HG Insights

Skill: CSM Health Scoring

One health grade per account, every point traced to a named HG signal, and no grade when the evidence is too thin.

Overview

Grade every customer account Low, Medium or High on the same 0 to 100 composite: adoption (40), sponsor coverage (20), intent posture (20) and sentiment proxies (20), each sourced to an HG MCP Server signal. Missing or stale components are marked unscored and the score is rescaled, and Claude refuses to grade when the evidence is too thin.

Use cases

  • Portfolio triage with grades your team can defend

    Your CSM lead inherits a book of accounts. Claude scores each one on adoption, sponsor coverage, intent and sentiment, shows the points earned per component with the evidence and its date, and flags which accounts could not be graded and why.

  • Same composite, every workflow

    Account scorecards, advocacy finders and expansion plays all use one formula, so a Medium in one output means the same thing in every other. Nobody re-derives weights inline.

  • Renewal timing from your CRM, not a guess

    Renewal-cycle weight only applies when you supply the renewal date from CRM. Without it, Claude reports it as unknown and deducts nothing, instead of inferring a renewal from unrelated contract data.

View full skill

CSM Health Scoring

When to use

  • A workflow needs a single Low / Medium / High grade per customer account (composite score 0 to 100).
  • A workflow is about to invent its own weighting. Stop and reference this skill so all surfaces agree.

Tools you'll touch

  • get_vendor_information, product_search_and_enrich (action search), list_intent_topics, list_fai_departments: resolve product, topic and department ids. Cost: free.
  • company_install_time_series: adoption, the heavy weight. Cost: 3 per product row returned. Always pass product_ids.
  • company_technographic: product_last_verified_date for your product, used to trim the series. Cost: 2 per company.
  • contact_enrich: sponsor match rate against prior_contacts. Cost: 0.2 per matched contact with revealPhone:false; always pass it.
  • company_intent: intent posture on your category. Cost: 0.1 per call.
  • sec_filing_section (8-K) and web_search: strategic events. Cost: 1 per SEC call; 0.05 per basic search. Neither is cached.
  • company_fai: current department share for your product. Cost: 2 per call.

Out of scope: company_contracts. It covers GSI / ICT outsourcing contracts and federal awards only, with no renewal fields. Renewal-cycle weight comes from CRM only (see below).

Composite formula

ComponentMax PointsSourceScoring Rule
Adoption strength40company_install_time_series detection days + intensity_momentumAdoption matrix below
Sponsor coverage20contact_enrich match rate against prior_contacts20 × (contacts still at the customer ÷ prior contacts)
Intent posture20company_intent topic trend on resolved topic_ids≥ +1 → 20; between -2 and +1 → 10; ≤ -2 → 0
Sentiment proxies20Strategic events (10) + FAI department share (10)Rules below
Renewal-cycle weightoptionalour_contract_renewal_date from CRMAbsent: unknown, no deduction. Supplied: consumer rule on top

Adoption strength (40)

  1. Call company_install_time_series with your resolved product_ids and time_range:"last_6_months" or "last_12_months".
  2. Clean the series: drop the trailing null (current month); drop the month containing product_last_verified_date unless that date is at month end, and any later month (it is partly loaded). A null inside the series means no detection that month; count it as 0.
  3. Average detection days (data_points[].intensity, 1 to 31) over the last 3 complete, loaded months. 20 or more = heavy; 5 to 19 = regular; under 5 = sparse.
  4. Classify intensity_momentum: ≤ -20 declining; -10 to -20 early decline; otherwise stable or growing.
Detection BandStable or GrowingEarly DeclineDeclining
Heavy (≥ 20 days)403020
Regular (5 to 19 days)251810
Sparse (< 5 days)1050

Several products: score each and take the mean. Never score current_intensity (all-time aggregate, another scale).

Sponsor coverage (20)

Bulk-enrich prior_contacts (up to 25 per call) with firstName, lastName, company_domain and revealPhone:false. A contact counts as still at the customer when the returned organization.domain is the customer. Unmatched people are omitted from the result: match by name or id, never by position, and report "no match" separately from "moved". Contact data comes from a third-party provider, not HG. No baseline: mark the component unscored.

