> ## Documentation Index > Fetch the complete documentation index at: https://docs.anysite.io/llms.txt > Use this file to discover all available pages before exploring further. # B2B Outbound Operator — ICP Discovery & Scoring at $1 per 1k Records > How a founder-led B2B team discovers 14,000 companies, scores them, and narrows to 150 hand-picked outbound targets per sprint — on a single db/linkedin/search/companies endpoint. ## Overview A founder-led B2B SaaS operator (under 10 employees, founded 2024) runs a boutique outbound program — high-quality, manually reviewed touches, \~150 per week — against a custom ICP framework. Instead of buying a sales-tech stack across firmographics, list-building, enrichment, and delivery, the team built a single pipeline on `db/linkedin/search/companies` plus four supporting endpoints. The database-backed search replaces a multi-vendor stack with one structured payload and a mini-DSL for keyword matching. Output is raw rows the team scores on their side — no inherited vendor schema, no per-record fees on data discovery. Discovered in a single sprint from one query (\~2s wall-clock) Hook + verified email + no role drift, per week Out-of-pocket LLM cost; discovery API calls free under MCP Unlimited ## The Challenge Boutique B2B outbound demands a tight ICP: not "B2B tech" but a specific cross-section of company type, size, geography, funding stage, and product positioning. Off-the-shelf sales-tech stacks force a coarse ICP — industry codes, employee ranges, hard categories. To narrow further, teams stitch four to six vendors with overlapping pricing and partial coverage, then export and re-classify on their own side anyway. The team needed: * **Bulk discovery in one call** — pull 1,000–10,000 candidates per sprint, not 50-result pagination loops * **Soft-signal search** — match phrases across `description`, `short_description`, `specialities`, `hashtags`, and `locations` simultaneously * **Strict structured filters** — country (ISO2), employee range, founded year, active-status flags — applied at the database layer, not client-side * **Sub-second iteration** — try ten different keyword sets in an hour, not three over a week * **Schema ownership** — every row returned in full LinkedIn shape so the team writes its own scoring rubric ## The Solution The `/db/linkedin/search/companies` endpoint exposes the database directly via a structured payload. Keywords get a mini-DSL (whitespace = AND, `|` = OR, `"phrase"` = exact phrase, `-token` = NOT). Filters are typed: `employee_count_min/max`, `founded_on_min/max`, `country_hq`, `country_any`, `is_active`, `has_description`, `min_description_length`, and more. ### 1. Cross-field keyword search via mini-DSL — 70% of calls A single `keywords` field searches across `name`, `description`, `short_description`, `specialities`, `hashtags`, and `locations` simultaneously. For 27 ICP categories ranging from autonomous outbound agents to compliance research tools, the team builds one DSL string per sprint: ```json theme={null} { "keywords": "\"AI SDR\"|\"AI BDR\"|\"autonomous outbound\"|\"AI sales agent\"|\"AI research agent\"|\"AI signal\"|\"AI inbound qualification\"|\"AI prospecting\"", "country_hq": ["US","GB","DE","NL","FR","CA","AU","IE","SG"], "employee_count_min": 11, "employee_count_max": 200, "is_active": true, "has_description": true, "min_description_length": 100, "count": 1000 } ``` This single call returns 1,000 ICP-candidate companies with full LinkedIn fields in \~2 seconds wall-clock. ### 2. Token-aware structured geo + size filters — 20% of calls A common pitfall in coarse country filters is substring matching: the `US` token used to false-match against `Austria`, `Australia`, `Austin`, `Houston`. The `country_hq` and `country_any` params parse the ISO2 code directly out of `headquarter_location` and the `locations[]` array, eliminating that class of false positive. ```json theme={null} { "industry_name": "fintech|payments", "country_hq": ["US","GB","DE"], "employee_count_min": 51, "founded_on_min": 2020, "page_verification_status": true, "count": 500 } ``` ### 3. Per-field DSL for surgical narrowing — 10% of calls When the cross-field `keywords` casts too wide a net, the team narrows by specific fields: * `specialities: "data engineering|ML"` — match only in declared specialities, ignoring marketing-copy description * `name: "ai|agent"` — match in company name (catches AI-native startup branding) * `description: '"sales intelligence" -consultancy'` — boolean exclusion of consulting hits ## The Data