Customer Health Score: How to Build a CHS Model for SaaS (Complete 2026 Guide)

TL;DR: A Customer Health Score (CHS) is a single 0–100 number that predicts whether a customer will renew, expand, or churn. Strong CHS models blend behavioral signals (product usage, login frequency, feature adoption), commercial signals (NPS, support tickets, invoice status), and qualitative signals (interview sentiment, stated pain points). Teams that combine quantitative data with conversational AI interviews — like those run through Koji — cut early-warning lead time by 30 to 60 days compared to usage-only scores.

What is a Customer Health Score?

A Customer Health Score is a composite metric that summarizes the probability a customer will continue to derive value from your product. It is the most widely-adopted leading indicator in modern Customer Success teams, with adoption among 71% of B2B SaaS companies above $10M ARR.

Unlike NPS (which measures sentiment after the fact) or churn rate (which measures the corpse), the CHS is forward-looking. A well-designed score lights up 30 to 90 days before the renewal conversation, giving CSMs time to intervene.

Most CHS models output one of three formats:

• Numeric score (0–100): Granular, used for dashboards and ML inputs
• Tier (Green / Yellow / Red): Simple, used for triage and CSM ticketing
• Probability score: Output of a churn-prediction model (e.g., 23% likelihood to churn in next 90 days)

The right format depends on your CS team's maturity. Early-stage teams should start with tiers; mature teams can layer in probability models.

The 4 Pillars of a Strong CHS Model

Most world-class CHS frameworks combine four signal categories:

1. Product Usage Signals (Behavioral)

What the customer actually does in your product. Examples:

• Daily/weekly active users (DAU/WAU) as a % of seats
• Time to first value (TTFV) for new accounts
• Core feature adoption (% of paid features in active use)
• Login frequency of admin/power users
• Trend direction (usage rising, flat, falling over last 30 days)

These are the strongest leading indicators. A customer who hasn't logged in for 14 days is at risk regardless of NPS.

2. Engagement Signals (Relational)

How the customer engages with you, the vendor:

• Number of trained users (relative to seats)
• QBR attendance and stakeholder engagement
• Email open rates for product updates
• Community/event participation
• Support ticket volume and severity

3. Commercial Signals (Account)

Objective account-level data:

• NPS, CSAT, CES scores (and trends)
• Days until renewal
• Invoice status (paid on time, past due)
• Expansion indicators (added seats, requested upgrade quotes)
• Champion stability (key contact still in role)

4. Qualitative Signals (Voice of Customer)

The pillar most teams underweight or skip entirely. Yet it is the strongest predictor of intent — because customers tell you what spreadsheets cannot.

• Sentiment from interview transcripts (positive, neutral, negative)
• Stated frustrations raised during QBRs or check-ins
• Mentions of competitive evaluation in calls
• Internal change signals (champion leaving, budget freeze, M&A activity)
• Specific feature requests (high engagement) vs. silence (disengaged)

This is where platforms like Koji become powerful. Quarterly AI-moderated check-in interviews — running automatically via the Koji API or scheduled webhook — generate structured sentiment and theme data that feed directly into your CHS as a weighted signal. Most CHS models that incorporate this qualitative layer outperform usage-only models by 20 to 40% in churn prediction accuracy.

How to Calculate a Customer Health Score (Formula)

The most common formula is a weighted-sum model:

CHS = (Usage Score × W1) + (Engagement Score × W2) + (Commercial Score × W3) + (Qualitative Score × W4)

A typical weighting for a B2B SaaS product:

Each pillar is normalized to a 0–100 sub-score using your own benchmarks. For example:

• Usage sub-score: 100 if DAU > 60% of seats, 50 if 30–60%, 0 if < 30%
• NPS sub-score: Promoter = 100, Passive = 50, Detractor = 0
• Qualitative sub-score: Auto-tagged from AI-interview sentiment (positive = 100, mixed = 50, negative = 0)

The weights are not set in stone. Test them against churned accounts: look at your last 50 churns and ask which signals would have caught them earliest. Adjust weights until the model would have flagged at least 80% of those accounts 60+ days before churn.

Example: A Working CHS Model for B2B SaaS

Let's walk through "Acme CRM," a hypothetical 200-seat customer:

Usage signals:

• 130 of 200 seats logged in this week (65% DAU)
• Core feature adoption: 4 of 6 features active
• Usage trend: +5% MoM
• → Usage sub-score: 80

Engagement signals:

• 140 of 200 users trained
• QBR attended last quarter
• 2 open support tickets, both P3
• → Engagement sub-score: 75

Commercial signals:

• NPS score: 8 (Promoter)
• 7 months until renewal
• All invoices paid on time
• → Commercial sub-score: 90

Qualitative signals:

• Q3 AI-moderated check-in interview: sentiment positive, champion mentioned "expanding to UK team"
• No competitive mentions, no escalations
• → Qualitative sub-score: 90

Total CHS:

(80 × 0.35) + (75 × 0.20) + (90 × 0.25) + (90 × 0.20)
= 28 + 15 + 22.5 + 18
= 83.5 → Green / Expansion candidate

Now imagine the qualitative signal had flagged "competitor under evaluation" instead. The score would drop to ~65 — still Yellow, but with a clear narrative of why. CSMs can act on narrative; they cannot act on a usage dip alone.

