Closed Lost Revival Agent

Challenge

Marq's sales team was sitting on hundreds of closed-lost deals with no reliable way to know which ones were worth revisiting. Re-engagement decisions were ad hoc, dependent on individual reps remembering to check, and often skipped entirely. The result: missed revenue from deals that could have been revived with the right timing and context.

Approach

• Researched and compared 5 deep research APIs (Perplexity, Exa, OpenAI, Gemini, Tavily) and recommended a tiered approach: Perplexity for MVP, Exa + Gemini for production
• Designed the evaluation logic to assess close reason changeability, deal size, competitive landscape, and time elapsed
• Architected a three-stage pipeline: data gathering, AI research, LLM evaluation
• Built and tested end-to-end in n8n with real closed-lost deal data
• Proposed 5 suppression rules to prevent duplicate outreach

How It Works

HubSpot workflow triggers webhooks at 60, 180, and 360 days after a deal is closed-lost. From there:

1. n8n pulls deal properties, company details, email history, and call records from HubSpot
2. Perplexity Sonar Deep Research runs autonomous web research on the company, looking for rebrands, leadership changes, funding rounds, or other signals
3. Gemini 3 Flash evaluates revival potential based on deal context and research findings
4. If qualified: drafts a personalized re-engagement email, creates a new revival deal in HubSpot, and creates a task for the sales rep
5. If not qualified: ends silently with no noise in the CRM

Deliverables

• Full 12-node n8n workflow (HubSpot trigger, data enrichment, Perplexity research, Gemini evaluation, conditional branching, deal/task creation)
• LLM evaluation prompts engineered for structured JSON output
• API comparison report with recommendation rationale
• Suppression logic proposal (awaiting VP of Sales input before deployment)
• Test results across 4 real closed-lost deals

Results

• Eliminated manual deal review for re-engagement decisions
• Personalized outreach drafted automatically, ready for rep review before sending
• Qualification logic validated against real deal data
• Core workflow complete and tested, pending suppression rules and final deployment

Tools Used

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