Intent Data Calibration

How to read HG intent honestly: what each source means, how far to trust it, and the cross-reference rules that keep claims defensible.

When to use

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

Tools you'll touch

• list_intent_topics: resolve topic keywords to ids before any topic filter. Cost: free.
• company_intent: HG topics and TrustRadius activities for one company. Cost: 0.1 credits per call.
• search_companies (intent.topics.ids, intent.signal_score): topic to companies. Cost: 1 per 100 companies returned.
• company_technographic: confirm the incumbent before any displacement claim. Cost: 2 per company.

Two sources, two field groups

Source	Field Group	What It Measures	Confidence	How To Use
HG topics (bidstream, contextual)	fields:["topics"]	Research activity across ad and content networks, scored 0 to 100	Low to medium, directional	Context and corroboration; a trigger only on a resolved topic
TrustRadius activities	fields:["activities"]	Someone at the account viewed reviews, comparisons or category pages on TrustRadius	High when relevant	Lead signal when it matches your vendor or category
Summary	default (no fields)	Counts from both sources	Overview only	Never cite as a signal

company_intent returns only the summary unless you ask. TrustRadius activity is invisible unless you pass fields:["activities"]. Request both groups in one call when you need both: fields:["topics","activities"].

Which filters apply where

Filter	HG Topics	TrustRadius Activities
topic_ids, signal_level, buyers_journey_names, context_type_names, granularity	Yes	No
vendor_ids, product_ids, signal_date	Yes	Yes
product_category_ids (from get_product_category)	No	Yes

A stage or level filter does not narrow activities. Pass granularity:"global" only when "topics" is requested.

Reading TrustRadius activities

Activities carry no score, level or stage. Read activity_label (reviews, comparison or category page), daily_views and activity_date, plus the vendors and products named. At large organisations the activity list includes browsing that has nothing to do with your category. Filter with vendor_ids or product_category_ids before treating any activity as a buying signal. An unfiltered activity at a large company is not a lead.

Intent context definitions

Valid context_type_names: Displacement, Whitespace, Expansion, Complementary.

• Displacement. Research suggesting replacement of a current vendor. Requires a confirmed install. Check company_technographic for the incumbent (current means product_last_verified_date within 90 days) before any displacement framing. Pair with vendor_ids to scope to one competitor.
• Whitespace. Research in a category with no vendor installed.
• Complementary. Research on capabilities adjacent to an existing install.
• Expansion. Research to extend an existing vendor relationship.

A topic tagged Displacement on a CRM category shows that someone read CRM-replacement content. It does not show that the account is replacing its CRM.

The "Evaluating" calibration

Evaluating is an HG topic stage: product page, pricing or comparison activity. Treat it as medium confidence. It does not mean a trial, a POC or a shortlist. buyers_journey_names usually holds both Researching and Evaluating, so the stage alone rarely separates accounts.

Never tell a prospect "you are evaluating X". Acceptable framings:

• "Activity on [vendor] product and pricing pages since [month]" (descriptive, sourced)
• "Researching the [category] space" (broad, defensible)
• "Engaging with [vendor]-comparison content" (matches a TrustRadius comparison activity)

Unacceptable framings:

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

Saturation and coverage

• Large companies show thousands of HIGH topics at score 100. A HIGH level alone says nothing; filter by resolved topic_ids.
• Smaller companies may return no topics. Report "no intent coverage", not "no interest".
• data_available:false is a definitive no-coverage answer.

Known limitation

The MCP may under-report bidstream intent compared with the HG platform. For high-stakes claims (a propensity score that drives budget, a priority list of 100+ accounts), verify findings in the HG platform before finalising.

Scoring propensity with intent

Weight signals by confidence and keep the sources separate:

1. TrustRadius activity, relevance-filtered (high weight). Comparison or review activity on your vendor or category, filtered by vendor_ids or product_category_ids. Flag the activity and date.
2. HG topic on a resolved topic (medium weight, capped). signal_level HIGH, recent last_seen_at, rising trend (≥ +1 weak, ≥ +2 strong; provisional thresholds). Never let it dominate the composite.
3. Install context (gate, not weight). A confirmed incumbent from company_technographic unlocks displacement framing; without it, cap the claim at "researching".
4. Operating signals (independent). Read raw company_ai_maturity scores, not stage labels.

Contracts are not a propensity weight: company_contracts covers GSI/ICT outsourcing and federal awards only and carries no renewal-likelihood data.

Cite each component with its source class. "Composite 7/10: TrustRadius comparison activity (2026-09-30) + HG topic data observability (score 91, HIGH, last seen 2026-09-24) + confirmed incumbent install" reads honestly. "Composite 7/10" alone does not.

Citation rules

Name the source class (HG topic or TrustRadius activity) the first time company_intent is cited. Later mentions can shorten. When mixing intent with technographic data, cite each: readers conflate "researching a warehouse" with "running a warehouse".

Common pitfalls

1. Adding the two sources together unweighted. The composite reads confident and is wrong.
2. Never requesting activities. TrustRadius is then silently absent from the analysis.
3. Expecting topic filters to narrow activities. Use vendor_ids, product_ids, product_category_ids or signal_date.
4. Displacement without an install. Always check company_technographic first.

Reference

• V2 tool docs: https://phoenix.hginsights.com/docs/mcp-tools/v2/company-intent
• V2 tool docs: https://phoenix.hginsights.com/docs/mcp-tools/v2/list-intent-topics
• Full docs for LLMs: https://phoenix.hginsights.com/docs/llms-full.txt
• Related skills: hg-mcp-calling-rules, hg-intent, hg-technographic, hg-contracts, opening-line-discipline