Champion Engagement Scoring

Use Claude (via Anthropic API) to analyze all touchpoints with a champion contact and produce a structured engagement score that predicts whether the champion is actively advocating, passively supportive, or disengaging.

Prerequisites

• Anthropic API key (ANTHROPIC_API_KEY)
• Champion interaction data collected in Attio (notes, emails, meeting transcripts)
• At least 3 interactions logged for the champion (scoring is unreliable with fewer data points)

How It Works

The scoring model evaluates five dimensions of champion engagement:

| Dimension | Weight | What It Measures | |-----------|--------|------------------| | Responsiveness | 25% | How quickly they reply, whether they initiate contact | | Internal Advocacy | 25% | Evidence they've introduced you to colleagues, forwarded materials | | Knowledge Depth | 20% | How well they understand your product's value prop | | Urgency Signals | 15% | Language indicating timeline pressure or budget momentum | | Risk Signals | 15% | Signs of disengagement, competitor mentions, role changes |

Steps

1. Gather Champion Interaction Data

Pull all interaction data for the champion from Attio using the attio-notes fundamental:

GET https://api.attio.com/v2/notes?parent_object=people&parent_record_id={champion_record_id}

Also pull email thread data from Instantly or your email tool's API. Compile into a single interaction timeline:

{
  "champion": {
    "name": "...",
    "title": "...",
    "company": "...",
    "champion_since": "YYYY-MM-DD"
  },
  "interactions": [
    {
      "date": "YYYY-MM-DD",
      "type": "email | meeting | linkedin | call",
      "direction": "inbound | outbound",
      "summary": "...",
      "response_time_hours": null | number,
      "sentiment": "positive | neutral | negative"
    }
  ]
}

2. Send to Claude for Scoring

Using the anthropic-api-patterns fundamental, send the interaction data to Claude:

POST https://api.anthropic.com/v1/messages

Headers:
  x-api-key: {ANTHROPIC_API_KEY}
  anthropic-version: 2023-06-01
  Content-Type: application/json

Body:
{
  "model": "claude-sonnet-4-20250514",
  "max_tokens": 1024,
  "messages": [{
    "role": "user",
    "content": "You are a sales intelligence analyst scoring champion engagement. Analyze the following champion interaction history and return a structured JSON score.\n\nChampion data:\n{interaction_timeline_json}\n\nScore each dimension 0-100:\n1. Responsiveness: How quickly do they reply? Do they ever initiate contact? Score 80+ if average response <24h and they've initiated at least once.\n2. Internal Advocacy: Evidence they've introduced colleagues, forwarded materials, scheduled internal meetings. Score 80+ if they've made 2+ introductions.\n3. Knowledge Depth: Do they reference specific features, use cases, or ROI metrics? Score 80+ if they articulate value in their own words.\n4. Urgency Signals: Language about timelines, budget cycles, upcoming decisions. Score 80+ if they mention a specific decision date.\n5. Risk Signals: INVERSE score — 100 means NO risk. Deduct for: competitor mentions (-30), longer response times over time (-20), cancelled meetings (-25), role change (-15).\n\nReturn ONLY valid JSON:\n{\n  \"responsiveness\": {\"score\": N, \"evidence\": \"...\"},\n  \"internal_advocacy\": {\"score\": N, \"evidence\": \"...\"},\n  \"knowledge_depth\": {\"score\": N, \"evidence\": \"...\"},\n  \"urgency_signals\": {\"score\": N, \"evidence\": \"...\"},\n  \"risk_signals\": {\"score\": N, \"evidence\": \"...\"},\n  \"composite_score\": N,\n  \"health\": \"strong\" | \"healthy\" | \"at_risk\" | \"disengaged\",\n  \"recommended_action\": \"...\"\n}"
  }]
}

3. Parse and Store the Score

Parse Claude's JSON response. Compute composite score using weights:

composite = (responsiveness * 0.25) + (internal_advocacy * 0.25) + (knowledge_depth * 0.20) + (urgency_signals * 0.15) + (risk_signals * 0.15)

Map composite to health status:

• 75-100: Strong
• 50-74: Healthy
• 25-49: At Risk
• 0-24: Disengaged

Write the score back to Attio using attio-custom-attributes:

PATCH https://api.attio.com/v2/objects/people/records/{champion_record_id}
{
  "data": {
    "values": {
      "champion_score": [{"value": composite_score}],
      "champion_last_engaged": [{"value": "latest_interaction_date"}]
    }
  }
}

4. Handle Edge Cases

• Fewer than 3 interactions: Return score with a confidence: "low" flag. Do not auto-update champion status based on low-confidence scores.
• All interactions are outbound: This is a risk signal. The champion is not initiating. Add -20 to responsiveness score.
• Champion changed roles: If the role change is within the same company, re-evaluate. If they left the company, set status to "Lost" immediately.

Error Handling

• 429 RATE_LIMITED from Anthropic: Back off exponentially. Start at 5s, double each retry, max 3 retries.
• 500 from Anthropic: Retry once after 10s. If it fails again, skip this champion and flag for manual review.
• Invalid JSON from Claude: Re-send the request with a stricter prompt. If it fails twice, extract scores manually from the text response using regex.

Alternatives

| Tool | Method | Notes | |------|--------|-------| | Anthropic Claude | Messages API | Best for nuanced interaction analysis | | OpenAI GPT-4o | Chat Completions API | Alternative LLM, similar capability | | Gong | Deal Intelligence API | Pre-built engagement scoring from call data | | Clari | Revenue Intelligence API | Enterprise engagement scoring | | Custom ML | Scikit-learn / PyTorch | Roll your own model if you have labeled training data |