Risk Mitigation Matching

Given a structured risk extracted from a sales call, match it to the most effective mitigation asset from a content library. Returns a ranked list of assets with an explanation of why each addresses the specific concern, plus a draft delivery message.

Prerequisites

• Anthropic API key (Claude)
• Structured risk data from call-transcript-risk-extraction
• Mitigation content library indexed in Attio or a structured JSON file (case studies, security whitepapers, ROI calculators, reference contacts, product documentation, guarantees)

Steps

1. Pull the risk data

Receive the risk object:

{
  "id": "risk-001",
  "category": "technical",
  "summary": "Worried about data migration complexity from legacy system",
  "concern_quote": "We've been burned before by migrations that took 6 months instead of 6 weeks",
  "severity": 8,
  "likelihood": 7,
  "mitigation_type_needed": "case_study"
}

2. Query the mitigation content library

Pull all available mitigation assets from Attio (or a local JSON file). Each asset should have:

{
  "id": "asset-001",
  "type": "case_study|security_doc|roi_proof|reference_contact|product_doc|guarantee|technical_demo",
  "title": "...",
  "risk_categories_addressed": ["technical", "timeline"],
  "industry": "...",
  "company_size": "...",
  "key_claims": ["Migration completed in 4 weeks", "Zero data loss"],
  "url_or_path": "..."
}

3. Match risk to assets via Claude API

POST https://api.anthropic.com/v1/messages
Authorization: x-api-key {ANTHROPIC_API_KEY}
Content-Type: application/json

Request body:

{
  "model": "claude-sonnet-4-20250514",
  "max_tokens": 1500,
  "messages": [{
    "role": "user",
    "content": "A sales prospect expressed this concern:\nRisk: {risk_summary}\nExact quote: \"{concern_quote}\"\nCategory: {category}\nSeverity: {severity}/10\n\nHere are available mitigation assets:\n{json_array_of_assets}\n\nRank the top 3 assets that best address this specific concern. For each:\n1. Asset ID and title\n2. Why it addresses this risk (reference specific claims from the asset)\n3. A 2-3 sentence delivery message that connects the asset to the prospect's exact words\n4. Confidence score 0.0-1.0 that this asset will reduce the prospect's concern\n\nReturn JSON:\n{\n  \"matched_assets\": [\n    {\n      \"asset_id\": \"...\",\n      \"title\": \"...\",\n      \"relevance_explanation\": \"...\",\n      \"delivery_message\": \"...\",\n      \"confidence\": 0.0\n    }\n  ],\n  \"gap_identified\": true/false,\n  \"gap_description\": \"If no asset scores above 0.6 confidence, describe what content should be created\"\n}"
  }]
}

4. Handle content gaps

If gap_identified is true (no asset has confidence >= 0.6), log this as a content gap:

• Fire PostHog event mitigation_content_gap with the risk category and summary
• Create an Attio note: "Content gap: need {mitigation_type_needed} for {risk_category} risk: {summary}"
• Fall back to a generic response framework for the risk category

5. Return the match result

Output the ranked asset list with delivery messages ready for the seller to send or the agent to deliver automatically.

Error Handling

• Empty content library: Return gap_identified: true for all risks. The team needs to build mitigation assets.
• No matching category: Expand search to all assets regardless of risk_categories_addressed. Cross-category assets sometimes address the underlying concern.
• LLM returns malformed JSON: Retry once. If still malformed, return top asset by simple category matching (no LLM).

Alternatives

| Tool | Method | Notes | |------|--------|-------| | Claude | Anthropic API | Best semantic matching | | GPT-4 | OpenAI API | Alternative LLM | | Embeddings search | OpenAI or Voyage embeddings | Better at scale with 100+ assets | | Manual lookup | Human searches content library | Fallback |