A customer submits a ticket every week for three weeks. Each time, the issue is different. Password problem, billing question, feature confusion. Each ticket gets resolved. CSAT is fine. Your metrics say everything is good.

A month later, they cancel. Your retention team asks why. "I was spending too much time dealing with issues." The signs were there. Nobody was looking at the right data.

Most support metrics are backward-looking. They tell you what happened: how many tickets, how fast you responded, how satisfied the customer was. They don't tell you what's about to happen.

But some patterns in support data are forward-looking. They predict churn before the customer decides to leave. If you track these, you can intervene before it's too late.

Repeat Contact Rate

This is the strongest churn predictor in support data. Customers who contact you 3 or more times in a 30-day period churn at 2x to 3x the rate of customers who contact you zero or one time.

Why? Because frequent contact means the product isn't working well for them. Each contact is a friction point. Even if every interaction is positive and every issue gets resolved, the cumulative effort is draining.

How to track it: count unique customers who submit 3+ tickets in a rolling 30-day window. Flag them automatically.

What to do about it: when a customer hits the threshold, trigger a proactive outreach. Not a generic "how can we help?" email. A specific message: "I noticed you've had a few issues recently. I'd like to make sure everything is working well for you. Do you have 15 minutes for a call?" This personal attention catches people before they decide to leave.

The outreach works best when it comes from someone with authority (a manager, a customer success person, even the founder for high-value accounts), not from the support agent who handled the last ticket.

Escalation Rate Trends

Your overall escalation rate might be 15%. That's fine. But if it was 10% three months ago and 15% now, something is getting worse.

Rising escalation rates usually mean one of two things:

Your product is getting more complex or buggy (new features breaking things, edge cases multiplying). This is a product signal.

Your front-line agents are less capable of resolving issues (new hires, training gaps, tool problems). This is an operational signal.

Either way, more escalations mean more effort for customers and more load on your senior team. Track escalation rate weekly and look at the trend line, not just the current number.

Sudden spikes in escalation rate often correspond to product releases. If you pushed a new version on Tuesday and escalations jumped 40% on Wednesday, you have a quality problem. Flag it to engineering immediately.

Response Time by Customer Segment

Average response time across all customers might look great: 2 hours. But what if your enterprise customers (who pay $5,000/month) are getting 4-hour response times because they happen to submit tickets during your busiest hours?

Break response time down by customer segment: by plan tier, by revenue, by tenure, by whatever segmentation matters for your business.

If your highest-value customers are getting slower responses than average, that's a churn signal. They're paying more and getting less. They know it, even if they don't say it yet.

Fix this with priority routing. AI classification tools like Supp can assign priority scores based on customer value, intent urgency, and ticket content. High-value customers get routed to the front of the queue automatically.

Sentiment Shift

A single negative interaction doesn't predict churn. People have bad days. But a pattern of declining sentiment over time is a strong signal.

If a customer's last 5 interactions show a CSAT trajectory of 5, 5, 4, 3, 2, they're heading for the exit. Even if the current interaction's CSAT is 3 (which most dashboards would show as "okay"), the trend says "leaving."

How to track it: for each customer, store their last 5 CSAT scores. Calculate the trend (simple linear regression or even just "is the last score lower than the average of the previous scores?"). Flag customers with declining trends.

Not every company has enough CSAT data per customer to do this. If your CSAT response rate is 15% and a customer only contacts you twice per year, you'll never get enough data points. This metric works best for high-contact businesses (SaaS, subscriptions, marketplaces).

Time Between Contacts and Silence

This one's counterintuitive. A customer who contacts you regularly and then suddenly stops is a churn risk.

Regular contact means the customer is engaged. They're using the product. They have questions. They care enough to ask. When that pattern breaks, something changed. Maybe they found an alternative. Maybe they stopped using the product. Maybe they're about to cancel and just haven't gotten around to it.

Track "time since last contact" for active customers. If a customer typically contacts you every 2 to 3 weeks and goes silent for 6 weeks, that's a signal worth investigating.

This metric works best when combined with product usage data. If the customer is still logging in and using the product but not contacting support, they're probably fine. If they've stopped logging in AND stopped contacting support, they've already mentally churned.

How to Build a Churn Warning System

You don't need a fancy predictive model. A simple scoring system works:

Give each customer a churn risk score based on:

• 3+ support contacts in 30 days: +2 points
• Declining CSAT trend (last score below average): +2 points
• Response time above SLA: +1 point
• Recent escalation: +1 point
• Silence (no contact when expected): +1 point
• Negative sentiment in most recent ticket: +1 point

Score of 0 to 2: low risk. Business as usual.

Score of 3 to 4: medium risk. Add to a watchlist. Proactive check-in within the week.

Score of 5+: high risk. Immediate outreach. Personal contact from a manager or CSM.

Update scores weekly. Review the high-risk list in your team meeting. Assign outreach to specific people with deadlines.

This system won't catch every churning customer. But it'll catch the ones that support data can predict, which is a meaningful portion. Combined with product usage data and billing signals (failed payments, downgrades), you'll have a full early warning system.

Supp's analytics dashboard tracks intent distribution, response times, and escalation patterns per customer. Exporting this data (CSV or JSON) into a spreadsheet or BI tool lets you build the scoring model above without writing code. The classification data is the foundation, because it tells you not just how often a customer contacts you, but what they're contacting you about.