What signals should go into a customer health score?

Most models combine four categories: product usage (depth, breadth, frequency), support signals (ticket volume, severity, sentiment), relationship signals (sponsor engagement, stakeholder turnover, QBR attendance), and commercial signals (payment status, contract terms). The right weighting is product- and segment-specific and should be set by back-testing against actual renewal and expansion outcomes, not guessed.

How is health scoring different from churn prediction models?

A health score is usually a rules-based weighted composite that a CS team can build and understand without data science resources. Churn prediction implies a statistical or machine-learned model trained and validated against historical churn outcomes. Many teams start with a rules-based health score and evaluate a predictive model later, once volume and clean outcome history justify the investment.

How often should the health score update?

Usage and support signals typically recompute daily or in near-real-time since a CS platform usually pipes in product and ticketing data continuously. Relationship and commercial signals, which are often manually entered, update on whatever cadence the CSM logs them — weekly is a common minimum so the score doesn't go stale between QBRs.

Should the score be visible to the customer?

Almost never directly — the score is an internal risk and prioritization tool, not a customer-facing metric, and sharing a raw 'health score' with a customer tends to create adversarial rather than collaborative conversations. What is shared externally is usually reframed as adoption or success metrics tied to the customer's own goals.

What should happen when an account turns red?

A defined playbook should trigger automatically: typically a CSM-led risk assessment within days, an internal escalation to CS leadership for visibility, and depending on severity, a joint save-plan conversation with the account's executive sponsor. The score's value comes entirely from the action attached to the band change.

Customer Health Scoring

A composite score that combines product usage, engagement, and relationship signals into a single indicator of an account's renewal and expansion risk.

Last updated July 11, 2026Also known as: account health score, customer health index, renewal risk score↓ Download SOP (Markdown)

Key facts

In one sentence: A composite score that combines product usage, engagement, and relationship signals into a single indicator of an account's renewal and expansion risk.
Primary owner: CS Operations / RevOps
Workflow: 8 steps, from “Define what 'healthy' means for this product and segment” to “Recalibrate signals and weights periodically”
North-star metric: Health-band predictive accuracy — typical target: typical target: meaningfully separated, e.g. 1.5–2x or more

What is customer health scoring?

Customer health scoring is the process of combining multiple signals — product usage depth and breadth, support ticket volume and sentiment, engagement with the CS team, contract and billing status, and relationship strength with key stakeholders — into a single score or banded rating (e.g., green/yellow/red) per account. The score gives customer success, sales, and leadership a shared, at-a-glance answer to 'is this account in good shape,' surfaced early enough to act on rather than discovered at the renewal conversation.

Health scoring is not the same as an NPS or CSAT survey score, which captures a customer's stated sentiment at one point in time from whoever responds. Health scoring is a continuously computed composite built primarily from behavioral and operational data, of which a survey response is at most one input among many. It is also not the same as churn prediction in the strict statistical sense — a health score can be a simple rules-based weighted composite, while churn prediction implies a model trained and validated against actual churn outcomes. Many teams start with the former and only invest in the latter once volume and data history justify it.

The process exists because renewal and expansion risk is rarely visible from the CRM's opportunity stage alone — an account can look fine on paper (on-time payment, no open tickets) while quietly disengaging, or look risky on a surface metric while actually being stable. A well-built health score aggregates the signals that individually are noisy into a composite that is directionally reliable, and — critically — is calibrated against what actually predicted churn and expansion in the past, not just what data happens to be easy to pull.

When to implement

Worth building once a CS team manages enough accounts that a CSM cannot personally track the state of every one from memory — typically 30+ accounts per CSM, or any pooled/tech-touch segment with no dedicated owner. Prerequisites: reliable product usage data (event tracking or a proxy for it), a CRM or CS platform to host the score, and at least a few renewal cycles of outcome history to calibrate the model against — without outcome history, the score is a plausible-looking guess.

