updates often enough to matter. Someone owns the response when the score moves.
If any of those three break, the score is theater.
The four input categories
Every credible health score pulls from four buckets. Skip a bucket and you lose signal.
Product usage
This is the most predictive bucket for most SaaS products. It includes login frequency, active users as a percentage of licenses, feature adoption depth, and time spent in core workflows.
The metric that matters depends on your product. For a sales tool, it might be weekly active users on the rep team. For a finance product, it might be month end usage by the controller.
Pick the actions that correlate with value realization, not vanity logins.
Support tickets
Volume and severity of tickets both signal trouble. A surge in P1 tickets is an obvious flag. A flat line of zero tickets can also be a flag. It often means the customer stopped using the product.
Track ticket sentiment when you can. A frustrated tone in a ticket is data.
NPS, CSAT, and qualitative signals
Survey data lags behavior, but it captures what behavior alone misses. A user can log in every day and still hate the product.
Use NPS for relationship health. Use CSAT for transactional health after support interactions. Use qualitative notes from CSMs and AEs for the things surveys miss.
Business outcomes
This is the bucket most teams ignore. Did the customer hit the goal they bought your product to hit?
If they bought your product to reduce ramp time and ramp time has not changed, the account is at risk no matter how green every other metric looks. Outcomes lag, but they are the truest signal of renewal intent.
For more on outcome tracking, see our customer success plan guide.
The math: weighted average vs ML model
You have two main options for combining inputs into a score.
Option 1: Weighted average
This is the right starting point for almost every team. Pick five to ten inputs. Assign each a weight that sums to 100. Each input gets a normalized score from zero to 100. Multiply, add, done.
Example weights for a SaaS company:
- Weekly active users vs licenses: 25 percent
- Feature adoption depth: 20 percent
- NPS: 15 percent
- Support ticket volume and severity: 15 percent
- Executive sponsor engagement: 15 percent
- Outcome tracking: 10 percent
The weights are a hypothesis. You will tune them once you validate the score against churn.
Option 2: Machine learning model
A predictive model trained on historical churn data can identify patterns a weighted average misses. Logistic regression or gradient boosted trees are the common choices.
The catch is that ML models need volume. If you have fewer than 200 churn events in your history, the model will overfit. You will get a score that looks precise and predicts nothing.
Most teams under 500 customers should stay with a weighted average. Add ML when you have the data and a CS ops team to maintain it.
How to validate the score
Here is the test most teams skip. Pull a list of accounts that churned in the last 12 months. Look at their health score 90 days before they churned.
If churned accounts had healthy scores 90 days out, your model is broken.
Run the same test for expansion. Pull accounts that expanded. What did their score look like 90 days before the expansion event?
A good health score has three properties. Churned accounts trended down before they churned. Expansion accounts trended up before they expanded. The distance between green and red is statistically meaningful.
Validate quarterly. Customer behavior shifts. The inputs that predicted churn last year may not predict churn this year.
How to operationalize the score
A validated score is useless without a play behind it.
Tier 1: Score crosses a threshold
When an account moves from green to yellow, something fires. A task. An email. A Slack alert to the CSM and the AE.
The play depends on the trigger. If usage dropped, the play is a check in call. If NPS fell, the play is an executive outreach. If the score dropped because the executive sponsor left, the play is a champion replacement campaign.
Map every input to a play. The score is the trigger. The play is the response.
Tier 2: Score crosses a tier line
Yellow to red is an escalation. The CSM should not be alone. The manager joins. The AE joins. Sometimes the VP joins.
A red account gets a save plan with a deadline. Two weeks. Four weeks. Whatever the renewal timeline allows.
Tier 3: Score informs forecasting
Health scores feed renewal forecasts. A green book of business at 95 percent forecast renewal. A yellow book at 70 percent. A red book at 30 percent.
This is also how you connect health to net revenue retention. NRR is the outcome. The health score is the leading indicator.
Common health score failures
Five patterns kill health scores. Watch for all of them.
Failure 1: Too many inputs
A score with 25 inputs cannot be debugged. When the score moves, nobody knows why. CSMs ignore it because they cannot act on it.
Five to ten inputs is the sweet spot. Each one should map to a clear play.
Failure 2: Static weights
Weights set at launch and never revisited. The product changes. The customer base changes. The weights need to change too.
Tune weights at least annually. More often if you ship major features.
Failure 3: No outcome data
Most health scores rely entirely on usage and support data because that data is easy to get. Outcome data is hard to get. So teams skip it.
Skipping outcomes means the score predicts engagement, not renewal. Customers can be engaged and still churn because the product did not deliver business value.
Get outcome data even if it is qualitative. Ask the CSM to rate outcome attainment quarterly.
Failure 4: Score without ownership
A health score that lives in a CS dashboard and never reaches the AE is half a score. Expansion lives on the AE. Churn risk lives on both.
Both functions need the score. Both functions need to act on it.
Failure 5: Treating the score as the answer
The score is a hypothesis about an account. The CSM still needs to verify. A green score on an account where the executive sponsor just left is not green. It is red and the model has not caught up.
Use the score as a starting point. Override when the qualitative signal is louder.
Building the dashboard
A useful customer health dashboard has three views.
The portfolio view shows score distribution across the book. Are 60 percent of accounts green? 30? This tells leadership whether the book is healthy in aggregate.
The trend view shows how scores have moved over time. A book trending down is the canary. A book trending up is expansion fuel.
The account view drills into one account. It shows the inputs, the weights, the score history, the plays that fired, and the actions taken. This is what the CSM uses in the weekly account review.
The dashboard should live where the seller already works. If your CSMs and AEs work in Salesforce, the dashboard should be in Salesforce. Pulling people out of their workflow to check a score in another tool is how scores get ignored.
Health score and account growth
Health scores predict churn. They also surface expansion opportunity.
A green account with high product engagement and strong outcomes is an expansion candidate. The health score tells you which accounts are ready. The account growth strategy tells you what to expand into.
The two work together. Health flags the timing. The growth plan defines the motion.
Related reading
Bring this into Salesforce with CRUSH
A health score is an input to an account plan, not a substitute for one. CRUSH brings health scores into the same Salesforce surface where account plans, relationship maps, and whitespace already live. When a score moves, the account plan updates. When the plan needs a new play, the score tells the team where to start.
That is the point of building a health score in the first place. It changes what the account team does next.
See how CRUSH connects health, planning, and growth in Salesforce.
