Building Automated Deal Flow Pipelines with AI and n8n
TL;DR
Build an automated deal flow pipeline that sources opportunities from RSS/APIs, screens against your investment criteria, scores deals quantitatively, and delivers weekly digests to your team — using n8n, Claude agents, and persistent memory.
Venture capital firms see thousands of deals per year. Angels and solo investors see hundreds. The bottleneck isn't finding opportunities — it's systematically evaluating them without burning all your time on screening.
n8n, now valued at $2.5B after a $180M Series C, is the workflow automation platform that makes AI deal flow pipelines practical. Combined with Claude agents for intelligent scoring and persistent memory for context, you can build a system that runs continuously and gets smarter over time.
Here's the architecture.
Pipeline Architecture
Data Sources → Ingestion → Screening → Scoring → Pipeline → Digest
↓
Human Review
Each stage is a separate workflow component that can be tested and tuned independently:
| Stage | Tool | What It Does |
|---|---|---|
| Ingestion | n8n triggers (RSS, webhook, cron) | Monitors 10-30 data sources for new deals |
| Deduplication | n8n + memory | Checks if we've already seen this company |
| Screening | Claude agent + criteria | Quick pass/fail against investment thesis |
| Scoring | Claude agent + scoring framework | Quantitative 0-100 score across 7 dimensions |
| Pipeline | n8n + Notion/Airtable | Stage tracking from Sourced through Close |
| Digest | Claude agent + template | Weekly summary to team (Slack/email) |
Stage 1: Data Source Ingestion
The first n8n workflow monitors your deal sources:
RSS Feeds (15-second setup each):
- TechCrunch Funding section
- Crunchbase funding alerts
- VentureBeat deal announcements
- Sector-specific blogs (e.g., AI, SaaS, fintech)
API Connections (requires credentials):
- Crunchbase API: New funding rounds in target sectors
- PitchBook: Deal alerts matching your criteria
- AngelList: New syndicates in your focus areas
- ProductHunt: Top launches in target categories
Scheduled Scraping (Playwright/Puppeteer):
- Y Combinator batch pages
- SEC EDGAR for S-1 filings
- Specific accelerator demo day pages
Each source pipes deals into a standardized format:
{
"company": "Acme AI",
"sector": "Enterprise SaaS",
"stage": "Series A",
"raised": "$12M",
"investors": ["Sequoia", "a16z"],
"source": "crunchbase",
"sourceDate": "2026-02-14",
"url": "https://...",
"description": "AI-powered customer support platform"
}
Stage 2: Deduplication and Enrichment
Before scoring, the pipeline checks memory:
- Already seen? Skip if processed in the last 90 days
- Related to an existing deal? Flag as potential follow-on
- In a tracked sector? Add sector context from prior research
The enrichment step pulls additional data:
- Funding history (all rounds, total raised)
- Team LinkedIn profiles (founders, key hires)
- Product metrics (if publicly available)
- Competitor context (from memory)
Stage 3: Quick Screen Against Criteria
A Claude agent runs a rapid pass/fail against your investment criteria:
## My Investment Criteria
- Stage: Seed to Series B
- Sector: AI/ML, Enterprise SaaS, Developer Tools
- Geography: US, Europe
- Minimum: Product launched, some revenue signal
- Red flags: Solo non-technical founder, pivoted 3+ times
Deals that clearly fail criteria get tagged PASS with reason. Deals that clearly match get tagged SCREEN. Ambiguous deals get tagged REVIEW for human triage.
This step filters out 60-70% of inbound deals immediately.