Step-by-step workflow

1
Define what 'healthy' means for this product and segment
Owner: CS leadership + Product
Before choosing inputs, agree what a healthy account actually looks like in outcome terms — renews, expands, references the product — and work backward to the leading indicators that correlate with those outcomes for this specific product and customer segment. A B2B collaboration tool and a compliance system have different usage patterns that signal health.
- Pull a sample of renewed/expanded vs. churned/downgraded accounts
- Identify usage and engagement patterns that differ between the two groups
- Document the health definition per segment if usage patterns differ materially
2
Select and weight the input signals
Owner: CS Operations / RevOps
Choose a small number of signal categories — product usage (depth, breadth, frequency, key-feature adoption), support (ticket volume, severity, resolution sentiment), relationship (executive sponsor engagement, stakeholder turnover), and commercial (payment status, contract terms) — and weight them by how strongly each actually predicted past outcomes, not by ease of access.
- Score usage signals: login frequency, active users, core-workflow completion, feature depth
- Score support signals: ticket volume trend, escalations, CSAT on tickets
- Score relationship signals: sponsor engagement, stakeholder/champion turnover, QBR attendance
- Weight each category against back-tested correlation with churn/expansion
3
Build the composite score and bands
Owner: CS Operations / RevOps
Combine weighted signals into a single score (commonly 0–100) and define bands (e.g., green/yellow/red) with clear, written criteria for each band, so 'yellow' means something specific and consistent across every CSM rather than a gut feeling dressed up as a number.
4
Surface the score with reasons, not just a color
Owner: CS Operations
Display the score on the account record alongside the specific signals driving it — which usage metrics dropped, which tickets escalated, which stakeholder went quiet — so the CSM can act on the cause, not just react to a color change.
5
Wire the score into playbooks and escalation
Owner: CS leadership + CSM team
Define what happens when an account crosses into yellow or red: a specific playbook (re-engagement outreach, executive check-in, risk escalation to CS leadership), an owner, and an SLA. A score with no attached action is a dashboard decoration.
6
Validate the model against renewal and expansion outcomes
Owner: CS Operations + RevOps
Each quarter, compare renewal and expansion rates by health band from the prior period. If red accounts renew at similar rates to green accounts, the model isn't predictive and the weights need revisiting rather than being trusted at face value.
7
Feed the score into forecasting and account planning
Owner: CS leadership + Sales leadership
Use health bands as an input to renewal forecasting, expansion pipeline qualification, and account planning discussions with sales — a red account should change the renewal forecast and trigger a joint save plan, not sit unnoticed in a CS-only dashboard.
8
Recalibrate signals and weights periodically
Owner: CS Operations / RevOps
Revisit input weights, thresholds, and even the signal set itself as the product, ICP, or customer base changes — a health model built for last year's product usage patterns silently drifts out of relevance without a deliberate refresh.

Roles & responsibilities

Role	Responsibility
CS Operations / RevOps	Owns the scoring model design, signal weighting, and periodic validation against outcomes.
Customer Success Manager	Acts on the score day-to-day; feeds back qualitative context the model can't see.
CS leadership	Defines what 'healthy' means for the business, owns playbooks tied to each health band, and holds the model to renewal/expansion outcomes.
Product / Product Analytics	Provides usage event data and helps identify which product behaviors correlate with retention.
Sales leadership	Uses health bands in renewal forecasting and expansion qualification; partners on save plans for at-risk accounts.

Tool stack

Customer success platform
Gainsight · ChurnZero · Vitally · Totango — typically where the composite score is built and surfaced
Product analytics
Amplitude · Mixpanel · Pendo — source of the usage-depth and adoption signals the score depends on
CRM
Salesforce · HubSpot CRM — surfaces the score to sales and links it to renewal/expansion opportunities
Support / ticketing
Zendesk · Intercom · Freshdesk — source of ticket volume, severity, and sentiment signals
Survey / voice of customer
Delighted · Qualtrics · in-app NPS via Pendo — supplements behavioral data with stated sentiment

