How to tell a customer is about to churn from a support conversation

The warning signs are usually in the last message, not the account data

Churn prediction models look at usage data such as login frequency and feature adoption, plus account level details like seat count. Useful signals, but they're lagging indicators. By the time usage drops, the decision to leave has often already been made. A support conversation can show you the moment it's being made.

Patterns worth watching

Going quiet after a real question stands out clearly. A customer asks something specific, gets an answer, and never replies again, not even a quick thanks. That silence after real engagement reads differently than a customer who never followed up in the first place.

A repeat problem tells a different story. One bug report is just a bug report. The same customer flagging the same category of issue twice in a month is a pattern, and it usually means the first fix either didn't land or didn't get communicated clearly.

A frustrated message with no resolution in sight deserves more attention than most support tools give it. Tone is a signal. A message that's short and clipped, or one that's visibly annoyed, ends a conversation without a clear next step, and that carries more weight than a neutral message that ends the same way.

Why this is hard to catch by hand

None of these patterns are subtle once you're looking at a single conversation. The problem is volume. If you're handling even a hundred conversations a month, going back through them looking for who went quiet right after a frustrated message isn't something anyone has time to do consistently. It tends to get caught by accident, usually after the customer has already left.

You can score a specific account yourself with the churn risk calculator, using how many days it's been since their last reply and how that last message read. It's a simplified version of what a system watching every conversation automatically would flag on its own, which is the actual point of building one.

Try it on an account you're already unsure about. If it comes back high, that's your cue to reach out before the usage numbers ever move.