Stage 4: Quantitative Scoring
Deals that pass screening get scored across 7 weighted dimensions:
| Criterion | Weight | Scoring Guide |
|---|---|---|
| Market Size (TAM) | 15% | >$10B = 10, $1-10B = 7, <$1B = 4 |
| Growth Rate | 15% | >50% YoY = 10, 20-50% = 7, <20% = 4 |
| Team Strength | 20% | Serial founders = 10, experienced = 7, first-time = 4 |
| Product-Market Fit | 15% | Revenue + retention = 10, revenue only = 7, pre-revenue = 4 |
| Competitive Moat | 15% | Strong (network/IP) = 10, moderate = 7, weak = 4 |
| Deal Terms | 10% | Below market = 10, at market = 7, above = 4 |
| Strategic Fit | 10% | Core thesis = 10, adjacent = 7, outside = 4 |
The Claude agent researches each dimension and produces a score card:
Score: 73/100 — RECOMMEND FIRST LOOK
Market Size: 8/10 (15%) = 1.2 — $15B TAM, growing
Growth Rate: 7/10 (15%) = 1.05 — ~35% YoY
Team: 9/10 (20%) = 1.8 — Serial founder, strong CTO
PMF: 6/10 (15%) = 0.9 — Revenue but retention unclear
Moat: 7/10 (15%) = 1.05 — Data advantage, switching costs
Terms: 7/10 (10%) = 0.7 — At market for stage
Fit: 8/10 (10%) = 0.8 — Core thesis alignment
Thresholds:
- Score >= 70: Recommend Full Due Diligence
- Score 50-69: Conditional — needs specific data points
- Score < 50: Pass with documented reason
Stage 5: Pipeline Management
Scored deals enter a pipeline with stage tracking:
Sourced → Screened → First Look → Full Due Diligence → IC Review → Term Sheet → Closed
Each stage transition requires:
- Screened → First Look: Score >= 50, human confirms
- First Look → Full Due Diligence: Score >= 65, DD Lead initiates
- Full Due Diligence → IC Review: DD report complete, score >= 70
- IC Review → Term Sheet: Investment committee approval
- Term Sheet → Closed: Legal and final terms
The pipeline syncs to Notion (for visual kanban) or Airtable (for structured data), providing the team with a real-time view.
Stage 6: Weekly Digest
Every Monday morning, a Claude agent compiles the week's pipeline activity:
# Deal Flow Digest: Week of Feb 10, 2026
## Pipeline Summary
| Stage | Count | Change |
| ---------- | ----- | ------ |
| Sourced | 47 | +23 |
| Screened | 12 | +8 |
| First Look | 3 | +2 |
| Full Due Diligence | 1 | +0 |
## Top Scored Deals This Week
1. **Acme AI** — Score: 78/100 — AI customer support, Series A, $12M
2. **DataForge** — Score: 73/100 — Developer data tools, Seed, $4M
3. **NeuralOps** — Score: 71/100 — AI operations platform, Series A, $8M
## Deals Advanced
- Acme AI: Screened → First Look (strong PMF signals)
## Deals Passed
- FooBar Inc: Passed at Screening (solo non-technical founder)
- BazCorp: Passed at Scoring (TAM < $500M)
This gets delivered via Slack, email (Resend), or both.
Why This Compounds Over Time
The persistent memory layer is what transforms this from a workflow into an intelligence system:
- Sector context grows: Every deal analyzed adds to your sector knowledge. By deal #50, the agent has deep context on competitive dynamics, typical metrics, and market positioning.
- Scoring calibrates: Track which scored deals you actually invested in and how they performed. Use this to tune scoring weights.
- Pattern recognition: After 6 months, the system identifies patterns — "Companies in sector X with metric Y above Z tend to score well" — that inform future screening.
Getting Started
Minimum viable pipeline (2-3 hours to set up):
- 5 RSS feeds in n8n (TechCrunch, Crunchbase, 3 sector blogs)
- Claude agent for quick screen against your criteria
- Notion database for pipeline tracking
- Weekly digest to your email
Full system (use IACOS):
- Clone investor-intelligence repo
- Configure IACOS with your investment criteria
- Connect n8n workflows for automated ingestion
- Set up MCP servers for persistent memory and web research
- See the IACOS Setup Guide for step-by-step instructions
The infrastructure investment pays for itself after the first month. One missed deal that your pipeline would have caught easily justifies the setup time.
For the complete system architecture, visit AI Investment Intelligence in the Research Hub, or explore the Investor Intelligence tools on frankx.ai.
Read on FrankX.AI — AI Architecture, Music & Creator Intelligence
Stay in the intelligence loop
Field notes on AI systems, production patterns, and builder strategy.
Continue Reading

How AI Agents Are Transforming Due Diligence in 2026
How orchestrated multi-agent systems compress investment due diligence from 40+ hours to minutes while improving audit thoroughness and trust.
Read article
The Ultimate n8n Workflow in 2026
The complete 2026 guide to building production n8n workflows: self-host vs cloud, the AI Agent node, MCP Server Trigger and Client Tool, content repurposing, RAG, and wiring n8n as the ...
Read article
Best No-Code AI Agent Builders 2026: n8n vs Lindy vs Relay vs Make
A hands-on comparison of the top no-code AI agent builders in 2026 — n8n, Lindy, Relay.app, Make, Zapier, Gumloop — judged on ease, power, self-hosting, MCP support, and price.
Read article