HG Peer Cohort Selection

When to use

• A workflow is producing a peer-comparison brief or benchmark.
• A prompt is selecting "similar companies" and risks picking obvious-but-wrong matches.
• An author is choosing a peer set and needs a defensible, repeatable recipe.

Tools you'll touch

• company_firmographic: the target's NAICS, size, revenue and HQ; then verify candidates. Cost: 0.1 per company.
• search_industries_naics_sic: expand an industry group to its 6-digit leaves. Free.
• search_companies: find candidates. Cost: 1 per 100 companies returned. See hg-search-companies.

The four-axis rule

A defensible cohort aligns on four axes:

1. Industry: the 4-digit NAICS industry group as the grain. 6-digit is too narrow (functionally identical firms often sit in neighbouring leaves); 2-digit is too coarse (a sector like manufacturing bundles consumer goods with heavy industry). search_companies only matches 6-digit NAICS 2012 leaves, so expand the group to its leaves first.
2. Employees. A range around the target, set with firmographics.employees{min,max}. Default: one third to three times the target's headcount.
3. Revenue. Same approach with firmographics.revenue{min,max} (USD). Two firms with the same headcount and a 5x revenue gap have different IT-spend baselines.
4. Geography. Region is usually enough: firmographics.region_names ("AMER", "APAC", "EMEA", "LATAM"). Use country_codes only when regulation makes the country matter. Both filter on HQ, so a multi-region company is placed by its HQ.

Three axes aligned is defensible; four is ideal.

Recipe

1. Profile the target with company_firmographic: naics_code, employees_total (or employees_band), revenue_total (or revenue_band), country_code, company_level. If the target resolved to a subsidiary, move to the brand-level entity first (hg-subsidiary-handling).
2. Expand the industry group. Take the first four digits of naics_code and call search_industries_naics_sic with q:"<4 digits>", taxonomy:"naics", naics_leaf_only:true. Pass all returned leaves. Codes introduced in NAICS 2017 or 2022 don't exist in this taxonomy.
3. Set size and revenue ranges from the target's figures. If only a band is known, use the band edges.
4. Search with company_level set and at least two filter groups. The firmographic axes form one group; add a second one, usually the target exclusion below or an installs filter when the comparison is about a technology.
5. Verify the survivors with company_firmographic (batches of 25, matched by company_id). The search returns identity only, so check every axis on the enriched record before including a peer.

Generic example (placeholder codes and ids):

{
  "firmographics": {
    "company_level": "CORPORATE_PARENT",
    "naics_codes": [{"codes": ["<leaf 1>", "<leaf 2>", "<leaf 3>"], "inclusion_method": "ANY_PRESENT"}],
    "employees": {"min": 1500, "max": 15000},
    "revenue": {"min": 400000000, "max": 4000000000},
    "region_names": {"ids": ["EMEA"], "inclusion_method": "ANY_PRESENT"}
  },
  "company_identifiers": {
    "company_ids_including_corporate_relatives": {
      "ids": ["<target hg_id>"],
      "inclusion_method": "NONE_PRESENT"
    }
  },
  "sorts": [{"field": "id", "direction": "ASC"}],
  "limit": 50
}

The exclusion keeps the target and its whole corporate family out of its own peer set.

Why naive industry codes mislead

Two companies in the same 6-digit code can differ threefold on headcount and operate under different regulators. A peer list citing both is technically correct and practically misleading. Conversely, a mid-market software publisher and a mid-market IT services firm sit in different codes but may compete for the same budgets. The four axes catch both cases; industry alone doesn't.

Don't use the HG industry_name buckets for tech peers: there is no Software bucket.

When to expand the cohort

• Fewer than 5 verified peers → drop the geography axis (e.g. AMER + EMEA together).
• Still fewer than 5 → drop revenue (keep industry, employees, geography).
• Still fewer than 5 → say the company is in a niche: "there are no clean peers; the closest comparables are...".

Never expand by changing the industry group: that defeats the purpose. Never widen by removing company_level: you'd get subsidiaries and sites, not peers.

Common pitfalls

1. Passing a 4-digit NAICS code to search_companies. It matches nothing. Expand to 6-digit 2012 leaves.
2. Leaving out company_level. The cohort fills with subsidiaries and sites of a few groups.
3. Skipping the revenue axis when comparing spend. Same headcount, very different budgets.
4. Including the target (or its subsidiaries) in its own peer set. Exclude it with company_ids_including_corporate_relatives.
5. Ranking peers from the search result. It returns identity only; verify with company_firmographic.
6. Stating peer facts from memory. Quote HG's enriched figures, and bands where totals are null.

Reference

• V2 tool docs: https://phoenix.hginsights.com/docs/mcp-tools/v2/search-companies
• V2 tool docs: https://phoenix.hginsights.com/docs/mcp-tools/v2/search-industries-naics-sic
• Full docs for LLMs: https://phoenix.hginsights.com/docs/llms-full.txt
• Related skills: hg-mcp-calling-rules, hg-search-companies, hg-firmographic, hg-subsidiary-handling