Marq - AI CRM Automation: Churned Customer Revival Agent

Marq was sitting on a large pool of churned customers and had no systematic way to know which ones might come back. I built an AI win-back agent that evaluates every fully churned deal at 180, 540, and 900 days, cross-references the original churn reason against Marq's current capabilities, drafts a personalized win-back email, and notifies the account owner via Slack - turning a dormant book of business into a structured revival pipeline.

Challenge

Churn reasons aren't static. A customer who left in 2024 because Marq lacked a specific feature might be a great fit in 2026 if that feature shipped. A customer who left because of a budget freeze might have budget again. A customer whose champion left during a reorg might now be reachable through new leadership. None of that signal was being acted on.

The agent also had to behave differently than its sister project for prospects. Former customers know the product, remember the experience, and need a win-back tone - not a cold pitch. And the most-mentioned competitor in churn descriptions was Canva, so the LLM prompts needed specific guidance on how to reposition Marq against it.

Approach

I built the agent in n8n, triggered by a HubSpot webhook the moment a churned deal hits 180, 540, or 900 days in Marq's churn pipeline. The agent runs suppression checks, gathers deal history and CRM context, enriches it with Marq's product knowledge base via RAG, runs deep research on the company through Perplexity Sonar to surface rebrands, leadership changes, or funding rounds, and sends everything to Gemini for a revival assessment.

The evaluation prompt is built around a churn_reason_resolved flag: the LLM has to explicitly cross-reference the original churn reason against Marq's current capabilities before recommending revival. If a customer left because of a missing feature, that feature has to actually exist now. If they left for budget reasons, enough time has to have passed for that to plausibly have changed. This is what stops the agent from re-pitching customers into the same wall they hit the first time.

If the deal qualifies, the agent drafts a personalized win-back email - tone calibrated for someone who already knows the product, with built-in repositioning guidance for the competitor most frequently named in past churn conversations. It then creates a revival deal in HubSpot and sends the original account owner a Slack DM with the full context. If it disqualifies, it logs the reasoning to a monitoring channel so nothing disappears silently.

Re-engagement intervals (180, 540, and 900 days) were chosen with VP Operations to give former customers enough cooldown that a win-back attempt feels timely rather than tone-deaf.

Results

• A previously dormant pool of churned customers brought into a structured, automated revival pipeline
• Churn-reason-aware evaluation: revival only triggers when the original blocker has actually been resolved by Marq's current product
• Win-back emails drafted automatically with built-in competitor-repositioning guidance, ready for the account owner to review before sending
• Account owners notified via Slack DM the moment a revival opportunity is created in HubSpot
• Core workflow built and end-to-end tested; pending final suppression-rule sign-off from VP Sales before production activation

Tools Used

n8n, HubSpot API v3, Gemini 3 Flash (via Google AI Studio), Perplexity Sonar Deep Research, OpenRouter, Google Drive API, Slack API