Customer Health Score: Build an Early Warning System for Churn

Why Existing Metrics Are Not Enough

You can tell when a customer is slipping away. Logins have dropped. Support tickets have a sharper tone. The power user has stopped replying to your emails. But the first official signal you get is a cancellation in your CRM, 60 days later.

The problem is that most organisations measure customer health using lagging indicators — NPS, renewal rate, support CSAT. By the time those move, the decision has usually already been made.

Customer Health Score is the answer to that problem. It combines leading and lagging indicators into a single number that tells you whether an account is in good shape, at risk, or on its way out. The goal is 60-90 days of lead time — the window where you can still act.

What Is a Customer Health Score?

A Customer Health Score is a composite number (typically 0-100 or a colour like green/yellow/red) that aggregates several data signals about the customer's relationship with you. In isolation it means nothing. It is only useful as a trigger for action.

Three levels:

• Green (healthy): The customer is using the product, NPS is positive, support is quiet. Focus on upsell and referrals.
• Yellow (at risk): Unusual patterns — declining usage, more tickets, neutral NPS. Proactive outreach from Customer Success.
• Red (critical): The score indicates high churn probability. Escalated to account management and possibly leadership.

These are triggers, not reports. A red customer should fire a workflow the same day, not end up in a quarterly dashboard.

Leading vs. Lagging Indicators

The biggest mistake in most Health Score models is overweighting lagging indicators. You end up with a model that predicts what you already know.

A good model blends both, weighted toward leading indicators for prediction. Research from Gainsight, ChurnZero and others consistently shows that usage behaviour is the single strongest predictor of B2B SaaS churn.

The Five Dimensions of a Typical Health Score

Most models cover five dimensions. You do not need all five from day one, but you should have at least three.

1. Product usage (weight: 30-40%)

The strongest predictor. What do you track?

• Login frequency: Daily/weekly active users per account
• Feature adoption: How broad is usage across the product's core features?
• Depth: Time in product, number of actions per session
• Admin activity: Is the customer creating new users? Configuring? That is the engagement signal.

Warning signs: -30% MoM drop in DAU, admin-level inactivity for 14+ days, no new users created in a quarter.

2. Engagement (weight: 15-20%)

Interaction with you outside the product.

• Email open/click rates: If these fall, mindshare falls
• Event attendance: Webinars, training modules
• Community activity: Where relevant
• Survey response rate: Silent customers are often a bad sign

3. Sentiment (weight: 20-25%)

What is the customer saying?

• NPS and transactional CSAT: See what is NPS
• Sentiment in open text responses
• Tone in emails and support conversations
• QBR attendance and feedback: In the enterprise segment

4. Support (weight: 10-15%)

Not just volume, but patterns.

• Number of open critical tickets
• Average resolution time for the customer's tickets
• Tickets per month trend: A 50%+ increase over three months is a flag
• Escalations to management

5. Commercials and contract (weight: 10-15%)

• Payment history: Delays, failed charges
• Contract length: Shorter terms = higher flight risk
• Seat utilisation: Is the customer paying for licences that are not used?
• Historical upsell/downgrade

How to Build Your First Health Score in 6 Steps

1. Define what "healthy" means for your product

Get Customer Success in a room for two hours. Ask them to describe their 10 healthiest and 10 least healthy customers. What are the common traits? That becomes the backbone of your signals.

2. Pick 3-5 starter signals

Begin with the smallest set that makes sense. For B2B SaaS that is typically: MAU per account, feature adoption, NPS, open support tickets, paid on time.

Add complexity later once the basic model runs.

3. Define scoring per signal

Each signal is scored 0-100. Example for login frequency:

0-30:  no logins in the last 14 days
31-60: 1-3 logins in the last 14 days
61-85: 4-10 logins in the last 14 days
86-100: 11+ logins in the last 14 days

Adjust the thresholds to your product. A B2B integration used once a month should not be judged like a daily-use SaaS.

4. Choose weights — start heuristic

Begin with heuristic weights based on CS team experience:

• Usage: 35%
• NPS/Sentiment: 25%
• Support: 15%
• Engagement: 15%
• Commercials: 10%

These are starting points, not truths. Adjust after 6 months once you have data.

5. Set thresholds for green/yellow/red

Rule of thumb for B2B SaaS:

Thresholds have to be calibrated to your actual dataset. A score of 60 can be "yellow" at your company and "red" elsewhere.

6. Build the workflow that reacts

The score is meaningless if nothing happens when it turns red. Concretely:

• Green customer drops to yellow: Automatic notification to the CSM, task created in the CRM, outreach within 7 days.
• Yellow customer drops to red: Escalated to Head of CS, executive sponsor contacted, structured close-the-loop conversation within 48 hours.
• Red customer does not stabilise within 30 days: Win-back plan or deliberate offboarding.

A Real-World Example

A B2B SaaS company we worked with had 14% annual churn. They had no visibility into which customers were at risk before the cancellation hit. We built a simple Health Score with five signals:

After three months of calibration the model caught 74% of the customers who churned in the following 6 months, with 60-90 days of lead time. The CS team reached out proactively to every red customer. After 12 months, churn had fallen to 9%.

Not because of the score. Because of the workflow the score triggered.

Common Pitfalls

Too many signals from the start. 20 signals sounds precise. It becomes practically impossible to explain why a customer is red. Start with 3-5.

Only lagging indicators. If your model is primarily built on NPS and renewal rate, it will only detect churn when it is too late. Use leading indicators as your primary signals.

No segmentation. Enterprise customers have completely different usage patterns than SMB. Build separate models for your main segments, or adjust the thresholds per segment.

No feedback loop into the model. When a "green" customer suddenly cancels, that needs to be analysed. What did the model miss? Recalibrate weights and thresholds every quarter.

Score without workflow. The most commonly observed problem. A dashboard with red and green customers, but nobody owns the response. Without ownership and an SLA the score is decoration.

Relationship to Other CX Metrics

Customer Health Score does not stand alone. It is complementary:

The best setup combines all four. NPS and CSAT feed into the Health Score. CES surfaces friction that later manifests as declining usage. Health Score is where it all culminates in one prioritised action list.

Getting Started

You do not need a dedicated platform to build your first Customer Health Score. You probably already have 80% of the data you need.

1. Get the CS team together and agree on 3-5 signals
2. Build a simple model in your BI tool or spreadsheet
3. Define green/yellow/red thresholds
4. Agree on an SLA for reacting to red customers
5. Run it for a quarter, then calibrate

The best Health Score programmes are not the ones with the most data points. They are the ones that actually get Customer Success to act before it is too late.

And as always: the score is the first step. The action is the whole difference. Start with close the loop.