HG Spend

When to use

• A workflow needs an order-of-magnitude IT-spend number for benchmarking, sizing or peer comparison.
• A prompt is about to claim "their X spend is high/low" and needs a defensible comparison set.
• A workflow needs AI spend in dollars.
• An author is about to inline a spend benchmark. Reference this skill instead.

Tools you'll touch

• company_spend: modeled IT spend in USD by category and country. Cost: 3 per company (batch up to 25; a batch of 10 costs 30).
• company_ai_spend: modeled AI spend in USD by AI category and country. Cost: 2 per company.
• product_search_and_enrich (action enrich): returns a spend_category_id per product. Cost: 1 per matched product.
• company_cloud_spend is not a dollar tool: it lists cloud, CDN, hosting, DNS and SaaS vendors with first_seen dates and has no dollar figures. See hg-ai-and-cloud-signals.

Results cache for 30 days org-wide. Pass hg_ids or domains, not both, and match results by company_id, never by position.

What HG actually returns

Modeled estimate, not invoices.

• The total reflects what HG estimates the company spends operating its IT footprint, including services, not just license fees.
• Response: companies[].spend.all (or ai_spend.all), rows of spend (USD), category_name, category_id, country_name, country_code, plus all_count.
• Rows are flat: one row per category × country, with no parent or level field. Rows such as "Total IT", "Total External IT", "Software" and "Services" are separate, overlapping totals, not parts that add up.
• A domain can resolve to a small subsidiary rather than the group. If spend looks far too small for the brand the user named, check company_level in company_firmographic and re-query on the parent id (see hg-subsidiary-handling).

How to call it

Resolve category ids from the data. get_product_category is the product taxonomy, not the spend categories. To filter:

1. Call company_spend unfiltered for one company and read category_id / category_name from the rows; or take spend_category_id from a product enrich.
2. Filter later calls with category_ids.

category_names is a case-insensitive substring match (max 10). A name that matches no category returns 0 rows silently, and a short name can match several categories ("Security" may hit more than one row). Prefer category_ids.

Pagination. max_results is 1 to 25 rows per company; there is no has_more. Page with offset (0, 25, 50, and so on) until a page returns fewer rows than max_results.

{ "domains": ["acme.com"], "category_ids": ["<id read from a prior spend row>"], "max_results": 25 }

How to read it

Never sum overlapping categories. Pick the one row that answers the question:

• Overall IT budget: "Total IT".
• Spend with vendors and service providers (the most comparable to a budget line): "Total External IT".
• Internal spend (own staff time and infrastructure), when both rows exist: "Total IT" minus "Total External IT". Label it as derived.
• Sum across countries only within one category, and only when no all-country row for that category is present.

Shares. A category share is the category row ÷ the "Total IT" row for the same country scope. State the scope.

Peer comparison is the defensible benchmark. Batch the account with 5 to 20 peers of similar industry and size in one company_spend call and compare shares. A rough working heuristic for security is a high-single-digit to low-double-digit share of IT spend in critical-infrastructure industries (financial services, healthcare, public sector) and a lower, wider band elsewhere. It is a heuristic, not an HG statistic; never present it as a benchmark.

Why low security spend is a signal, not a gap. A low reading can mean (a) the company outsources security to an MSSP and the spend sits in services, (b) security is embedded in cloud-platform fees rather than a separate line, or (c) genuine under-investment. Check the technographic stack before claiming "underspending".

AI spend (company_ai_spend)

• Same shape and rules: flat rows, category_ids preferred, category_names substring, 25 rows per page with offset.
• Categories include "Total AI Spend", "AI Software", "AI Services" and "AI Hardware". "Total AI Spend" overlaps the others; don't add them together.
• For AI maturity scores rather than dollars, use company_ai_maturity (see hg-ai-and-cloud-signals).

Common pitfalls

1. Treating modeled spend as invoiced. "They spend $50M on cloud" is a modeled estimate; "they have $50M of cloud invoices" is something HG can't tell you. Hedge with "modeled" or "estimated".
2. Adding "Total IT", "Software" and "Services" together. They overlap.
3. Filtering with a guessed category name. 0 rows looks like "no spend".
4. Stopping at the first 25 rows. Page with offset.
5. Comparing across industries without normalization. A 6% security share at a SaaS company and at a regional bank read very differently.
6. Quoting company_cloud_spend as dollars. It has none.
7. Missing a subsidiary resolution. Subsidiary-scale numbers for a group-level question.

Citation rules

Cite company_spend and company_ai_spend as "HG modeled spend" the first time they appear in a deliverable. After that, "modeled" is enough; don't repeat it on every line. See hg-citation-discipline for source priority and density.

Reference

• V2 tool docs: https://phoenix.hginsights.com/docs/mcp-tools/v2/company-spend
• V2 tool docs: https://phoenix.hginsights.com/docs/mcp-tools/v2/company-ai-spend
• Full docs for LLMs: https://phoenix.hginsights.com/docs/llms-full.txt
• Related skills: hg-mcp-calling-rules, hg-ai-and-cloud-signals, hg-modeled-vs-observed, hg-subsidiary-handling, hg-citation-discipline