gtm-scorecard

name	gtm-scorecard
description	Diagnose any company's GTM infrastructure health in 60 seconds — scores 7 dimensions from public data using web search only.
metadata	{"version":"2.0.0"}

GTM Health Scorecard v2

Diagnose any company's GTM infrastructure health from public data. Zero dependencies.

Input: A company domain (e.g., ramp.com) Output: 3 red flags → 7-dimension score → diagnosis → priority fixes With :prospect: All of the above + best contact + outreach DM draft

No pip install. No gather.py. Core scorecard needs no API keys — just web search. Two optional Exa boosters (SPA job-description bodies + people discovery) activate only if EXA_API_KEY is set; without it the scorecard runs fully on web search. Note the firewall distinction: Exa /contents returns raw page text (eyeball-able — fine to cite), but Exa /search structured entity fields are provider-asserted — treat them like any provider claim under Step 0.6, never as confirmed stack.

Step 0: Load GTM Context

If .agents/gtm-context.md exists, read it for context about your own company, ICP, and GTM motion. This helps tailor the fixes and outreach to your sales motion. It does NOT feed the diagnosis (see firewall below).

If it doesn't exist, the scorecard still works — it just produces generic fixes instead of ones calibrated to your offering.

Step 0.5: Evidence Firewall (READ THIS — non-negotiable)

The scores, red flags, diagnosis, and evidence section must be derived ONLY from public data gathered during THIS run (web search, the company's site, live JDs). Nothing else.

Do NOT let any of the following influence the scores, red flags, or diagnosis:

• Memory files, CLAUDE.md, sticky notes, or any persistent/ambient context about the user
• Whether the user knows, is meeting, is selling to, or has a relationship with the target company
• The user's own product/playbook (e.g. assuming the target needs the specific thing the user happens to sell)
• Prior runs or training knowledge presented as if it were observed this run

Why: pulling internal context inflates the output — it manufactures findings that fit the user's hopes instead of the target's reality. A scorecard the user walks into a real call with must reflect what's verifiable, or it burns credibility the moment the target contradicts it.

Allowed, with labels:

• Inferences beyond gathered data → mark (inferred) and phrase as a hypothesis to confirm, never as fact.
• Fix prescriptions / outreach framing → may reference the user's capabilities, but ONLY from .agents/gtm-context.md, and only after the diagnosis is locked. Frame these as "your angle," not as something observed at the target.

If you catch yourself writing a specific fix (named registries, data sources, tools) that came from knowing the user rather than from the target's public footprint, stop — either cut it or downgrade it to an open question.

Step 0.6: Tool & Tech-Stack Citation Rule (READ THIS — non-negotiable)

Every time you name a specific tool, CRM, or platform as part of the target's stack, it MUST carry all three of:

1. A verbatim quote of the exact sentence it came from (not a paraphrase, not a summary).
2. The source URL it came from.
3. A live/dead tag — (live as of [date]) or (removed/cached, [date]). If the only place the text survives is a search-engine snippet of a page that now 404s, it is dead — say so, and cap your confidence accordingly. A dead source can never be stated as a present-tense fact ("they use X").

Hard rules — these exist because the skill has gotten them wrong:

• Never collapse a "such as X, Y, or Z" list into a single asserted tool. A line like "experience with CRM software such as Zoho, Salesforce, or Hubspot" lists examples of acceptable candidate experience — it does NOT tell you which CRM the company runs. The correct read is "no specific CRM confirmed." Report the whole list verbatim and label it as a qualification example, never "CRM: Zoho."

• Separate "what the candidate should know" from "what the company runs." Tools in a Requirements / Qualifications / Nice-to-have section describe the applicant. Only tools in a Responsibilities / "our stack" / "you'll use" / "we run on" context describe the actual stack. Tag which section every tool came from. A tool that appears only in qualifications is unconfirmed, not detected.

• Never promote the first tool in a truncated search snippet. If a snippet shows "...CRM such as Zoho" and cuts off, fetch the full line before citing anything — the list almost certainly continues. Picking the first word off a truncation is how "CRM: Zoho" happened off a line that actually named three.

