outbound-lead-qualification

Outbound Lead Qualification

Takes any list of outbound prospects and qualifies each against your ICP. Unlike inbound-lead-qualification (which has intent signals from the lead coming to you), this skill works with cold lists where you have no intent data — qualification is purely on fit.

When to Auto-Load

Load this composite when:

User says "qualify these prospects", "score this lead list", "which of these are worth reaching out to"
User provides a CSV, Google Sheet, or list of LinkedIn URLs for outbound
An upstream signal composite (funding, hiring, news) produced a prospect list that needs qualification
User exported leads from Apollo, Clay, LinkedIn Sales Nav, or similar tools

Do NOT load when:

Leads are inbound (demo requests, signups, content downloads) — use inbound-lead-qualification instead
User wants triage/prioritization of inbound — use inbound-lead-triage

Architecture

[Prospect List] → Step 1: Load ICP & Parse → Step 2: Enrich Sparse Records → Step 3: Company Qualification → Step 4: Person Qualification → Step 5: Reachability & Signal Check → Step 6: Score & Tier → Step 7: Output CSV

Step 0: Configuration (Once Per Client)

On first run, establish the ICP definition and tool preferences. Save to clients/<client-name>/config/outbound-lead-qualification.json.

{
  "icp_definition": {
    "company_size": {
      "min_employees": null,
      "max_employees": null,
      "sweet_spot": "",
      "notes": ""
    },
    "industry": {
      "target_industries": [],
      "excluded_industries": [],
      "notes": ""
    },
    "use_case": {
      "primary_use_cases": [],
      "secondary_use_cases": [],
      "anti_use_cases": [],
      "notes": ""
    },
    "company_stage": {
      "target_stages": [],
      "excluded_stages": [],
      "notes": ""
    },
    "geography": {
      "target_regions": [],
      "excluded_regions": [],
      "notes": ""
    }
  },
  "buyer_personas": [
    {
      "name": "",
      "titles": [],
      "seniority_levels": [],
      "departments": [],
      "is_economic_buyer": false,
      "is_champion": false,
      "is_user": false
    }
  ],
  "hard_disqualifiers": [],
  "hard_qualifiers": [],
  "list_source_context": {
    "source": "Apollo | Clay | LinkedIn Sales Nav | event | scraped | manual | signal-composite | other",
    "source_detail": "",
    "original_targeting_criteria": ""
  },
  "crm_access": {
    "tool": "Supabase | HubSpot | Salesforce | CSV export | none",
    "access_method": "",
    "tables_or_objects": []
  },
  "qualification_prompt_path": "path/to/lead-qualification/prompt.md or null"
}

If lead-qualification capability already has a saved qualification prompt: Reference it directly — don't rebuild ICP criteria from scratch.

On subsequent runs: Load config silently.

Step 1: Load ICP Criteria & Parse Leads

Process

Load the client's ICP config (or qualification prompt from lead-qualification capability)
Parse the prospect list — accept any format:
- CSV file with any column structure
- Google Sheet URL (read via Rube MCP)
- LinkedIn profile URLs — one or more inline
- Apollo/Clay/Sales Nav export
- Pasted list of names/companies/emails
- Output from an upstream signal composite
Detect the list source and note it — source context affects scoring nuance:
- Signal-sourced lists (funding, hiring, news): Higher baseline relevance — these were already filtered by a signal
- Tool exports (Apollo, Clay): Usually pre-filtered by some criteria — check what
- Event attendees: Have implicit topic interest
- Purchased/scraped lists: Lowest baseline quality — expect higher disqualification rates
Inventory available data per lead:
- Have: Fields present in the input
- Need: Fields required for qualification but missing
- Gap report: "X leads have LinkedIn URL, Y have company + title, Z have only name/email"

Output

Parsed lead list with field inventory
Gap report for the user
Source classification

Human Checkpoint

If >60% of leads are missing both company name AND title AND LinkedIn URL, warn: "This list is very sparse. I can try enriching from emails/names, but expect lower accuracy. Proceed or provide a richer list?"

