tam-builder

name	tam-builder
description	Build and maintain a scored Total Addressable Market (TAM) using Apollo Company Search. Upserts companies to Supabase, scores ICP fit, assigns tiers, and auto-builds a persona watchlist for Tier 1-2 companies using Apollo People Search (free). Supports build, refresh, and status modes. Designed for CLI now, automation-ready for trigger.dev later.
tags	["lead-generation"]

TAM Builder

Build and maintain a scored Total Addressable Market. Uses Apollo Company Search to discover companies, upserts them to Supabase, scores ICP fit (0-100), assigns tiers (1/2/3), and auto-builds a persona watchlist for Tier 1-2 companies using Apollo People Search (free).

Three modes:

• build — First-time TAM construction from Apollo search
• refresh — Update existing TAM: re-score, detect tier changes, deprecate stale companies
• status — Read-only report of current TAM state

Prerequisites

1. Apollo API Key

Add to .env:

APOLLO_API_KEY=your-api-key-here

2. Supabase Project

Same project used by other skills. Schema in tools/supabase/schema.sql. This skill writes to companies and people tables.

3. Verify Environment

Ensure .env has:

APOLLO_API_KEY=...
SUPABASE_URL=https://xxx.supabase.co
SUPABASE_SERVICE_ROLE_KEY=eyJ...

Config Format

Create a JSON config per client/segment:

{
  "client_name": "happy-robot",
  "tam_config_name": "voice-ai-midmarket",

  "company_filters": {
    "organization_num_employees_ranges": ["51,200", "201,500", "501,1000"],
    "q_organization_keyword_tags": ["call center", "contact center"],
    "organization_locations": ["United States"]
  },

  "scoring": {
    "weights": {
      "employee_count_fit": 30,
      "industry_fit": 25,
      "funding_stage_fit": 20,
      "geo_fit": 
       
    
          
      
          
       
     
  

   
     
     
     
        
         
    
      
  

   
   

Save to: skills/capabilities/tam-builder/configs/{client-name}.json

Database Write Policy

CRITICAL: Never write to Supabase without explicit user approval.

The TAM builder touches the companies and people tables. Unwanted writes are hard to clean up — they pollute the database, create duplicates on re-runs, and contaminate existing good data.

Required flow before any database write:

1. Run --preview to show the total TAM universe count and cost implications
2. Run --sample (with --test for 1 page) to show scored results in-memory — NO database writes
3. Present sample results to user: tier distribution, example Tier 1/2 companies, scoring sanity check
4. Get explicit user approval before proceeding to a build that writes to Supabase
5. Only then run without --sample to execute the actual build

Why this matters:

• If sample results look wrong, the user adjusts filters and re-runs — no cleanup needed
• Multiple test iterations don't create orphaned rows in the database
• Existing good data from prior builds is never accidentally overwritten
• The user stays in control of what enters the database

The agent must NEVER pass --yes on a first run. The --yes flag is only for automated/scheduled refreshes where the user has pre-approved the operation.

Usage

Sample (review before committing — ALWAYS DO THIS FIRST)

python3 skills/capabilities/tam-builder/scripts/tam_builder.py \
  --config skills/capabilities/tam-builder/configs/{client}.json \
  --mode build --sample --test

This searches Apollo (1 page, ~100 companies), scores them in-memory, and prints tier distribution + top companies. No database writes. Review output with user before proceeding.

Build (first-time — only after user approves sample)

python3 skills/capabilities/tam-builder/scripts/tam_builder.py \
  --config skills/capabilities/tam-builder/configs/{client}.json \
  --mode build [--test] [--yes] [--dry-run] [--skip-watchlist]

Refresh (update existing)

python3 skills/capabilities/tam-builder/scripts/tam_builder.py \
  --config skills/capabilities/tam-builder/configs/{client}.json \
  --mode refresh [--test] [--yes]

Status (read-only report)

python3 skills/capabilities/tam-builder/scripts/tam_builder.py \
  --config skills/capabilities/tam-builder/configs/{client}.json \
  --mode status

Pipeline: Build Mode

Step 0: --preview → total count + cost estimate (no DB writes)
Step 1: --sample --test → search 1 page, score in-memory, show results (no DB writes)
Step 2: User reviews sample → approves, adjusts filters, or caps scope
Step 3: Full build → Apollo Company Search → Upsert to Supabase → Score → Tier → Watchlist

Phase details (Step 3 only — after user approval):

Phase 1: Apollo Company Search → Upsert raw companies → Score ICP fit → Assign tiers
Phase 2: (skipped in build mode — no prior data to deprecate)
Phase 3: Persona Watchlist — pull 2-3 personas per Tier 1-2 company (free)

Pipeline: Refresh Mode

Phase 1: Apollo Company Search → Upsert/update companies → Re-score → Detect tier changes
Phase 2: Deprecation — companies missing 2+ consecutive refreshes get deprecated
Phase 3: Persona Watchlist — pull personas for new/promoted Tier 1-2 companies,
         disqualify personas at deprecated companies

ICP Scoring (0-100)

Pure function, no API calls. Weighted scoring across 5 dimensions from config:

• employee_count_fit — headcount in target ranges?
• industry_fit — industry matches targets?
• funding_stage_fit — funding stage in targets?
• geo_fit — HQ location in target geos?
• keyword_match — org keywords overlap config keywords?

Score thresholds (configurable): >=75 = Tier 1, >=50 = Tier 2, else Tier 3.

Deprecation Rules (refresh only)

• First miss (not returned by search): metadata.refresh_miss_count = 1, keep active
• Second consecutive miss: tam_status = 'deprecated'
• Employee count drops to 0: immediate deprecation
• Companies with tam_status = 'converted' are always exempt

Watchlist — Persona Sync

Scenario	Behavior
New Tier 1-2 company	Pull 2-3 personas immediately
Company promoted Tier 3→2	Pull personas during refresh
Company deprecated	Disqualify monitoring personas
Company demoted Tier 1→3	Keep existing personas, stop refreshing

Mode Caps

Parameter	Test	Standard	Full
Max pages	1	50	200
Max companies	100	5,000	20,000

Flags

Flag	Effect
--test	Limit to 1 page / 100 companies
--sample	Search + score in-memory, NO database writes. Shows tier distribution and top companies for user review. Combine with --test for a quick 1-page sample.
--yes	Skip confirmation prompts. Only use for pre-approved automated refreshes, never on first runs.
--dry-run	No API calls, show what would happen
--preview	Show total count only (build mode)
--skip-watchlist	Skip persona watchlist phase