Key metrics

Metric	Definition	Formula	Typical target
Health-band predictive accuracy	How much better green accounts renew/expand versus red accounts — the core validity check of the model.	Green-band renewal rate ÷ red-band renewal rate	typical target: meaningfully separated, e.g. 1.5–2x or more
Red-account renewal rate	Renewal rate of accounts flagged red, showing whether the model catches real risk in time to act.	Renewed red accounts ÷ total red accounts	typical target: below green/yellow renewal rate by a wide, actionable margin
Score coverage	Share of the active customer base with a computed, up-to-date health score.	Accounts with a current score ÷ total active accounts	typical target: above 95%
Time in red before action	Days between an account entering the red band and a documented save-plan action.	First save-plan action date − date entered red	typical target: under 5 business days
False positive rate	Share of red-flagged accounts that renew/expand normally despite the flag, indicating over-sensitive scoring.	Red accounts that renewed/expanded ÷ total red accounts	typical target: below 30%

Common failure points

Failure	Symptom	Fix
Score built on convenient data, not predictive data	The composite doesn't separate renewing from churning accounts; CSMs stop trusting it.	Back-test every signal against actual renewal/expansion outcomes before including it, and drop ones with no lift.
One blended score with no visible drivers	CSMs see 'yellow' but not why, and can't act on the score without redoing the diagnosis manually.	Surface the specific signals driving the score alongside the composite number.
No action attached to band changes	Accounts sit in red for months with nobody assigned to respond.	Define a specific playbook, owner, and SLA for every band transition, especially into yellow and red.
Model never revalidated	Health bands stop correlating with outcomes as the product and customer base evolve, but nobody notices.	Quarterly validation comparing renewal/expansion rates by band; recalibrate weights when the correlation weakens.
Relationship signals ignored in favor of usage data alone	Accounts with strong usage churn anyway after a champion leaves or a sponsor disengages.	Include stakeholder turnover and sponsor engagement as explicit signal categories, not an afterthought.
Score hidden from sales and leadership	A red account renews as a surprise-loss because only CS ever saw the risk.	Feed health bands into renewal forecasting and account planning visible to sales and CS leadership jointly.

Frequently asked questions

What signals should go into a customer health score?: Most models combine four categories: product usage (depth, breadth, frequency), support signals (ticket volume, severity, sentiment), relationship signals (sponsor engagement, stakeholder turnover, QBR attendance), and commercial signals (payment status, contract terms). The right weighting is product- and segment-specific and should be set by back-testing against actual renewal and expansion outcomes, not guessed.
How is health scoring different from churn prediction models?: A health score is usually a rules-based weighted composite that a CS team can build and understand without data science resources. Churn prediction implies a statistical or machine-learned model trained and validated against historical churn outcomes. Many teams start with a rules-based health score and evaluate a predictive model later, once volume and clean outcome history justify the investment.
How often should the health score update?: Usage and support signals typically recompute daily or in near-real-time since a CS platform usually pipes in product and ticketing data continuously. Relationship and commercial signals, which are often manually entered, update on whatever cadence the CSM logs them — weekly is a common minimum so the score doesn't go stale between QBRs.
Should the score be visible to the customer?: Almost never directly — the score is an internal risk and prioritization tool, not a customer-facing metric, and sharing a raw 'health score' with a customer tends to create adversarial rather than collaborative conversations. What is shared externally is usually reframed as adoption or success metrics tied to the customer's own goals.
What should happen when an account turns red?: A defined playbook should trigger automatically: typically a CSM-led risk assessment within days, an internal escalation to CS leadership for visibility, and depending on severity, a joint save-plan conversation with the account's executive sponsor. The score's value comes entirely from the action attached to the band change.

Download the SOP

The standard operating procedure for this process — purpose, roles, step-by-step procedure with checklists, metrics, and failure modes — is available as a Markdown file you can drop into Notion, Confluence, or any wiki and adapt.

↓ Customer Health Scoring SOP (.md)

Related processes

Cite this page

“Customer Health Scoring: definition, workflow, roles, metrics & SOP.” b2bprocess.com, updated 2026-07-11. https://b2bprocess.com/customer-health-scoring

What is customer health scoring?

Step-by-step workflow

Define what 'healthy' means for this product and segment

Select and weight the input signals

Build the composite score and bands

Surface the score with reasons, not just a color

Wire the score into playbooks and escalation

Validate the model against renewal and expansion outcomes

Feed the score into forecasting and account planning

Recalibrate signals and weights periodically