💬 Score customer churn risk and send save-play emails with Claude and Gmail
⚡ 188 views · 💬 Lead Nurturing & AI Sales Agents
Description
Quick overview
Catch at-risk paying customers before they cancel. This workflow receives customer health signals through a webhook, scores churn risk from 0–100 with Claude, logs every account to a data table, and automatically sends a personalized save-play email to the customers who are quietly slipping away.
How it works
- A Webhook receives customer health signals — last login, usage change, email open rate, support sentiment, days to renewal, open tickets and plan value — as JSON, one customer per call.
- A Set node normalizes the payload and applies safe defaults, so missing fields never break the run.
- Claude (Anthropic Chat Model + structured output parser) scores churn risk 0–100, assigns a RED / YELLOW / GREEN tier, lists the top signals, and drafts a personalized HTML save-play email.
- A Shape node flattens the AI output and the original data into one clean health record.
- The record is logged to a data table for tracking and reporting.
- A Switch routes by risk: RED triggers a save-play email, YELLOW a lighter nurture email, GREEN is logged and monitored. Email nodes retry automatically 3×.
Setup
- Connect your Anthropic credential on the AI Churn Score model node, and your Gmail (OAuth2) credential on the two email nodes.
- Make sure the customer_health_log data table exists with the mapped columns (customer_id, churn_score, risk_tier, action_taken, logged_at and more); create it if you are importing into a fresh instance.
- Activate the workflow and POST customer signals to the webhook URL. Test with the sample payload, then connect your CRM, billing or product-analytics tool to send real signals.
Requirements
- n8n instance with the Data Table feature enabled Anthropic API key (Claude) Gmail account connected via OAuth2
Customization
- Swap the Anthropic model for OpenAI or any other chat model without touching the flow Adjust the RED / YELLOW / GREEN thresholds or scoring weights directly in the system prompt Replace Gmail with Slack, Outlook or your CRM to change how alerts are delivered Add a Schedule trigger to batch-scan your whole customer list instead of pushing signals one by one
Additional info
Every downstream node references the Shape node explicitly, so fields never arrive null after the data-table insert — a common failure point in AI + data-table workflows. Fits any recurring-revenue business: SaaS, agencies, coaching, subscription e-commerce, and paid communities.
🔗 Nodes Used
Webhook, Gmail, AI Agent, Anthropic Chat Model, Structured Output Parser, Data table
📥 Import
Download workflow.json and import into n8n:
Workflow menu → Import from File