Step 2: Enrich Sparse Records

When to Run

Any lead missing company, title, or industry
Leads with LinkedIn URLs but no structured profile data
Skip for leads that already have full company + title + industry data

Process

Path A — LinkedIn URL available: Run the batch enrichment script from the lead-qualification capability:

python3 skills/lead-qualification/scripts/enrich_leads.py INPUT_CSV \
  --output ENRICHED_CSV \
  --cache-hours 24

Use --dry-run first to show cost estimate. Cost: ~$3 per 1,000 profiles.

Path B — No LinkedIn URL, but have name + company:

Web search for "[Name] [Company] LinkedIn" to find the profile URL
Then run Path A enrichment
Batch these lookups — don't search one at a time

Path C — Only email available:

Extract domain from email
If corporate domain: look up company, search for the person by name + company
If personal email (gmail, yahoo, etc.): flag as enrichment_limited — qualify on whatever data is available

Path D — Name only:

Too ambiguous without additional context. Flag as insufficient_data.
Recommend the user provide more identifying information.

Output

Enriched CSV with: enriched_title, enriched_company, enriched_industry, enriched_location, enriched_connections, enriched_education, enriched_experience_years, enriched_headline, enriched_about, enrichment_status
Enrichment summary: "Enriched X/Y leads. Z failed. W had cached data."

Step 3: Company Qualification

Process

For each lead's company, evaluate against every ICP company dimension:

Dimension 1 — Company Size

Check employee count against ICP range
Sources: enrichment data, LinkedIn company page, web search
Score: match | borderline | mismatch | unknown
Note: For subsidiaries/divisions, evaluate the relevant unit

Dimension 2 — Industry

Check against target and excluded industry lists
Be smart about classification: "AI-powered HR platform" matches both "AI/ML" and "HR Tech"
Score: match | adjacent (related but not core target) | mismatch | unknown

Dimension 3 — Company Stage

Seed, Series A, Series B+, Growth, Public, Bootstrapped
Sources: Crunchbase, news, enrichment data
Score: match | borderline | mismatch | unknown

Dimension 4 — Geography

Check HQ location and/or the specific person's location
For remote-first companies, check where the majority of the team is
Score: match | borderline | mismatch | unknown

Dimension 5 — Use Case Fit

Based on what the company does, could they plausibly use the product?
For outbound, this is the most important dimension — there's no intent signal, so fit must be strong
Sources: company website, product description, job postings, tech stack signals
Score: strong_fit | moderate_fit | weak_fit | no_fit | unknown

Output

Each lead gets a company_qualification block:

{
  "company_size": { "score": "", "value": "", "reasoning": "" },
  "industry": { "score": "", "value": "", "reasoning": "" },
  "stage": { "score": "", "value": "", "reasoning": "" },
  "geography": { "score": "", "value": "", "reasoning": "" },
  "use_case": { "score": "", "value": "", "reasoning": "" },
  "company_verdict": "qualified | borderline | disqualified | insufficient_data"
}

Step 4: Person Qualification

Process

For each lead's contact person, evaluate against buyer persona criteria:

Dimension 1 — Title/Role Match

Check title against buyer persona title lists
Handle variations: "VP of Marketing" = "Vice President, Marketing" = "VP Marketing"
Adjust for company size: a "Director" at a 10-person startup ≠ "Director" at a 10,000-person enterprise
Score: exact_match | close_match | adjacent | mismatch | unknown

Dimension 2 — Seniority Level

Map to: Individual Contributor, Manager, Director, VP, C-Level, Founder
Check against ICP seniority requirements
For outbound, seniority matters more — you're cold-reaching, so you need someone with authority to respond
Score: match | too_junior | too_senior | unknown

Dimension 3 — Department

Engineering, Product, Marketing, Sales, Operations, Finance, HR, etc.
Check against ICP department targets
Score: match | adjacent | mismatch | unknown

Dimension 4 — Authority Type

Based on title + seniority + company size, classify:
- economic_buyer — Can sign the check
- champion — Wants it, can influence the decision
- user — Would use it daily, can validate need
- evaluator — Tasked with research, limited decision power
- gatekeeper — Can block but not approve
- unknown

Dimension 5 — Reachability Assessment

Unlike inbound (where the lead reached out), outbound requires you to reach them
Check: Do we have a work email? LinkedIn URL? Phone?
Score: highly_reachable (work email + LinkedIn) | reachable (one channel) | low_reachability (personal email only or no contact info) | unreachable

