HG Intent

When to use

• A workflow needs a fresh signal that something is happening at an account.
• A prompt is generating outreach openings — intent gives the "why now".
• An author is tempted to lead with intent — read the "trigger, don't ground" rule below first.

Tools you'll touch

• company_intent — intent-topic readings per company
• intent_category — category-level filtering helper

What HG actually returns

Intent measures observed research behavior across HG's ad and content network. Each row gives:

• topic — the intent topic (e.g., "Tableau", "Customer Data Platform", "GDPR Compliance")
• score — a 0-100 reading of how much research activity HG has observed for this company × topic, normalized against the company's baseline
• stage — the buyer's-journey stage (see below)
• context_type — the kind of content the activity matched (general research, displacement signals, vendor-specific, regulatory, etc.)
• first_seen / last_seen — recency of the signal

The Phoenix-side credit cost is 2 credits per call (pricing.ts).

v1 vs. v2 schema (important for migration)

The v1 company_intent endpoint and the v2 /v2/intent endpoint return different schemas. v2 is recommended for new work — see hg-insights-api-v2.md#intent for the full mapping. Key differences:

• v2 splits "buyer's journey" out as a first-class field; v1 conflates it with score.
• v2 returns context types as a list; v1 returns a single primary context.
• v2 supports max_results consistently across paginated calls; v1 has endpoint-specific caps.

Buyer's-journey stages

Stage	Reading
Research	Early — scanning the category. Outreach lands as "hello, here's our point of view".
Awareness	Comparing options. Outreach lands as "here's how we differ from incumbent X".
Decision	Shortlisting / evaluating vendors. Outreach lands as "here's why us".
Negotiation	Active procurement. Outreach is too late unless you're already on the list.

Stage gates the message. A "Decision" stage outbound that opens with "have you considered modernizing your data stack?" reads tone-deaf — they're past that question.

The "trigger, don't ground" rule

Intent is a temporal signal — "something is happening". It is not a factual signal — "this is what they have". Two failure modes if you confuse the two:

1. Leading prose with intent: "The company is evaluating Tableau alternatives" — but you don't know if they currently run Tableau, what their analytics stack looks like, or what budget is in play. Pair intent with company_technographic + company_spend to ground the claim.
2. Replacing technographic with intent: an outreach that says "they're researching Snowflake" without checking the data warehouse stack can land at a company already running Snowflake at scale — embarrassing.

Canonical pattern: intent triggers the play, company_technographic + company_spend ground the open, FAI names the buyer. See hg-tool-chain-account-research.

Common pitfalls

1. Mistaking score for a forecast. Score 100 means "highly active research", not "they will buy". Plenty of companies research and don't act.
2. Ignoring last_seen. A score-100 reading from 6 months ago is stale.
3. Conflating v1 and v2 schemas in the same prompt. Pick one; document which.
4. Quoting the topic name as if it were a vendor. "Tableau intent" includes Tableau and Tableau-alternative research — read context_type to disambiguate.

Citation rules

Cite company_intent with both the score AND the date: "HG Intent: Tableau (score 100, Decision stage, May 2026)". The date is non-negotiable — stale intent is the most-cited offender in workflow output.

Reference

• hg-insights-api.md#intent
• hg-insights-api-v2.md#intent