• The scorecard is web-only — it uses NO enrichment/data providers. If a tool claim ever traces to a provider (Apollo, BuiltWith, ZoomInfo tech-detection, a prior run, or memory), it is out of bounds for the scorecard — do not use it. Provider tech-detection is website-tag scraping (it flags marketing pixels/forms, not the sales stack) and is exactly the kind of unverifiable signal this skill is supposed to avoid. Eyeball-able beats provider-asserted, always.

When you cannot meet the three-part citation bar, write the honest version: "No [tool type] confirmed" + whatever the verbatim source actually supports. A missing tool finding is fine. A fabricated-confidence one burns the call.

---

Command: /gtm-scorecard

Step 1: Get the Domain

If user didn't provide one, ask. Clean input: strip https://, http://, trailing slashes.

Step 2: Research (Web Search — 6 searches, strict order)

Run these searches sequentially. Extract specific data points, not summaries.

Search 1 — What the company does + stage:

[company name] about company funding crunchbase

Extract: one-line description, funding stage, approximate headcount, HQ location.

Search 2 — SDR/AE hiring activity:

[company name] careers SDR OR BDR OR "sales development" OR "account executive"

Extract: exact role titles, count of open sales roles. If results are thin, also try:

[domain] site:greenhouse.io OR site:lever.co OR site:ashbyhq.com

Search 3 — Ops/infrastructure hiring:

[company name] careers "revenue operations" OR "sales operations" OR RevOps OR "GTM operations" OR "sales enablement"

Extract: exact role titles, count of open ops roles. The absence of results IS data.

Search 4 — JD deep dive (CRITICAL): Pick the 1-2 most interesting SDR/AE roles from Search 2. Search for the full JD text:

[company name] [exact job title from Search 2] job description responsibilities

Read the JD carefully for strain signals (see framework below). This is where the insight lives.

Search 4b — VERIFY JDs are live (CRITICAL): Before using any JD as evidence, verify it's currently posted: site:[domain] [job title] careers

Or check the company's careers page directly: [domain]/careers OR [domain]/jobs

If the role appears filled, removed, or the posting says "no longer available":

• Do NOT reference that JD in red flags, evidence, or outreach
• Note it as "(role previously posted, now filled)" in evidence section
• Score based on currently live roles only
• If no live sales roles exist, the company may have completed a hiring cycle — adjust diagnosis accordingly

This matters because: Referencing a filled role in outreach destroys credibility instantly. The recipient knows that JD is gone. You look like you're working from cached data, not live intelligence.

Search 5 — Tech stack signals:

[company name] sales tools salesforce hubspot outreach gong

Also check: tool mentions in any JDs found. Companies reveal their stack in "requirements" sections.

Apply Step 0.6 to everything you find here. A tool named in a JD's requirements/qualifications describes the candidate, not the stack — it is unconfirmed. Only a responsibilities / "you'll use" / "our stack" mention confirms the company actually runs it. Carry verbatim quote + URL + live/dead tag for every tool you cite, and never distill a "such as X, Y, or Z" list down to one tool.

Search 5b — Enrichment-stack JD sweep (CRITICAL for Dimension 2):

The enrichment/prospecting stack (Apollo, Clay, ZoomInfo, etc.) is internal, seat-based tooling — it never loads on the public website, so site fingerprinting and tech-stack search will not reveal it. The ONLY public surface where it leaks is the body of an ops/sales job description ("experience with Apollo/Outreach required"). Therefore:

1. Read EVERY live JD body, not just the 1-2 from Search 4. Modern boards (Greenhouse/Lever/Ashby) render JD detail pages as SPAs, so a plain page fetch returns an empty stub. Two reliable ways to get the full text:
   • ATS posting API (fastest): Ashby → https://api.ashbyhq.com/posting-api/job-board/[slug]?includeCompensation=true; Greenhouse → https://boards-api.greenhouse.io/v1/boards/[slug]/jobs?content=true; Lever → https://api.lever.co/v0/postings/[slug]?mode=json. These return full description bodies as JSON.
   • Exa /contents (optional, if EXA_API_KEY set) — preferred over rendering: when no ATS API is available, pull the JD page's crawled text directly. Cleaner and faster than spinning up the web-scraping skill:

     curl -s -X POST https://api.exa.ai/contents \
       -H "x-api-key: $EXA_API_KEY" -H "Content-Type: application/json" \
       -d '{"urls":["<job-url>"],"text":{"maxCharacters":6000}}'
     

     Returns raw page text (eyeball-able — fine under Step 0.6). Verify the returned page's domain matches the target; Exa occasionally returns a wrong-entity or binary page (~20% noise in testing) — discard those.
   • Fallback — invoke the web-scraping skill to render the SPA detail pages and extract the full JD text when no public API is available and Exa returned nothing usable.
2. Scan each body for this tool list and report tool → which role named it: Apollo, Clay, ZoomInfo, Clearbit, 6sense, Cognism, Lusha, LeadIQ, Outreach, Salesloft, Gong, Chorus, Salesforce, Segment, Census, plus any CRM/automation tool.
3. Score Dimension 2 from this sweep, not from a guess. If you have read all live JD bodies and none name an enrichment tool, that is an evidenced low score (drop the (inferred) tag) — phrase it as "across all N live JDs, the only revenue tool named is X." The honest ceiling: you still cannot prove a tool is absent from internal seats, so never claim "confirmed they don't use Apollo" — only "not named in any public JD."

Search 6 — Team structure:

[company name] "head of revenue operations" OR "VP sales operations" OR "director RevOps" LinkedIn

Extract: does a dedicated ops leader exist? How large is the sales org vs ops org?

Optional Exa booster (if EXA_API_KEY set) — surfaces profile evidence faster than LinkedIn-via-WebSearch, and is strong on tiny / repositioned companies:

curl -s -X POST https://api.exa.ai/search \
  -H "x-api-key: $EXA_API_KEY" -H "Content-Type: application/json" \
  -d '{"query":"RevOps or Sales Operations leader at [company]","category":"people","numResults":5,"contents":{"highlights":true}}'

Cite the returned profile URL (eyeball-able). Per Step 0.6, treat Exa's structured title/role as a claim to confirm against the page, not an asserted fact. Pin to the target's own domain — names collide across companies.

Step 3: Score Using the 7-Dimension Framework

Score each dimension 1-10. Cite specific evidence. Higher = healthier.

SCORING RULES:

• If data is missing, say so. Flag low-confidence scores with (inferred).
• "Not detected" ≠ "confirmed absent." Be explicit about what you couldn't verify.
• Combine gathered data with your training knowledge of the company.
• Any tool/tech you cite as evidence must clear the Step 0.6 citation bar (verbatim quote + source URL + live/dead tag; qualifications-section tools are unconfirmed; never collapse a "such as X, Y, or Z" list into one tool). If a score leans on a tool you can't cite to that bar, lower your confidence and mark it (inferred).
• Be opinionated. A confident wrong score that sparks conversation is more valuable than a hedged score that says nothing — but opinionated ≠ fabricated. Be bold on the diagnosis, never on unverified facts.

---

THE 7 DIMENSIONS

1. Sales/Ops Ratio

What it measures: Balance between revenue roles (SDR, BDR, AE, Sales Leadership) and infrastructure roles (RevOps, Sales Ops, GTM Engineer, Enablement).

Score	What it looks like
9-10	Ops proportional to sales headcount. Dedicated RevOps + enablement.
7-8	Ops present but slightly understaffed. Ratio ≤5:1 sales:ops.
5-6	One ops person supporting a growing team. Ratio ~8:1.
3-4	Many sales roles, minimal or zero ops visible. Ratio >10:1.
1-2	Actively scaling SDRs/AEs with ZERO ops anywhere — hiring or filled.