Output

Each lead gets a person_qualification block:

{
  "title_match": { "score": "", "value": "", "reasoning": "" },
  "seniority": { "score": "", "value": "", "reasoning": "" },
  "department": { "score": "", "value": "", "reasoning": "" },
  "authority_type": "",
  "reachability": { "score": "", "channels": [], "reasoning": "" },
  "person_verdict": "qualified | borderline | disqualified | insufficient_data",
  "mismatch_type": "null | right_company_wrong_person | right_person_wrong_company"
}

Step 5: Reachability & Signal Check

Process

Signal Overlap Check: For each lead, check if any signal composites have already flagged this company:

Search Supabase signals table (if available) for the company
Check if the company appeared in recent funding/hiring/news signal runs
If match found: Flag as signal_boosted with signal type and date

CRM/Pipeline Check:

Search CRM for the company in active deals
If match found: Flag as in_pipeline — the outbound might conflict with an existing deal
Search customer database: Flag as existing_customer for expansion plays
Search outreach logs: Flag as previously_contacted with outcome

Timing Signals:

Recent funding round → Good timing for budget conversations
Recent leadership change → Potential openness to new vendors
Recent hiring spree in relevant department → Growing team = growing need
Recent layoffs → Bad timing, budget likely frozen
Recent competitor mention → Awareness of the problem space

Output

Each lead gets:

{
  "signal_flags": [],
  "pipeline_status": "new | existing_customer | in_pipeline | previously_contacted",
  "pipeline_detail": "",
  "timing_signals": [],
  "timing_verdict": "good_timing | neutral | bad_timing"
}

Step 6: Score & Tier

Scoring Logic

Combine all dimensions into a final qualification verdict. Weights are shifted compared to inbound because there's no intent signal — fit must carry the entire decision.

Composite Score Calculation:

Dimension	Weight	Possible Values
Use Case Fit	30%	strong=100, moderate=60, weak=20, no_fit=0, unknown=30
Industry	20%	match=100, adjacent=60, mismatch=0, unknown=30
Person Title/Role	15%	exact=100, close=75, adjacent=40, mismatch=0, unknown=30
Company Size	10%	match=100, borderline=50, mismatch=0, unknown=30
Person Seniority	10%	match=100, too_junior=20, too_senior=60, unknown=30
Company Stage	5%	match=100, borderline=50, mismatch=0, unknown=30
Geography	5%	match=100, borderline=50, mismatch=0, unknown=30
Reachability	5%	highly_reachable=100, reachable=60, low=20, unreachable=0

Hard overrides (bypass the score):

Any hard disqualifier present → disqualified regardless of score
Any hard qualifier present → qualified regardless of score (but still show full breakdown)
Existing customer → Route separately as expansion opportunity
In active pipeline → Flag for sales coordination, don't disqualify

Signal boosters (add to final score):

Signal-sourced lead (from funding/hiring/news composite): +5 points
Positive timing signal: +5 points
Multiple contact channels available: +3 points

Signal dampeners (subtract from final score):

Bad timing signal (layoffs, budget freeze): -10 points
Previously contacted, no response: -5 points
Previously contacted, rejected: -10 points (but don't auto-disqualify — circumstances change)

Verdict thresholds (stricter than inbound — no intent signal):

Score >= 80: tier_1 — Strong fit, prioritize outreach
Score 65-79: tier_2 — Good fit, include in sequences
Score 50-64: tier_3 — Moderate fit, lower priority or batch outreach
Score 35-49: borderline — Marginal fit, consider only if pipeline is thin
Score < 35: disqualified — Does not fit ICP, do not reach out

Sub-verdicts for routing:

tier_1_signal_boosted — Tier 1 AND has timing/signal boost → Outreach immediately
tier_1_standard — Tier 1, no special signals → High-priority sequence
tier_2_standard — Good fit → Standard sequence
tier_3_batch — Moderate fit → Batch/automated sequence
borderline_hold — Marginal → Hold for now, revisit if pipeline is thin
disqualified_wrong_company — Company doesn't fit
disqualified_wrong_person — Company fits but person doesn't → find the right person
disqualified_unreachable — Fits but no way to contact
existing_customer_expansion — Already a customer → Route to CS/AM
pipeline_conflict — Already in active deal → Coordinate with deal owner

Output

Each lead gets:

{
  "composite_score": 0,
  "verdict": "",
  "sub_verdict": "",
  "outreach_priority": 1,
  "top_qualification_reasons": [],
  "top_disqualification_reasons": [],
  "summary": "One sentence: why this lead is/isn't a fit"
}

Step 7: Output CSV

Parallelized Qualification

Before producing the output, the actual qualification (Steps 3-6) MUST be parallelized for lists over 10 leads.

