The Churn Hunter – Spotting Customers Who Are Leaving Before They Ghost

A Christmas Surprise

During a December retention review, a mid-market IT provider found a problem hiding in plain sight. Nearly four in ten of their customers were dormant, showing no meaningful engagement in over 60 days. Revenue was still coming in, but the relationships were already gone. The company had no system to tell which customers were salvageable and which were truly lost.

Where the logic failed

The team initially built an agent to look for “quiet” accounts. It flagged 47 “risky” customers. The Head of Customer Success called ten of them and found that most were perfectly happy.

The issue was that quiet periods are common in managed services, especially when things are working and don’t reliably indicate churn. The team had confused low usage with churn risk.

The Solution: Detector vs. Diagnoser

To add useful nuance, they built a two-stage system and you can too.

The Detector (Rules): A cheap, automated query that identifies anomalies.
The Diagnoser (Agent): An intelligent review of the context to understand why the anomaly happened.

Phase 1: The Detector (The Smoke)

Run a daily script to flag accounts where at least one of these hard signals is true.

Phase 2: The Diagnoser (The Fire)

For every account on the Review List, feed a context brief to the Agent. Its job is to classify the specific Strategic Drift driving disengagement.

Example Input Brief:

Customer: Acme Corp

Contract Type: Annual Managed Service

Tickets (Last 30d): 0

Usage Trend: Flat

Last QBR Sentiment: Mixed

Enrichment News: Missed earnings targets, New CFO hired

Support Themes: Price, Competitor mention

Notes (Last 90d): Project complete, Budget freeze mentioned

System Prompt:

Task: Analyze the customer brief and classify the account into one of four risk categories. Your output must be a single, structured diagnosis for this account.

Output Format: Provide a structured response with the following four fields: risk_score, drift_category, reasoning_summary, and strategic_recommendation.

Reasoning Logic:

• The Value Gap (“Job Done”): The customer has achieved their initial goal and now perceives the service as a recurring cost with no new value. Look for signals like completed projects combined with a sharp drop in usage.
• The Economic Gap (“ROI Failure”): The customer is under financial pressure and views the product as a line item to be cut. Look for signals like budget freezes, missed earnings, layoffs, or a new CFO.
• The Relationship Gap (“Political Drift”): Your internal champion has left, or you’ve been “ghosted” ahead of a renewal. The business case may still be valid, but you’ve lost your political cover.
• The Fit Gap (“Stack Divergence”): The customer’s technical needs are changing, often due to a merger, acquisition, or a new internal IT strategy. Your product is at risk of being engineered out of their stack.

Output Format:

Provide a structured response with the following four fields:

• risk_score: A numerical score from 1-10.
• drift_category: One of four specific categories: Value Gap, Economic Gap, Relationship Gap, or Fit Gap.
• reasoning_summary: A single sentence explaining why you chose that category.
• strategic_recommendation: The specific, high-level action the team should take next.

Example Output:

Phase 3: The Operational Workflow

The agent’s structured diagnosis must route automatically into the Customer Success system of record, such as Salesforce or HubSpot.

1. Ingest & Route Don’t dump the data into a spreadsheet. Set up a simple automation (using Zapier, Make, or a custom script) to process the agent’s daily output:

• Trigger: New structured diagnosis from Agent.
• Action: Create “High Priority Task” in CRM.
• Assignment: Route to Account Owner (if Relationship Gap) or Technical Account Manager (if Fit Gap).

2. The Playbook Mapping The CRM task description should automatically include the Strategic Diagnosis and the specific Human-Led Playbook to run.

Executive Takeaway

For the research behind this, see how AI is changing churn detection in UK B2B.

Most retention metrics flatter reality by counting payment, not intent. This two-stage system surfaces churn early enough to respond, transforming a lagging indicator into a real-time operational workflow. It directs your team’s attention with precision, ensuring that even if you still lose customers, you lose them on your own terms.