Overview

Run Claude-driven account research without burning through your monthly credit budget. Claude learns which tool to reach for at each step (cheap firmographic lookup before expensive full enrichment), pre-filters territory sweeps to the accounts that actually matter, and prefers cached signals when freshness isn't the goal.

Use cases

500-account territory sweeps that don't bankrupt the org's credits
Claude pre-filters by cheap firmographic (1 credit/account) to the ~50 accounts worth deeper research, runs technographic + intent on the filtered set, and reserves full enrichment for the top 10. A territory pass that would've cost 10,000+ credits comes in under 1,500.
Re-runs that don't pay for already-known data
When the same workflow already loaded firmographic for an account, Claude reuses it instead of re-fetching. The second time you ask 'is this account in EMEA?' costs zero credits, not one.

View full skill

HG Credit Awareness

When to use

An author is choosing how to enrich an account list — fresh, cached, or derived.
A workflow is about to call company_research or full enrichment for every record in a list.
A prompt is iterating over a peer cohort and the cost adds up.

Phoenix per-tool credit costs

These are Phoenix-side credits charged to the org's plan, sourced verbatim from webapp/src/lib/pricing.ts (TOOL_CREDITS). They are distinct from any underlying HG API costs — the Phoenix layer abstracts the upstream metering.

Tool	Credits	Notes
`company_firmographic`	1	Direct lookup, cheap.
`search_companies`	dynamic	1 credit per company returned.
`company_technographic`	2
`company_cloud_spend`	2
`company_fai`	2
`company_intent`	2
`intent_category`	0	Catalog metadata.
`company_spend`	3
`company_contracts`	3	Per call (the call may return many contracts).
`company_operating_signals`	3
`company_install_time_series`	dynamic	3 credits per product returned.
`company_research`	dynamic	Sum of successful sub-tool costs.
`contact_search`	2
`contact_enrich`	3	Per record.
`sec_filing_section`	1
`web_search`	dynamic	0.01 (basic) / 0.02 (advanced), via MCP `_meta.creditCost`.

The truth source is TOOL_CREDITS. If this skill ever quotes different numbers than the file, the file wins.

Decision rules

Prefer cached over fresh when the answer doesn't change quickly.

company_firmographic — refresh weekly at most; this data moves slowly.
company_technographic — refresh monthly; intensity bands move quarterly at fastest.
company_intent — must be fresh; intent decays in 30-90 days.
company_contracts — refresh quarterly unless watching a specific renewal window.

Prefer derived signals over fresh enrichment when both work.

"Is the company in EMEA?" → use the firmographic record you already have, not a fresh company_firmographic call.
"Are they on the cloud?" → check the technographic stack you already loaded, not company_cloud_spend.

Cap territory sweeps. A workflow that calls company_research for every account in a 500-account territory bursts 500+ credits per call × N tools. Always:

Pre-filter the territory by firmographic (1 credit/account) to ~50-100 candidates.
Run technographic + intent on the filtered set.
Reserve full enrichment for the ~10-20 accounts the workflow is actually going to act on.

Common pitfalls

Calling company_research on every record because it's "comprehensive". It's the most expensive call; use it for the final 10, not the initial 500.
Re-pulling firmographic when the same workflow already loaded it. Cache within a single execution.
Quoting credit costs from memory instead of TOOL_CREDITS. The numbers change; the file is the truth source.

Reference

webapp/src/lib/pricing.ts — TOOL_CREDITS (single source of truth)
hg-insights-api.md#credit-costs — per-endpoint Phoenix credit notes

Skill: HG Credit Awareness