inbound-lead-triage（入境线索分类）

Inbound Lead Triage

Pulls all inbound leads from a specified period, classifies them by source urgency, qualifies against ICP, enriches with company/person context, and produces a prioritized action queue. Turns a messy inbox of form fills into a clear "who to call first" list.

When to Auto-Load

Load this composite when:

• User says "triage my inbound leads", "review new leads", "check demo requests"
• User wants to process a batch of inbound leads from any period
• User asks "who should I follow up with first?"
• A scheduled workflow triggers periodic inbound review

Architecture

[Lead Sources] → Step 1: Collect → Step 2: Classify & Rank → Step 3: Qualify → Step 4: Enrich → Step 5: Route & Respond
                    ↓                    ↓                        ↓                  ↓                    ↓
              Raw lead list      Urgency-ranked list       ICP-scored list    Context-rich list    Action queue + drafts

Each step has a clear input/output contract. Tools are configured once per client, not hardcoded.

Step 0: Configuration (Once Per Client)

On first run, ask the user to configure their lead sources and preferences. Save to clients/<client-name>/config/inbound-triage.json.

{
  "lead_sources": {
    "demo_requests": {
      "source_tool": "HubSpot | Salesforce | Typeform | CSV | other",
      "access_method": "API | CSV export | MCP tool | manual paste",
      "fields_available": ["name", "email", "company", "title", "message"]
    },
    "free_trial_signups": {
      "source_tool": "product database | Stripe | CSV | other",
      "access_method": "API | CSV export | manual paste",
      "fields_available": ["name", "email", "company", "signup_date", "plan"]
    },
    "content_downloads": {
      "source_tool": "HubSpot | Marketo | CSV | other",
      "access_method": "API | CSV export | manual paste",
      "fields_available": ["name", "email", "company", "content_title", "download_date"]
    },
    "webinar_registrations": {
      "source_tool": "Zoom | Luma | CSV | other",
      "access_method": "API | CSV export | manual paste",
      "fields_available": ["name", "email", "company", "webinar_title", "attended"]
    },
    "chatbot_conversations": {
      "source_tool": "Intercom | Drift | Crisp | CSV | other",
      "access_method": "API | CSV export | manual paste",
      "fields_available": ["name", "email", "company", "conversation_summary", "intent"]
    }
  },
  "urgency_overrides": {},
  "response_preferences": {
    "demo_request_sla": "< 1 hour",
    "trial_signup_sla": "< 4 hours",
    "default_sla": "< 24 hours"
  },
  "qualification_prompt_path": "path/to/existing/qualification-prompt.md or null"
}

On subsequent runs: Load config silently, skip setup.

If user provides a raw CSV or pastes leads inline: Skip source config — just classify what's given.

Step 1: Collect Leads

Input

• Time period (e.g., "last 24 hours", "this week", "since Monday")
• Lead sources to check (all configured sources by default, or user specifies)

Process

1. Pull leads from each configured source for the specified period
2. Normalize into a common schema regardless of source:

{
  "name": "",
  "email": "",
  "company": "",
  "title": "",           // if available
  "source_type": "",     // demo_request | free_trial | content_download | webinar | chatbot | other
  "source_detail": "",   // which form, which webinar, which content piece
  "timestamp": "",
  "raw_message": "",     // form message, chat transcript, or null
  "raw_fields": {}       // all original fields preserved
}

1. Deduplicate by email — if same person appears in multiple sources, merge into one record with all source types listed (this is a stronger signal)

Output

• Normalized lead list with source classification
• Dedup report: "X leads from Y sources, Z duplicates merged"

Human Checkpoint

Present the collected leads count by source. Ask: "These are the leads I found. Proceed with triage?"

Step 2: Classify & Rank by Urgency

Urgency Tier System

Rank every lead into one of four urgency tiers based on their source type and behavioral signals.

