Intent Data Calibration

How to read HG's intent signals honestly: what each source means, where its confidence lies, and the cross-reference rules that keep claims defensible.

When to use

• A workflow scores account propensity using intent signals.
• A brief or PVP makes a claim about a target "evaluating", "displacing", or "researching" a product.
• A segmentation pass uses intent topics as a filter.

Two types of intent (NOT interchangeable)

Type	Source	Confidence	What it means	How to use
Bidstream (contextual)	Third-party ad exchanges and content networks	Low to medium (directional only)	Account is researching topics in this category	Corroborating context only, never as a commercial trigger
TrustRadius (buyer intent)	TrustRadius review and comparison activity (first-party)	High (verified behavioral signal)	Someone at the account is actively reviewing vendors	Flag as a lead signal when present, a meaningful buying behavior

The two are not interchangeable. A propensity score that adds them together without weighting will read confidently and be wrong. The bidstream signal is contextual: it tells you the topic is on the account's radar, not that anyone at the account is in market.

Intent context definitions

• Displacement. Account researching to replace a current vendor. MUST cross-reference with install data to confirm an incumbent exists. Without the install confirmation, the displacement framing is wrong.
• Whitespace. Account researching a category where no vendor is currently installed.
• Complementary. Account researching adjacent capabilities to expand an existing install.
• Expansion. Account researching to expand an existing vendor relationship.

The displacement-requires-install rule is the most common failure point. A bidstream signal showing intent against a CRM category does NOT establish that the account is displacing a CRM, only that someone on the account viewed CRM-related content. Cross-reference with company_technographic to confirm the incumbent install before any displacement claim.

The "evaluating" calibration warning

"Evaluating" means someone at the account is visiting product-related pages: product pages, pricing pages, demo requests. It does NOT mean they are in an active trial or POC.

Never tell a prospect "you are evaluating X" based on this signal. It overstates certainty and destroys credibility if wrong. Acceptable framings:

• "Activity on [vendor] product pages in the last [N] days" (descriptive, source-cited)
• "Researching the [category] space" (broad, defensible)
• "Engaging with [vendor]-comparison content" (specific to the signal class)

Unacceptable framings:

• "Evaluating [vendor]" (overstates)
• "In a trial of [vendor]" (fabricates)
• "About to switch from [incumbent] to [vendor]" (fabricates a commitment)

Known limitation: Phoenix intent

Phoenix MCP may underreport bidstream intent signals when queried directly. For high-stakes claims (a propensity score that drives campaign budget, an SDR-priority list of 100+ accounts), verify intent findings manually in the HG platform before finalizing. The HG platform UI surfaces the full bidstream plus TrustRadius mix, including signals that may not surface through the MCP tools.

Topic discovery

When the workflow needs to filter accounts by named intent topic rather than full-account intent, use list_intent_topics to enumerate the canonical topic taxonomy plus intent_category to map free-form category names to HG topic IDs. Both are zero-credit discovery tools and should be called before any company_intent query that filters by topic.

How to score propensity with intent

A defensible composite score weights signals by confidence:

1. TrustRadius buyer intent (high weight). Verified vendor-comparison activity. Boost the score; flag the specific topic.
2. Bidstream (medium weight, capped). Directional. Add as context; never let it dominate the composite.
3. Operating signals (per the HG Operating Signals skill, if invoked). Independent of intent; check cloud posture, AI maturity, network modernization separately.
4. Contracts (per the HG Contracts skill). Renewal-cycle proximity is a separate, structural signal: high weight when inside the 122-day window.

Cite each signal with its source class. A score line like "Composite 7/10: TrustRadius (high, last 14d) + bidstream (med, 90d) + contract expiry (122d)" reads honestly. A score line like "Composite 7/10" without source provenance does not.

Citation rules

company_intent should always be cited with the source class (TrustRadius vs bidstream) the first time it is referenced. Subsequent mentions can shorten. When mixing intent with technographic or contracts, call out the source for each: readers conflate "researching Snowflake" with "running Snowflake" and "about to renew Snowflake" without it.

Reference

• hg-insights-api.md#intent-endpoints: v1 company_intent schema
• hg-insights-api-v2.md#intent-company-intent: v2 /v2/intent schema and source-class fields