Seventy percent of churn is predictable. The signals were there. Nobody had a system to read them.
Most CS teams inherit a customer health score that’s really just NPS plus login frequency, wrapped in a red-yellow-green color scheme, reviewed in a monthly business review nobody finds useful. That’s not a health score. That’s a comfort blanket.
A health score model that actually predicts churn needs four specific inputs, threshold logic that fires before damage is done, and an operator who knows what to do when a flag goes red. Here’s how to build it.
The Four Inputs That Actually Matter
Every customer health scoring system worth running draws from the same four layers. Miss one and you’ll have blind spots. Combine all four and you can see churn coming 60 to 90 days out.
1. Product engagement. This is the foundation. You’re looking for frequency, depth, and trend. Frequency is how often users log in. Depth is which features they’re using, specifically whether they’re using the features tied to your core value proposition. Trend is the direction over the last 30 and 60 days. A customer who logs in daily but only uses one surface-level feature is not an engaged customer. They’re an at-risk one.
2. Support burden. High ticket volume isn’t automatically a red flag. The composition matters. Tickets about how to do something are yellow. Tickets about things not working as expected are red. Repeated tickets on the same issue without resolution are a fire alarm. Track ticket volume, ticket category, and days-to-resolution. A customer submitting five tickets a month about broken workflows is signaling something your product data alone won’t tell you.
3. Commercial trajectory. This layer looks at the customer’s financial relationship with you. Are they on a contract that’s growing, flat, or shrinking? Have they turned down an upsell conversation in the last 90 days? Did they push back on renewal terms? These are lagging indicators compared to the first two, but they’re concrete. A customer with flat ACV for three consecutive periods and no expansion conversations is not a healthy customer, regardless of what their login data says.
4. Relationship depth. This is the hardest to quantify and the most important. You want to know: who in their org actually uses the product, who owns the relationship on their side, and when you last had a meaningful conversation that wasn’t about a support issue. One champion at the director level with no backup is a single point of failure. If that champion leaves, the contract goes with them. Measure recency of executive contact, number of active internal users, and whether you have relationships at more than one level of the org.
How to Weight the Inputs
The weighting depends on your product and sales motion, but here’s a framework that works for most B2B SaaS companies with contracts above $25K ARR:
| Input | Weight | Primary Signal |
|---|---|---|
| Product engagement | 35% | Core feature usage trend (30-day) |
| Support burden | 25% | Ticket category + resolution time |
| Commercial trajectory | 25% | ACV movement + expansion signal |
| Relationship depth | 15% | Champion coverage + exec recency |
Each input should score 0 to 100. Composite score of 75 or above is healthy. 50 to 74 is watch. Below 50 is intervene now. These thresholds sound simple, and they are. The value isn’t in the thresholds. It’s in building the scoring logic so the number actually moves when something real changes, not just when someone manually updates a field in your CRM.
Threshold Logic That Triggers Action
A customer success health score is worthless without intervention rules attached to it. The score is not the output. The action is the output.
When a customer drops from healthy to watch, the trigger is an asynchronous check-in within five business days. Not a QBR. Not a formal meeting request. A direct message or a short call from the CSM that sounds human: “Noticed your team’s usage pattern shifted a bit over the last month. Wanted to make sure everything’s landing the way it should.”
When a customer drops from watch to intervene, the trigger is an escalation within 48 hours. The CSM brings in a senior operator or account lead. The conversation shifts from relationship maintenance to active problem-solving. You need to understand what changed, whether there’s a fixable issue, and whether there’s a competitive threat you don’t know about yet.
The 90-day window matters most. That’s when usage patterns set in, when the internal champion either becomes an advocate or starts second-guessing the purchase, and when the customer’s perception of your product is most malleable. If you don’t have threshold logic firing in the first 90 days, you’re reacting to churn instead of preventing it.
What Breaks Most Health Score Models
Two failure modes kill customer health scoring before it starts.
The first is manual inputs. If a CSM has to update a field in the CRM to change a health score, the score will always be stale. The data needs to flow automatically from your product, your support platform, and your billing system. If your CRM isn’t pulling usage data via API, the health score is a guess dressed up as a number. This is a RevOps architecture problem before it’s a CS problem.
The second is treating the score as a reporting tool instead of an operating tool. Health scores that live in a dashboard nobody opens between QBRs are decorative. The score should be visible in the CSM’s daily workflow, connected to task triggers, and reviewed in weekly team standups. If your CS team is only looking at health data when preparing for renewal conversations, you’ve already missed the window to intervene.
For more on how CSAT, NPS, and CES interact with a health scoring model, this breakdown of the three metrics is worth reading alongside this framework.
How Phi’s CS Pod Runs This in the First 90 Days
When Phi’s customer success pod embeds in a client org, the first 90 days aren’t about relationship building. They’re about building the scoring infrastructure and catching anything already at risk.
Week one is an audit. We look at what customer health metrics are being tracked, where they live, and how automated the data flow actually is. Most companies have fragments of a health score: someone tracks NPS, someone else monitors ticket volume, a third person keeps an eye on login rates. Nobody has connected them into a single composite score with threshold logic.
By week three, the scoring model is live and connected to the CRM. By week six, intervention playbooks are running for any account that dropped below the watch threshold during the audit window. The accounts that were already at risk get the most attention first.
AtoB ran this exact process across thousands of fleet accounts. The result was a 40% improvement in CSAT and a retention engine that scaled without adding proportional headcount.
Churn doesn’t announce itself. It accumulates quietly across four data layers while your CS team is busy preparing slide decks for QBRs. The companies that get retention right have stopped treating health scores as a reporting exercise and started treating them as an operating system. If yours isn’t connected to automated triggers and real intervention playbooks, you’re not measuring health. You’re measuring history.
