Lead Intel

Input: scored_leads.json (pass verdicts only) Batch: process all passing leads at once

Mid-Market

For each A and B tier lead, generate:

• pain_signal — detailed pain hypothesis (statement, not question)
• lead_intel — 2–3 sentence personalized insight connecting their situation to the solution
• email_subject — suggested subject line (<60 chars, no clickbait)
• email_opening — first 1–2 sentences of a personalized cold email

Input: enriched_leads.json if available, otherwise scored_leads.json Batch: A-tier first, then B-tier

Enterprise

For each A and B tier lead, everything from Mid-Market plus:

• trigger_detail — specific trigger event driving outreach timing (from account research)
• email_cta — tailored call-to-action based on authority level and likely priorities

Input: enriched_leads.json (should be available for A/B leads) Batch: process individually for maximum personalization

Data Operations

Use jq for filtering and merging, pandas for batch intel generation:

# Filter qualified leads (SMB)
jq '[.[] | select(.score.verdict == "pass")]' "$CAMPAIGN_DIR/scored_leads.json"

# Filter qualified leads (Mid/Enterprise)
jq '[.[] | select(.score.verdict == "A" or .score.verdict == "B")]' "$CAMPAIGN_DIR/enriched_leads.json"

# Merge intel fields into lead records
jq --argjson intel "$INTEL_ARRAY" '
  [range(length)] | map(. as $i | input[$i] + $intel[$i])
' "$CAMPAIGN_DIR/scored_leads.json" > "$CAMPAIGN_DIR/intel_leads.json"

import pandas as pd, json

# Load leads into DataFrame for batch processing
with open(f"{campaign_dir}/scored_leads.json") as f:
    df = pd.DataFrame(json.load(f))

# Filter by verdict
qualified = df[df["verdict"].isin(["A", "B"])].copy()

# Add intel columns (populated by the LLM per-lead)
qualified["pain_signal"] = ""
qualified["lead_intel"] = ""
qualified["email_subject"] = ""
qualified["email_opening"] = ""

# Save
qualified.to_json(f"{campaign_dir}/intel_leads.json", orient="records", indent=2)

Process

1. Load inputs. Use jq to read scored/enriched leads, business context, ICP, and market KB.

2. Filter leads. Use jq to select only leads meeting the mode's criteria:

   • SMB: jq '[.[] | select(.score.verdict == "pass")]'
   • Mid / Enterprise: jq '[.[] | select(.score.verdict == "A" or .score.verdict == "B")]'

3. Generate intel. For each qualified lead, use the business context and any available enrichment data to generate personalized outreach content.

4. Apply templates. Reference reference/intel_templates.md for tone, format, and quality guidelines.

5. Save output. Use jq or pandas to write intel_leads.json — each entry is the lead record merged with intel fields.

6. Report summary. Use jq or pandas to count leads with intel generated and display sample of best pieces.

7. Update pipeline state. Use jq to mark step 6 complete:

   jq '.completed_steps += [6] | .current_step = 6' \
     "$CAMPAIGN_DIR/pipeline_state.json" > tmp.json && mv tmp.json "$CAMPAIGN_DIR/pipeline_state.json"
   

8. Hand off to quality-gate.

Intel Templates

See reference/intel_templates.md for guidelines on tone, format, and quality.