Tier 1 — Respond NOW (< 1 hour SLA)

• Demo requests (explicit buying intent)
• Chatbot conversations where the prospect asked about pricing, integration, or implementation
• Any lead that mentions a competitor by name
• Any lead that mentions a timeline ("looking to switch by Q2", "evaluating this month")
• Multi-source leads: appeared in 2+ sources (e.g., downloaded content AND requested demo)

Tier 2 — Respond TODAY (< 4 hours SLA)

• Free trial signups (active evaluation intent)
• Chatbot conversations with product questions (not pricing)
• Webinar attendees who actually attended (not just registered)
• Content downloads of bottom-funnel content (case studies, ROI calculators, comparison guides, pricing pages)

Tier 3 — Respond within 24 hours

• Webinar registrants who did NOT attend
• Content downloads of mid-funnel content (how-to guides, industry reports, templates)
• Free trial signups with personal email domains (gmail, yahoo — lower commercial intent)

Tier 4 — Batch response / nurture

• Content downloads of top-funnel content (blog posts, general ebooks, infographics)
• Webinar registrants for broad/educational topics with no other signal
• Any lead where company or title is missing and can't be enriched

Signal Boosters (Move Up One Tier)

• Company matches existing ICP (known good industry, size, stage)
• Person has a title that maps to buyer persona
• Company is already in pipeline (existing deal — warm context)
• Person previously engaged (downloaded other content, attended other webinar)
• Company was flagged by a signal composite (funding, hiring, leadership change)

Signal Dampeners (Move Down One Tier)

• Personal email domain with no company info
• Bot-like behavior (form filled in < 3 seconds, nonsense fields)
• Existing customer (unless upsell signal — keep at tier but flag differently)
• Competitor employee (flag separately — competitive intelligence, not a lead)

Urgency Override

If the user configured custom urgency rules in urgency_overrides, apply those after the default classification.

Output

• Lead list sorted by urgency tier, then by timestamp within each tier
• Each lead tagged with: urgency_tier, urgency_reason, signal_boosters[], signal_dampeners[]

Step 3: Qualify Against ICP

Process

For each lead, run a lightweight ICP qualification check. This is NOT the full lead-qualification capability — it's a fast pass to separate likely-fit from unlikely-fit.

1. If a qualification prompt exists (from lead-qualification capability): Apply it in fast mode — check hard disqualifiers only, skip deep enrichment
2. If no qualification prompt: Use the company context to do a basic fit check:

- Does the company size/industry/stage roughly match ICP? - Does the person's title map to a buyer/user persona? - Any obvious disqualifiers? (wrong geography, wrong industry, too small/large)

Qualification Categories

• Strong fit — Company and person clearly match ICP
• Likely fit — Partial data but what's available looks good
• Unknown fit — Not enough data to determine (needs enrichment)
• Poor fit — Clearly outside ICP (but still log — they came inbound for a reason)
• Flag: Competitor — Employee of a known competitor
• Flag: Existing customer — Already a customer (route to CS/upsell, not sales)

Output

• Each lead tagged with: icp_fit, fit_reasoning (one sentence)
• Leads with "poor fit" or "flag" categories are separated into a secondary list

Step 4: Enrich with Context

Process

For leads in Tier 1 and Tier 2 (and strong/likely fit Tier 3), gather context that helps the responder have a better conversation.

Company enrichment (per unique company):

• What does the company do? (one sentence)
• Company size, stage, industry
• Any recent news or signals? (check if any signal composites have flagged this company)
• Are they in our pipeline already? (check CRM/Supabase)
• Previous engagement history (other people from this company who engaged)

Person enrichment:

• Current role and tenure
• LinkedIn headline (if available via enrichment tools)
• Any prior engagement with us (previous downloads, webinar attendance, email replies)
• Mutual connections or warm intro paths (if detectable)

Engagement context:

• What content did they download? (summarize the topic — the responder should know what the lead is interested in)
• What did they say in the demo request form? (quote key phrases)
• What was the chatbot conversation about? (3-sentence summary)
• Which webinar did they attend? (topic relevance to product)

Tool Flexibility

• Use whatever enrichment tools are configured (Apollo, LinkedIn scraper, web search, Supabase)
• If no enrichment tools are configured, do a basic web search for company + person
• Skip enrichment for Tier 4 leads (not worth the cost — batch-nurture them)

Output

• Each Tier 1-3 lead has an enrichment block with company context, person context, and engagement context
• Enrichment is best-effort — missing fields are noted, never blocks the process

Step 5: Route & Recommend Response

For Each Lead, Determine:

1. Response action:

2. Draft response (for Tier 1-3 leads with personal outreach):

Draft a response email that:

• References the specific action they took ("I saw you requested a demo of...", "Thanks for downloading our guide on...")
• Connects their likely interest to a relevant product capability
• Uses enrichment context naturally (don't be creepy — reference public info only)
• Has a clear CTA appropriate to the tier:

- Tier 1: "Are you free [specific time slots] this week for a quick call?" - Tier 2: "Would it be helpful if I walked you through [relevant feature]?" - Tier 3: "I put together [relevant resource] that builds on what you downloaded. Happy to chat if you have questions."