Pipeline Daily ICP refresh runs as a chained pipeline: | Step | Action | Powered by | | ---- | ------------------------------------------------------------------------------------------ | ------------------------------------ | | 1 | Pull 1,000–10,000 ICP candidates by keyword DSL | `db/linkedin/search/companies` | | 2 | Enrich with funding stage + total + growth score | `crunchbase/company` (cached lookup) | | 3 | Score each company against a 27-category ICP framework (HOT/WARM/MAYBE/EXCLUDE/COMPETITOR) | LLM with prompt-cached system prompt | | 4 | Filter to HOT tier (top \~2-5% of pool) | `query_cache` on the cache\_key | | 5 | Find 5–15 senior decision-makers per company | `linkedin/sn_search/users` | | 6 | Fetch last-90-day posts for buying-intent scoring | `linkedin/user/posts` | | 7 | Find email via vanity-resolved profile URL | `linkedin/user/find_email_by_url` | | 8 | Generate personalized 24–30 word hook anchored to a real post or homepage | LLM (higher-tier model for quality) | | 9 | Export ready-to-send leads to CSV for the outreach platform | `export_data` | End-to-end run time: \~1.5 hours for 14,000 companies. Cost out-of-pocket: \~\$15–20 in LLM calls. Discovery API calls: free under MCP Unlimited. ## Results & Scale | Pipeline metric | Value | | -------------------------------------------------------- | -------------------------------------------- | | Companies discovered (single sprint) | 14,000 | | Wall-clock for the discovery query | \~2 sec | | Companies scored as HOT (focus narrowing) | 319 (2.3% — 20x narrower than naïve filter) | | HOT companies in the primary spearhead segment | 211 (66% of HOT) | | Decision-makers found across HOT-linked companies | 16,000 | | Emails resolved via Tier 2 finder | 8,100 (62% hit rate after vanity resolution) | | Personalized hooks generated with verifiable source URLs | 245 (0 errors) | | Ready-to-send leads (hook + email + no role drift) | 180 | | Validation metric | Value | | ------------------------------------------------------------ | ------------------------------------------------- | | Crossover with an externally-sourced 279-company target list | 84% (234 of 279 already in pipeline) | | Of crossover, classified as HOT by the team's scoring | 99% (231 of 234) — independent confirmation | | Crunchbase enrichment match rate | 25% (3,298 of 13,972 companies have funding info) | The team operates the entire outbound funnel — discovery, enrichment, scoring, intent detection, personalization, delivery prep — without subscribing to a separate firmographics, list-building, enrichment, or delivery vendor. ## Key Anysite Endpoints Used | Endpoint | Purpose | Volume share | | --------------------------------- | ---------------------------------------------- | ------------ | | `db/linkedin/search/companies` | Bulk ICP discovery via SQL-style mini-DSL | 35% | | `linkedin/sn_search/users` | Senior decision-maker enrichment in batch | 25% | | `linkedin/user/posts` | Buying-intent signal harvest from recent posts | 20% | | `linkedin/user/find_email_by_url` | Email resolution for outreach delivery | 15% | | `crunchbase/company` | Funding/stage signal for tier scoring bumps | 5% | ## Why Anysite The combination of LinkedIn-depth firmographic queries + flexible mini-DSL + structured filters is unique to the platform. Recreating it elsewhere requires stitching four to six vendors — firmographics, sales-nav search, list-building, enrichment, funding intelligence — into a pipeline with per-record pricing at every layer. For sales teams that want to **own their ICP definition** instead of inheriting one from a vendor, the SQL-style discovery endpoint is the unlock: write your own scoring rubric, run it against the underlying database in seconds, and pay only for the rows that survive your filter. ## Key Takeaways * **Own your ICP, don't inherit it.** A structured, SQL-style discovery endpoint lets the team write its own scoring rubric instead of accepting a vendor's coarse categories. * **One endpoint replaces a four-to-six-vendor stack.** Firmographics, sales-nav search, list-building, enrichment, and funding intelligence collapse into a single pipeline. * **Discovery is cheap; precision is the value.** 14,000 companies narrow to 180 hand-picked touches — a 2.3% HOT rate, independently confirmed at 99% against an external target list. * **Boutique scale without sales-ops headcount.** A sub-10-person team ships 150 hand-picked outbound touches per week — work that would otherwise need a full sales-ops function and a five-figure annual tooling budget.