Intent posture (20)

Resolve your category's topics with list_intent_topics (one lowercase keyword), then:

{ "company_domain": "acme.com", "fields": ["topics"], "granularity": "global", "topic_ids": ["<ids>"] }

Take the median trend across your topics that come back. Thresholds are provisional (unit unconfirmed; observed values run about -3 to +7): ≥ +1 → 20 points; above -2 and below +1 → 10; ≤ -2 → 0. data_available:false, or none of your topics returned, means no coverage: unscored, not 10.

Sentiment proxies (20)

Strategic events (10). company_firmographic returns no events. Use:

  • sec_filing_section with companyTicker, filingType:"8-K", dateFrom 90 days back, and section "5.02" (officer changes), "2.01" (M&A) or "2.05" (exit or restructuring costs). US issuers only; it returns the most recent matching filing.
  • web_search for private companies or press coverage. Cite result URLs, not the answer summary.

No strategic-reset event (new CIO or CTO, M&A, layoffs or restructuring) in 90 days → 10; one event → 5; two or more → 0.

FAI department share (10). Call company_fai with product_ids for your product and sort_field:"department_usage_share". Rows repeat per role and location, so dedupe by department. department_usage_share is the product's usage mix across departments in %, summing to 100. The department that should own your product is among the top 3 by share → 10; present but lower → 5; absent → 0. FAI is a point-in-time reading with no trend field.

Renewal-cycle weight

From CRM only. If our_contract_renewal_date is absent, report "unknown" and don't deduct. If supplied, the consumer may apply its own rule (for example -10 at 120 days or less). Never derive it from company_contracts.

Grade mapping

CompositeGrade
70 or moreHigh
40 to 69Medium
Under 40Low

When components are unscored, rescale: composite = points earned ÷ max points available × 100, and say which components are missing.

Minimum to grade. Give a grade only when adoption is scored and at least 60 of the 100 points are scorable. Otherwise report "Not enough evidence to grade" with the components that were scored.

Product not detected. If company_technographic and the time series return no rows for your product, adoption is "not detected": not 0, and not silently unscored. Say it first. Either the customer isn't live, or HG can't see the deployment; ask the user to confirm with CRM usage data. No grade follows (see the minimum above).

Per-component minimum evidence

A component counts only if its evidence is fresh (see hg-recent-signals):

  • Adoption: product_last_verified_date within 90 days (is_active:true). Otherwise stale.
  • Sponsor coverage: contact_enrich is a live lookup; fresh.
  • Intent: topic last_seen_at within 90 days.
  • Sentiment: events within 90 days; FAI last_verified_at within 12 months.

Stale: mark it "stale, not scored" and reduce the maximum. Never invent a missing component.

Required content

Layout is up to the formatting skills. Whatever the layout, the answer contains:

  • The grade (Low, Medium or High) and the composite out of 100, or "Not enough evidence to grade".
  • For each component: points earned out of its maximum, the evidence value, its date, and the source tool in brackets. Example: "Adoption 25/40: 14 detection days a month over Jun to Aug 2026, intensity_momentum -3 [company_install_time_series]".
  • Components not scored, and why (no baseline, no coverage, stale, product not detected); when rescaled, the points available.
  • For adoption: the 3-month detection-day average with the months named, and intensity_momentum.
  • Renewal-cycle weight: the CRM date, or "unknown".

Common pitfalls

  1. Treating company_contracts as the customer's contract with you. It has no renewal data.
  2. Scoring adoption on current_intensity, or keeping the partly loaded last month.
  3. Sourcing strategic events from company_firmographic. It returns none.
  4. Reading FAI rows without deduping by department.
  5. Sponsor coverage without a baseline, or with the default phone reveal.
  6. Scoring "no intent coverage" as neutral, or inventing any missing component. Mark it unscored and rescale.

Reference