What most people miss: A company posting 5 SDR roles and 0 ops roles isn't just "growing sales." Their SDRs are about to drown in manual work because nobody is building the systems that make outbound scale. Every SDR hire without ops is a burnout timer.

---

2. Enrichment Maturity

What it measures: Whether the company has data enrichment tooling (Apollo, Clay, ZoomInfo, Clearbit, 6sense) or SDRs are manually researching prospects.

Score	What it looks like
9-10	Multiple enrichment tools (ZoomInfo + Clearbit + 6sense). Intent data present.
7-8	At least one enrichment tool detected or mentioned in JDs.
5-6	No enrichment detected but company likely uses it. Mark (inferred).
3-4	No enrichment detected AND JDs mention manual prospecting as core duty.
1-2	JDs describe SDRs building lists manually, researching via LinkedIn. Confirmed absent.

What most people miss: Enrichment isn't "nice to have." Without it, every SDR spends 30-40% of their time researching instead of selling. That's not a productivity problem — it's a unit economics problem. The company is paying $70K/yr for someone to do what Apollo does for $99/mo.

---

3. Outbound Infrastructure

What it measures: Whether the company has sequencing/outbound tools (Outreach, Salesloft, Instantly) for structured multi-touch campaigns.

Score	What it looks like
9-10	Sequencing tool + conversation intelligence (Gong/Chorus). Full stack.
7-8	Dedicated sequencing tool (Outreach, Salesloft).
5-6	Marketing automation (HubSpot, Marketo) but no dedicated sequencing.
3-4	CRM only. SDRs likely emailing from Gmail or basic CRM features.
1-2	No sequencing, no marketing automation. Outreach is fully manual.

What most people miss: A company using HubSpot for outbound sequences is burning their marketing domain's reputation. Dedicated sequencing tools exist because outbound and marketing email have different deliverability needs. HubSpot but no Outreach = outbound is either underperforming or about to get their domain blacklisted.

---

4. Data Hygiene

What it measures: Whether CRM data quality is maintained by ops or dumped on sellers.

Score	What it looks like
9-10	Ops roles mention CRM management. Analytics tools suggest data-driven culture.
7-8	CRM + ops roles exist. Likely maintained.
5-6	CRM detected but no clear ops ownership.
3-4	Sales JDs mention "maintaining CRM" or "data entry" as responsibilities.
1-2	SDR/AE JDs explicitly include CRM administration or database management.

What most people miss: When "maintain Salesforce data" appears in an SDR job description, that company doesn't have a data problem — they have a structural problem. They're paying their most expensive-per-hour-of-selling employees to do work that should be automated or owned by ops.

---

5. Automation Maturity

What it measures: Whether GTM tools are connected (Zapier, Make, Workato, Segment) or siloed with humans as the integration layer.

Score	What it looks like
9-10	Automation platform + data layer (Segment). Systems talk to each other.
7-8	Automation platform detected (Zapier, Make).
5-6	Multiple tools but no automation visible. Mark (inferred).
3-4	Multiple tools + no automation. Someone is exporting CSVs between systems.
1-2	Minimal tools + no automation + strain signals about manual processes.

What most people miss: A company with Salesforce + Outreach + Apollo but no Zapier/Make probably has someone manually exporting CSVs between tools. That person is either an ops hire (good) or an SDR doing it at midnight (terrible).

---

6. Role Scope Creep

What it measures: Whether sales roles are expected to do ops/infrastructure work.

Score	What it looks like
9-10	Clean separation. SDR JDs = prospecting + selling. No ops work.
7-8	Minor overlap — "maintain own pipeline hygiene." Normal.
5-6	Some creep — JDs mention tool management, reporting, process docs.
3-4	Clear creep — sales JDs include CRM admin, tool evaluation, "many hats."
1-2	Multiple strain signals. SDR JDs describe an ops role disguised as sales.

