Customer Health Score Model for B2B SaaS

What a health score is supposed to do

A health score has one job: order your book so the right account gets attention first. It is not a performance metric, not a forecast and not a substitute for a conversation.

That job sets the design rules. The score must be explainable in a sentence, the components must be visible so a low score tells you what to do, and the model must be checked against reality often enough that people trust it.

The five components and starting weights

Start with five components. Fewer misses real risk; more creates a model nobody can explain.

Scoring: a worked example

Score each component from 0 to 100, then apply the weights. An account with usage at 60, breadth at 45, outcome progress at 80, support experience at 70 and sponsor engagement at 90 scores: (60 × 0.30) + (45 × 0.20) + (80 × 0.20) + (70 × 0.15) + (90 × 0.15) = 18 + 9 + 16 + 10.5 + 13.5 = 67.

A 67 alone means little. What matters is that breadth at 45 is the weakest component: the account is being carried by a small group of users while the sponsor stays engaged. The action is an adoption push with the teams that are not using it, not a relationship call.

This is why component visibility is not a nice-to-have. The number sets priority; the component sets the play.

Bands and the action each one triggers

Attach a defined action to each band, otherwise the score becomes decoration.

• 80–100 Healthy: confirm value in the next review, look for expansion evidence.
• 60–79 Watch: name the weakest component, agree one corrective action with the customer.
• 40–59 At risk: written recovery plan, internal owner, executive visibility.
• Below 40 Critical: escalate, involve leadership, decide whether recovery is realistic.

Validate the model every quarter

Take every account that churned or downgraded in the last quarter and look at what the score said ninety days earlier. Anything green at that point means the model is missing a signal.

In practice, the missing signal is usually one of three things: a sponsor change nobody recorded, an outcome that was never defined so progress could not be scored, or usage that looked fine in total while the important workflow had quietly stopped.

Adjust one thing at a time and write down what you changed. A model with a documented history earns trust; one that gets silently re-weighted every quarter does not.

Common mistakes

Three failure patterns show up repeatedly. Scoring only what is easy to measure, which usually means login counts standing in for value. Letting the CSM override the score without recording why, which erases the learning. And reporting the average score across the book, which hides the distribution that actually matters.

Report the distribution and the movement between bands. That view drives decisions; an average never has.

Frequently asked questions

What components should a customer health score include?

Product usage against the expected pattern, adoption breadth, progress on the customer's outcomes, support experience including repeat contacts, and sponsor engagement. Five components with published weights is enough for most B2B SaaS teams.

How often should a health score be updated?

Refresh the data weekly and review the model itself quarterly against accounts that actually churned. Weekly refresh keeps the priority list current; quarterly validation keeps the model trustworthy.

Can a health score predict churn?

It can flag elevated risk earlier than intuition does, but it does not predict outcomes on its own. Treat a low score as a prompt for a conversation, and validate the model backward each quarter so you know how often it caught real losses.