Customer health score: AI signals, examples, and template

A customer health score should answer a simple question: is this account on track to renew, grow, and achieve its goals?

Most health scores do not quite answer that. They combine product usage, support tickets, NPS, and CRM fields into a number, but the number often hides the actual account story.

A better health score is explainable. It tells the team what changed, where the evidence came from, and what action should happen next.

What to include in a customer health score

Start with four categories.

1. Product and adoption signals

These show whether the customer is using the product in a way that supports value.

Examples:

• active users
• feature adoption
• usage depth
• admin activity
• implementation progress
• change in usage over time

Usage is useful, but it can be misleading alone. A low-usage account in onboarding is different from a low-usage account 30 days before renewal.

2. Relationship signals

These show whether the right people are engaged.

Examples:

• champion responsiveness
• executive sponsor attendance
• meeting cadence
• stakeholder changes
• buyer engagement
• decision-maker sentiment

Relationship signals are often the difference between a recoverable account and a surprise churn.

3. Outcome signals

These show whether the customer is achieving the business result they bought for.

Examples:

• stated business goals
• success criteria
• project milestones
• unresolved blockers
• ROI proof
• value confirmation from the customer

If outcomes are not documented, the renewal conversation becomes subjective.

4. Risk and friction signals

These show whether something is actively working against renewal or expansion.

Examples:

• support escalations
• unresolved product gaps
• pricing objections
• procurement blockers
• implementation delays
• negative sentiment
• competitor mentions

These are the signals AI can extract especially well from unstructured customer conversations.

A simple health score template

Use this model as a starting point:

Category	Weight	Example signals
Adoption	30%	usage trend, feature adoption, active users
Relationship	25%	champion engagement, meeting cadence, executive sponsor
Outcomes	25%	business goals, milestone progress, value proof
Risk	20%	support escalations, blockers, sentiment, procurement

Then require every score change to include:

• reason for the change
• source citation
• date of signal
• recommended next action

That last requirement prevents the score from becoming a black box.

Example: explainable health change

Weak health update:

"Account moved from yellow to red."

Strong health update:

"Account moved from yellow to red because the executive sponsor missed the last two meetings, the champion has not replied in 12 days, and the customer raised unresolved onboarding blockers in the latest email thread. Recommended action: send an executive recap and propose a recovery plan before the renewal checkpoint."

The second version is something a CSM can act on.

How AI improves health scoring

AI helps by reading the sources humans do not have time to inspect every day:

• meeting notes
• email threads
• CRM changes
• support conversations
• call transcripts
• calendar patterns

But AI should not simply produce a score. It should produce an explanation.

How Aartha approaches health scores

Aartha connects customer signals to a cited Customer Memory Graph. Health changes are grounded in real account evidence, so teams can understand why the score changed and what to do next.

That turns health scoring from a reporting exercise into an operating system for retention.