Parallelization protocol:

Run a calibration batch of 5-10 leads first, present to the user for approval
Once approved, split remaining leads into batches of ~15
Launch ALL batches simultaneously using parallel Task agents
Merge results preserving original row order
Validate completeness: qualified + disqualified + failed = total

CSV Structure

Produce a CSV with ALL input fields preserved plus qualification columns appended:

Core qualification columns:

qualification_verdict — tier_1 | tier_2 | tier_3 | borderline | disqualified
qualification_sub_verdict — tier_1_signal_boosted | tier_1_standard | tier_2_standard | tier_3_batch | borderline_hold | disqualified_wrong_company | disqualified_wrong_person | disqualified_unreachable | existing_customer_expansion | pipeline_conflict
composite_score — 0-100
outreach_priority — 1 (highest) to 5 (lowest)
summary — One sentence qualification reasoning

Enrichment columns (if enrichment was run):

enriched_title, enriched_company, enriched_industry, enriched_location
enriched_headline, enriched_experience_years
enrichment_status — success | cached | failed | skipped

Company qualification columns:

company_size_score — match | borderline | mismatch | unknown
industry_score — match | adjacent | mismatch | unknown
stage_score — match | borderline | mismatch | unknown
geography_score — match | borderline | mismatch | unknown
use_case_score — strong | moderate | weak | no_fit | unknown

Person qualification columns:

title_match_score — exact_match | close_match | adjacent | mismatch | unknown
seniority_score — match | too_junior | too_senior | unknown
authority_type — economic_buyer | champion | user | evaluator | gatekeeper | unknown
reachability_score — highly_reachable | reachable | low_reachability | unreachable
mismatch_type — null | right_company_wrong_person | right_person_wrong_company

Signal & pipeline columns:

pipeline_status — new | existing_customer | in_pipeline | previously_contacted
pipeline_detail — One sentence on the overlap
signal_flags — Any signal composite matches
timing_verdict — good_timing | neutral | bad_timing

Primary Output: Google Sheets via Rube MCP

Use RUBE_SEARCH_TOOLS to find Google Sheets tools, then:

Create a new sheet: [Campaign Name] - Qualified Prospects - [Date]
Write all columns with formatting:
- Bold header row with dark background
- Color-code verdict: green (tier 1-2), yellow (tier 3), orange (borderline), red (disqualified)
- Color-code outreach priority: 1=green, 2=blue, 3=yellow, 4=orange, 5=red
- Add filter row for sorting by verdict/priority
Create a "Summary" tab with the stats from below

Fallback: CSV

If Rube MCP is unavailable:

Write CSV to clients/<client-name>/leads/outbound-qualified-[date].csv
Tell the user the file path

Summary Report

After producing the output, present:

## Outbound Qualification Results: [Campaign Name]

**List source:** [source type and detail]
**Total prospects processed:** X

### Qualification Breakdown
| Tier | Count | % | Action |
|------|-------|---|--------|
| Tier 1 (strong fit) | X | Y% | Priority outreach |
| Tier 2 (good fit) | X | Y% | Standard sequences |
| Tier 3 (moderate fit) | X | Y% | Batch outreach |
| Borderline | X | Y% | Hold / revisit |
| Disqualified | X | Y% | Do not contact |

### Signal & Pipeline Flags
- Signal-boosted leads: X
- Existing customers (expansion): X (route to CS)
- Already in pipeline: X (coordinate with deal owner)
- Previously contacted: X

### Top Qualification Reasons
1. [reason] — X leads
2. [reason] — X leads

### Top Disqualification Reasons
1. [reason] — X leads
2. [reason] — X leads

### Data Quality
- Fully enriched: X leads
- Partially enriched: X leads
- Enrichment failed: X leads
- Insufficient data: X leads