Common CHS Model Mistakes

1. Overweighting usage. A customer with high logins but a frustrated champion is still going to churn.
2. Skipping qualitative inputs. Without conversational signal, the model is blind to intent until behavior changes — which is too late.
3. Setting and forgetting weights. Weights should be re-validated quarterly against actual churn/retention outcomes.
4. One-size-fits-all model. Enterprise and SMB customers churn for different reasons. Segment your CHS by tier.
5. Score without action playbook. A Red score means nothing without a defined intervention. Map every tier to specific CSM workflows.
6. Hiding the score from customers. Industry best practice (e.g., Gainsight, Catalyst) recommends sharing the score during QBRs — it forces transparency and aligns expectations.

How to Capture Qualitative Signals at Scale

Manually interviewing every customer every quarter is impossible. Most CS teams cover 10 to 20% of their book and rely on guesswork for the rest. AI-moderated interviews fix this.

Using a platform like Koji, you can:

• Auto-trigger a quarterly check-in interview via webhook for every account in the Yellow tier or higher
• Use structured questions (the six Koji types: open_ended, scale, single_choice, multiple_choice, ranking, yes_no) to capture both numeric and conversational signal
• Get auto-extracted sentiment that maps directly into your CHS qualitative sub-score
• Surface competitive mentions and churn risk language automatically via AI analysis
• Pipe results to your CRM or CS platform (HubSpot, Salesforce, Gainsight) so CSMs see qualitative context alongside usage

This turns the qualitative pillar from "what the loudest customer said in last QBR" to a structured, comparable dataset across your entire book.

CHS Tier Definitions & Playbooks

The most underused tier is Yellow. Most teams over-invest in Reds (already too late) and Greens (already happy). The Yellow zone is where CHS earns its ROI — small, well-timed interventions reverse the trajectory.

CHS vs. NPS vs. CSAT vs. CES

CHS is the operating metric — it tells your CSM team where to spend the next hour. The others are inputs into the CHS or independent diagnostic tools.

Rolling Out a Customer Health Score in 30 Days

Week 1: Foundations

• Pull last 50 churns and 50 expansions from your CRM
• Identify the 5 to 10 signals most correlated with each outcome
• Decide on 3 to 5 starter signals (don't aim for perfection)

Week 2: Calculate v1

• Pull data into a spreadsheet for every account
• Apply weights and generate v1 scores
• Validate against last quarter's actuals

Week 3: Integrate Qualitative

• Set up an AI-moderated quarterly check-in study (with Koji or similar)
• Push results into the CHS as a 20–25% weighted signal
• Tag competitive mentions and churn-risk language

Week 4: Operationalize

• Define tier playbooks (Green/Yellow/Orange/Red)
• Push CHS into your CRM or CS platform
• Train CSMs and set weekly review cadence

The model gets better over time. Plan to re-tune weights quarterly based on accuracy against actual outcomes.

Frequently Asked Questions

How is a Customer Health Score different from NPS?

NPS measures loyalty at one moment with one question. A Customer Health Score is a composite, continuous metric that combines NPS with usage, engagement, commercial signals, and qualitative interview data to predict renewal probability. NPS is one input into the CHS, not a replacement.

What's the right weighting for CHS signals?

A typical B2B SaaS weighting is 35% usage, 20% engagement, 25% commercial, 20% qualitative — but the right weighting depends on your churn drivers. Audit your last 50 churns to find what signals would have caught them earliest, then weight accordingly.

Can I build a Customer Health Score without a CS platform like Gainsight or Catalyst?

Yes. Many companies start with a Google Sheet or Looker dashboard that pulls from their CRM, product analytics, and survey/interview tool. The model matters more than the platform. Graduate to a dedicated CS tool once you have 200+ accounts to manage.

How often should I update the Customer Health Score?

Daily refresh of usage data; weekly refresh of engagement; quarterly refresh of qualitative signal via AI-moderated interviews. Avoid manual monthly updates — they bottleneck on the CS team and become stale.

How do AI interviews improve Customer Health Score accuracy?

AI-moderated interviews scale qualitative input from "the 10% of accounts your CSMs got to this quarter" to 100% of accounts. They surface intent signals (competitive mentions, frustration, champion exits) that usage data alone misses, improving churn prediction accuracy by 20 to 40%.

Should I share the Customer Health Score with my customers?

Yes — gradually. Sharing CHS during QBRs forces transparent value conversations and aligns expectations. Mature CS teams (Gainsight, HubSpot, Notion) all share scores externally. Start by sharing only the tier (Green/Yellow/Red) rather than the raw number.

Related Resources

• Win-Back Customer Interviews — Reactivate lapsed customers flagged by your CHS
• Structured Questions Guide — Design check-in interview questions using Koji's six question types
• Churned Customer Interviews — Calibrate your CHS against actual churn outcomes
• Customer Success Interview Guide — Conversation patterns for quarterly check-ins
• Stay Interviews at Scale — Run proactive retention interviews via AI
• Hubspot Research Integration — Pipe interview signals into your CRM for CHS calculation
• NPS Survey Guide — A core CHS input signal

Ready to power your CHS with conversational qualitative signal? Try Koji free — auto-moderated quarterly check-ins, sentiment scoring, and CRM integration included.