Observability (Customer Health Scoring)

Part of The Playbook. Before you can run a play, you need to know something's wrong. Observability is the monitoring layer of Customer Success: a customer health score that tells you the state of an account before the customer has to tell you themselves.

What actually goes into a health score

A health score that only measures sentiment is a lagging indicator dressed up as a dashboard. The useful ones combine:

• Product usage and adoption depth — not just "did they log in," but which features, how deep, and whether usage is trending up or down. This is the highest-weight signal, because usage drops before satisfaction scores do.
• Support signal — ticket volume and severity trend, not a raw count. A spike after a quiet stretch matters more than a steady baseline.
• Relationship and engagement — meeting attendance, response latency, whether the champion is still the champion or quietly changed jobs.
• Commercial signal — contract terms, upcoming renewal date, prior expansion or contraction history.
• CSM/TAM pulse — the qualitative gut-check from the human who actually talks to the account. Don't let the model override this; let it flag disagreement instead.

That's five, and five is close to the ceiling, not the floor. Gainsight's own published guidance lands on 4–6 signals as the practical range — usage, support trend, sentiment, executive engagement — and every dimension past that adds noise, not accuracy. Drop any signal the team has no ability to influence, however clean its data source is; a health score is a tool for action, not a trivia dashboard.

A real example of why blending beats any single signal: support-ticket volume alone routinely misdirects attention. Teams that scored health on raw ticket count found their loudest, highest-ticket accounts were often their most engaged customers — while quieter, genuinely at-risk accounts generated no signal at all and slipped through unnoticed. Blending usage alongside support volume catches what raw ticket counts hide.

Leading vs. lagging

Leading indicators (usage decline, feature abandonment, unanswered emails) predict a problem before it's visible. Lagging indicators (a formal complaint, a support escalation, a churn notice) confirm a problem that's already arrived. A health score built entirely on lagging indicators is just a churn report with extra steps — weight leading indicators higher, and treat lagging indicators as confirmation, not discovery.

Score bands and what they trigger

A workable default, tuned per business:

Band	Meaning	Action
0–40	Critical	Declare an incident
41–60	At-risk	Run the matching play, proactively
61–80	Healthy	Standard cadence, no special action
81–100	Thriving	Flag for expansion, hand to the growth motion

What breaks health scores

• Overcomplication. A score with 40 weighted inputs is a score nobody trusts and nobody can explain to a customer's exec sponsor. Fewer inputs, clearly defensible, beats a black box every time.
• Static snapshots. A single number hides the story. Track the trend — an account moving from 90 to 65 over three weeks is a different problem than one sitting at 65 for a year.
• Mistaking activity for engagement. Logins aren't adoption. A user opening the product every day to check one report they could get from an email digest is not a healthy account, even though the login graph looks great.

Where this feeds

The score sets the trigger for The Play Library and defines the target in SLOs for Customer Outcomes. When a score crosses into critical, it's the input to Incident Response for At-Risk Accounts.

Sources

• Gainsight — "Customer Health Score Explained: Metrics, Models & Tools" — 4–6 signal guidance
• ChurnZero — "Customer Health Scores in the Age of AI"