**Output:** [Google Sheet link or CSV path]

Handling Edge Cases

Large lists (500+ leads):

Run enrichment in batches of 50 to avoid Apify rate limits
Parallelize qualification aggressively — use 10 concurrent batches
Present progress updates every 100 leads

Duplicate companies across leads:

Qualify the company once, apply to all leads from that company
Person qualification runs individually
Flag multi-threading opportunity: "4 people from [Company] in this list — consider account-based approach"

Leads from signal composites:

These were already filtered by a signal (funding, hiring, news) — they have a baseline relevance
Don't re-check the signal; trust it. Focus qualification on ICP fit
Apply the +5 signal boost to the score

Previously disqualified leads appearing again:

If a lead was disqualified in a prior run but shows up in a new list, re-qualify
Circumstances change: new role, company grew, new funding
Note in reasoning: "Previously disqualified on [date] for [reason]. Re-evaluated because [what changed]."

Right company, wrong person:

Don't disqualify the company. Mark sub-verdict as disqualified_wrong_person
Recommend: "Company [X] is a strong fit. Consider finding [target title] there instead."

Leads with only personal email (gmail, yahoo):

Low reachability for outbound — personal emails are bad for cold outreach
Still qualify on fit. If fit is strong, recommend finding work email via Apollo or similar
Score reachability as low_reachability

Conflicting data between sources:

Prefer enriched LinkedIn data over self-reported/scraped data
Prefer recent data over stale data
Note conflicts in reasoning: "LinkedIn says VP of Engineering, input says CTO — using LinkedIn (more recent)"

Tools Required

Apify LinkedIn Enrichment — skills/lead-qualification/scripts/enrich_leads.py for batch profile enrichment
- Requires APIFY_API_TOKEN environment variable
- Run with --dry-run first to preview cost
Web Search — to research companies and fill gaps when enrichment is sparse
Fetch (web page) — to pull company pages, LinkedIn profiles
Rube MCP — for Google Sheets input/output
- RUBE_SEARCH_TOOLS — discover available tools
- RUBE_MANAGE_CONNECTIONS — ensure Google Sheets connection is active
- RUBE_REMOTE_WORKBENCH — execute sheet operations
CRM access — to check pipeline, existing customers, outreach history
Supabase client — for pipeline/signal lookups
Task tool — to parallelize lead processing across subagents (mandatory for Step 7)
Read/Write — for CSV I/O and config management
Glob/Grep — to find existing qualification prompt files

Example Usage

Qualify an Apollo export:

Qualify these prospects from Apollo: [CSV path or Google Sheet URL]

→ Agent loads ICP, enriches sparse records, qualifies, outputs tiered CSV.

Qualify leads from a signal composite:

Take the funding signal leads from last week's run and qualify them against our ICP.

→ Agent loads signal output, applies signal boost, qualifies, tiers for outreach.

Qualify event attendees:

Here's the attendee list from the AI conference: [CSV]. Qualify them for outreach.

→ Agent notes event source, enriches, qualifies with event-topic context.

Reuse existing qualification prompt:

Qualify these leads using @skills/lead-qualification/qualification-prompts/series-a-saas.md
— list: [Google Sheet URL]

→ Skips ICP setup, goes straight to enrichment + qualification.

Re-qualify with updated criteria:

Lower the company size minimum to 20 employees and add "developer tools" as a target industry.
Re-qualify the borderline leads.

→ Updates config, re-runs only borderline leads, re-tiers.

About this skill

When to Auto-Load

Architecture

Step 0: Configuration (Once Per Client)

Step 1: Load ICP Criteria & Parse Leads

Process

Output

Human Checkpoint

Step 2: Enrich Sparse Records

When to Run

Process

Output

Step 3: Company Qualification

Process

Output

Step 4: Person Qualification

Process

Output

Step 5: Reachability & Signal Check

Process

Output

Step 6: Score & Tier

Scoring Logic

Output

Step 7: Output CSV

Parallelized Qualification

CSV Structure

Primary Output: Google Sheets via Rube MCP

Fallback: CSV

Summary Report

Handling Edge Cases

Tools Required

Example Usage

Qualify an Apollo export:

Qualify leads from a signal composite:

Qualify event attendees:

Reuse existing qualification prompt:

Re-qualify with updated criteria: