Stakeholder Role Classification

Use Claude or another LLM to classify contacts at a target account into buying committee roles based on their title, seniority, department, and any interaction history. Produces a structured role assignment for each stakeholder.

Prerequisites

• List of contacts at a target account with: name, title, department, seniority level
• Optional: interaction history (emails, call notes, meeting transcripts)
• Anthropic API key or Claude MCP access

Buying Committee Roles

| Role | Definition | Typical Signals | |------|-----------|----------------| | Economic Buyer | Controls budget, signs the contract | C-Suite, VP Finance, "Head of" with P&L ownership | | Champion | Internal advocate who sells on your behalf | Engaged contact, asks for materials to share internally, refers you to others | | Influencer | Shapes the decision but does not own it | Technical leads, architects, team leads who evaluate tools | | Blocker | Actively or passively resists the purchase | Security, legal, procurement, or someone loyal to incumbent vendor | | End User | Will use the product daily | ICs, managers in the target department | | Gatekeeper | Controls access to the Economic Buyer | Executive assistants, Chiefs of Staff |

Steps

1. Classify by title and department (rule-based first pass)

Apply deterministic rules before using AI:

IF title contains "CEO" OR "CRO" OR "CFO" OR "VP" with P&L keywords → Economic Buyer
IF title contains "Procurement" OR "Legal" OR "Compliance" OR "Security" → Potential Blocker
IF title contains "Engineer" OR "Developer" OR "Analyst" (IC level) → End User
IF title contains "Director" OR "Manager" in target department → Influencer

This first pass is fast and free. Flag contacts that do not match any rule for AI classification.

2. Classify ambiguous contacts with Claude

API call:

POST https://api.anthropic.com/v1/messages
Authorization: Bearer {ANTHROPIC_API_KEY}
Content-Type: application/json
x-api-version: 2023-06-01

{
  "model": "claude-sonnet-4-20250514",
  "max_tokens": 500,
  "messages": [
    {
      "role": "user",
      "content": "You are classifying stakeholders in a B2B software buying process.\n\nCompany: {company_name}\nProduct being sold: {your_product_description}\nContact: {name}, {title}, {department}\nInteraction history: {summary_of_interactions_or_none}\n\nClassify this person into exactly ONE primary role: Economic Buyer, Champion, Influencer, Blocker, End User, or Gatekeeper.\n\nReturn JSON:\n{\"role\": \"...\", \"confidence\": \"high|medium|low\", \"reasoning\": \"one sentence\"}"
    }
  ]
}

3. Incorporate interaction signals

If you have call transcripts or email threads, add behavioral signals to improve classification:

• Champion signals: Asks "What materials can I share with my team?", introduces you to other stakeholders, responds quickly, volunteers internal context
• Blocker signals: Asks about competitor features, mentions "we already have a solution", delays meetings, CC's procurement/legal early
• Economic Buyer signals: Asks about pricing, ROI, contract terms, implementation timeline
• Influencer signals: Asks detailed technical questions, requests a sandbox or trial, evaluates against specific criteria

Use transcript-insight-extraction fundamental to pull these signals from Fireflies transcripts.

4. Score confidence and flag unknowns

For each classification, assign confidence:

• High: Title + behavior clearly indicate role (e.g., CFO asking about pricing = Economic Buyer)
• Medium: Title suggests role but no behavioral confirmation (e.g., VP Engineering who has not engaged yet)
• Low: Ambiguous title, no interaction data (e.g., "Senior Director, Strategy")

Flag all low-confidence classifications for human review during discovery calls.

5. Store classifications in CRM

Write role classifications to Attio using attio-custom-attributes:

{
  "data": {
    "values": {
      "stakeholder_role": [{"option": "Champion"}],
      "stakeholder_confidence": [{"option": "High"}],
      "stakeholder_sentiment": [{"option": "Supportive"}]
    }
  }
}

6. Re-classify as deals progress

Roles are not static. A Champion can become a Blocker if priorities shift. Re-run classification after every significant interaction (discovery call, demo, negotiation). Compare new classification to previous and flag changes for the sales team.

Via Clay (Claygent)

For batch classification without API calls:

Claygent prompt: "Given that {Full Name} is {Title} at {Company Name} in the {Department} department, classify their likely role in a B2B software buying decision. Choose one: Economic Buyer, Champion, Influencer, Blocker, End User, Gatekeeper. Return the role and a one-sentence explanation."

Cost: 5-10 credits per classification. Use for initial bulk classification of new accounts.

Tool Alternatives

| Tool | Method | Notes | |------|--------|-------| | Anthropic Claude | Messages API | Best reasoning for ambiguous cases | | OpenAI GPT-4 | Chat Completions API | Good alternative, similar quality | | Clay Claygent | Built-in AI column | Convenient for batch processing in Clay | | Gong | Deal intelligence | Infers roles from call participation patterns | | 6sense | Buying team detection | Enterprise-grade intent + role mapping |

Error Handling

• Over-classifying Champions: Not every friendly contact is a Champion. A Champion must have organizational influence AND willingness to advocate. Friendly but junior contacts are End Users.
• Missing Economic Buyer: If no one in the map is classified as Economic Buyer, the deal is at risk. Prioritize discovery to find them.
• Role conflicts: Two people classified as Economic Buyer suggests either a matrix organization or misclassification. Flag for human verification.
• API rate limits: Anthropic rate limits vary by plan. Batch classifications in groups of 10-20 with 1-second delays.