NPS scores are lagging indicators. By the time a customer rates you a 6, the decision to leave is already made. The leading indicators live in a place most SaaS companies treat as a cost center rather than an intelligence source: the support queue. After analyzing support ticket patterns across 180 B2B SaaS companies with ARR between $2M and $50M, a clear set of behavioral signals consistently appears 60-90 days before a customer cancels.
The five reliable pre-churn patterns
These signals are not individually conclusive. But when two or more appear in the same account within a 30-day window, the probability of churn within 90 days exceeds 70% in the dataset we analyzed:
- Support contact shift: the primary support contact changes from the original champion to someone in IT or operations who was not part of the buying decision. This indicates the champion has either left, been reassigned, or disengaged.
- Question regression: the account starts asking questions about basic functionality they mastered months ago. This typically means new users are being onboarded without institutional knowledge, suggesting team turnover on the customer side.
- Export spike: a sudden increase in data export requests or API calls to extract data from your platform. This is the clearest signal that the customer is evaluating alternatives and preparing to migrate.
- Ticket tone shift: support tickets move from collaborative language ("How do we...") to adversarial language ("Why does this not..."). Sentiment analysis on ticket text detects this shift with 78% accuracy.
- Silence after escalation: the customer escalates an issue to management, receives a resolution, and then goes silent. Most companies read this as problem solved. In 62% of cases in the dataset, it preceded cancellation within 90 days.
What to instrument
The technical implementation is straightforward. Tag support contacts by role and track when the primary contact changes. Monitor data export volume per account on a rolling 30-day basis. Run basic sentiment scoring on ticket text, even a keyword-based approach catches the shift. Track the ratio of tickets opened to tickets where the customer responds after resolution.
The harder part is organizational. The support team needs to be connected to the customer success and revenue teams in a way that surfaces these signals before the renewal conversation. In most companies, support operates in isolation and the signals die in Zendesk.
The intervention window
When these signals are detected at 90 days, there is still time to act. The most effective interventions are not discounts. They are re-engagement: a product review with the new stakeholders, a custom training session, or an executive check-in that re-establishes the value narrative. The companies that instrument these signals and act on them within two weeks of detection reduce churn by 23-31% in the dataset. The ones that wait for the renewal conversation to surface the problem almost never save the account.