AI Deal Analyzer

Analyze any B2B deal for risk level, estimated margin, and AI recommendation.

How It Works

1. Enter deal details (product, supplier, price, quantity)
2. Claude AI analyzes risk factors and margin potential
3. Get structured output: Risk Level, Estimated Margin, Recommendation

Tech Stack

• Claude API (Anthropic)
• Node.js
• HTML/CSS (no framework, keep it simple)

How to Run

ANTHROPIC_API_KEY=your_key node server.js

Open http://localhost:3000 in your browser.

Example Output

{
  "risk_level": "Medium",
  "estimated_margin": "18-24%",
  "recommendation": "Proceed with caution — request product samples before full order.",
  "key_factors": "New supplier, competitive pricing, high demand product"
}

Scoring Engine (deterministic logic before AI)

Before any Claude API call, the server applies deterministic scoring rules:

• Margin threshold check: if estimated margin falls below 10%, the deal is flagged as high risk regardless of AI output
• Supplier age heuristic: new suppliers (unverified history) automatically increase risk score by one level
• Quantity-to-price ratio: unusually low unit prices for high quantities trigger a fraud-risk flag
• Category risk multiplier: certain product categories (electronics, supplements) carry a baseline risk premium

These rules run first. The AI then receives the pre-scored deal as context, which anchors its reasoning rather than relying on general inference alone.

Structured Output (JSON example)

The API returns a consistent JSON structure on every request:

{
  "risk_level": "Medium",
  "estimated_margin": "18-24%",
  "recommendation": "Proceed with caution — request product samples before full order.",
  "key_factors": "New supplier, competitive pricing, high demand product",
  "flags": ["unverified_supplier", "margin_near_threshold"],
  "confidence": "high"
}

Fields are validated server-side before returning to the client. Malformed AI responses are retried once before returning an error.

System Design Flow

User Input (product, supplier, price, quantity)
        ↓
Deterministic Scoring Engine (margin check, supplier flags, category risk)
        ↓
Claude API Call (pre-scored context injected into prompt)
        ↓
JSON Response Parsing + Validation
        ↓
Structured Output → Frontend Display