3. Assign owner (if team structure is configured):

• Route by territory, account size, or round-robin based on client config
• If no team structure: all leads go to the user

Output Format

Primary: Prioritized Action Queue

Present as a clear, actionable table organized by urgency tier:

## Inbound Lead Triage: [Period]
Generated: [timestamp]

### Summary
- **Total leads:** X from Y sources
- **Tier 1 (respond now):** X leads
- **Tier 2 (respond today):** X leads
- **Tier 3 (respond within 24h):** X leads
- **Tier 4 (nurture):** X leads
- **Flagged:** X competitors, X existing customers

---

### 🔴 Tier 1 — Respond NOW

#### Lead: [Name] — [Company] — [Title]
- **Source:** Demo request via [tool] at [time]
- **Urgency reason:** Explicit demo request + company matches ICP
- **ICP fit:** Strong — [one sentence reasoning]
- **Company:** [one sentence what they do, size, stage]
- **Signal context:** [any signals from other composites]
- **Engagement:** "[quote from form message or chat]"
- **Recommended action:** Personal email + calendar link
- **Draft response:**
  > [ready-to-send email draft]

---

#### Lead: [Name] — [Company] — [Title]
...

---

### 🟡 Tier 2 — Respond TODAY
...

### 🟢 Tier 3 — Respond Within 24h
...

### ⚪ Tier 4 — Add to Nurture
[Batch list — name, company, source, content engaged with. No individual drafts.]

### 🔵 Flagged
- **Competitors:** [list with company name — route to competitive intel]
- **Existing customers:** [list with company name — route to CS]

Secondary: CSV Export

Also produce a flat CSV with all leads and their triage data:

• All original fields
• urgency_tier, urgency_reason, icp_fit, fit_reasoning
• recommended_action, response_draft
• company_context, person_context, engagement_context

Save to: clients/<client-name>/leads/inbound-triage-[date].csv

Handling Edge Cases

No leads in period: Report "No new inbound leads found for [period]" with a note on which sources were checked.

Lead with no email: Still triage if company + name are available. Flag as "no direct contact — needs manual lookup."

Lead from a very large company (enterprise): Boost urgency by one tier. Enterprise inbound is rare and high-value. Note: "Enterprise lead — consider executive-level response."

Lead that's already in an active outbound sequence: Flag: "Already in outbound sequence [campaign name] — check for overlap before responding." Don't send a second competing message.

Duplicate lead (same person, same source, same period): Keep the most recent submission. Note the repeat engagement as a signal booster.

Partial data (no company, no title): Classify based on what IS available. Source type still determines urgency tier. Note: "Incomplete data — consider enrichment before outreach."

High volume (100+ leads in period): For Tier 3-4 leads, skip individual enrichment. Batch-qualify company names against ICP, produce summary stats instead of individual briefs. Focus human attention on Tier 1-2 only.

Cadence

• 2-3x daily (for active teams): Triage new leads every few hours to hit SLA targets
• Daily AM (minimum): Morning triage of all leads from previous day/overnight
• Weekly (lightweight): Summary stats on inbound volume, source effectiveness, response times

Tools Required

The agent should have access to:

• CRM / form tool access — to pull inbound leads (HubSpot, Salesforce, Typeform, etc.)
• Web search — for company/person enrichment when other tools aren't available
• Lead enrichment tools — Apollo, LinkedIn scraper, or similar (optional, enhances quality)
• Supabase client — to check existing pipeline, outreach history, signal flags
• Email drafting — references email-drafting capability for response frameworks
• Calendar tool — to include available time slots in Tier 1 responses (gcalcli or similar)