What most people miss: "Wearing many hats" in a sales JD isn't startup charm — it's a confession that the company hasn't built infrastructure. The SDR expected to prospect, sell, manage CRM data, build lists, evaluate tools, AND hit quota burns out in 6 months. Then the company posts the same role again wondering why they can't retain SDRs.

---

7. Ops Hire Timing

What it measures: Whether the company is actively investing in ops infrastructure or running without it.

Score	What it looks like
9-10	Ops roles posted AND existing ops team visible. Actively investing.
7-8	At least one ops role open. Building.
5-6	No ops roles but company small enough that founder handles it. Not yet critical.
3-4	No ops roles, many sales roles, company past the stage where someone should own this.
1-2	"Build from scratch" or "establish processes" language. Zero ops for too long.

What most people miss: A "first ops hire" posting at a 50+ person company means they've been running sales on duct tape for at least a year. That first ops person walks into a mess — dirty CRM, no documented processes, tools nobody configured, and a sales team with workarounds nobody can explain. This is the highest-signal indicator that a company needs external GTM help RIGHT NOW.

---

Step 4: Identify the 3 Biggest Red Flags

From the 7 dimension scores, pick the 3 lowest-scoring dimensions. These are the RED FLAGS. For each, write one punchy sentence explaining why it matters — using the "What most people miss" insight, not just restating the score.

If a company scores 7+ across all dimensions, the red flags section becomes WATCH AREAS — things that could slip as they scale.

---

Step 5: Format the Output

═══════════════════════════════════════════
  GTM HEALTH SCORECARD: [Company Name]
  [domain] | [what they do in <10 words]
  [Funding stage] · ~[headcount] employees
═══════════════════════════════════════════

  ⚡ 3 RED FLAGS

  1. [Dimension name]: [Score]/10
     [1-2 sentence "what most people miss" insight applied to THIS company]

  2. [Dimension name]: [Score]/10
     [1-2 sentence insight]

  3. [Dimension name]: [Score]/10
     [1-2 sentence insight]

  ─── FULL DIAGNOSTIC ───

  Overall: [X]/10 [🟢 ≥7 | 🟡 4-6 | 🔴 ≤3]

  1. Sales/Ops Ratio       [emoji] [X]/10  [1-line evidence]
  2. Enrichment Maturity   [emoji] [X]/10  [1-line evidence]
  3. Outbound Infra        [emoji] [X]/10  [1-line evidence]
  4. Data Hygiene          [emoji] [X]/10  [1-line evidence]
  5. Automation Maturity   [emoji] [X]/10  [1-line evidence]
  6. Role Scope Creep      [emoji] [X]/10  [1-line evidence]
  7. Ops Hire Timing       [emoji] [X]/10  [1-line evidence]

  ─── DIAGNOSIS ───
  [2-3 sentences. Confident. No hedging. What's happening,
  why it's happening, and what breaks first.]

  ─── PRIORITY FIXES ───
  1) [Most urgent + why]
  2) [Second fix]
  3) [Third fix]

  ─── EVIDENCE ───
  • [Specific data point — role counts, JD quotes, tools detected]
  • [Specific data point]
  • [Low-confidence flags if any]
═══════════════════════════════════════════

Emoji key: 🟢 7-10 | 🟡 4-6 | 🔴 1-3

Overall score: Weighted average. Dimensions 1, 2, and 6 get 1.5x weight (strongest predictors of GTM strain). Round to nearest integer.

---

Step 6: Offer Next Steps

Want me to:
  → Deep-dive on any red flag?
  → Run :prospect to find the right contact + draft outreach?
  → Score another company?

---

Command: /gtm-scorecard:prospect

Runs the FULL /gtm-scorecard workflow, then adds:

Additional Search — Best Contact

[company name] "VP of sales" OR "head of sales" OR "sales leader" OR "CRO" LinkedIn

If no VP Sales, try:

[company name] founder OR CEO LinkedIn

Optional Exa booster (if EXA_API_KEY set): category:"people" reliably surfaces the right contact even on small companies (in testing, one discovery query surfaced named sales/BD leaders across 5 sub-50-person pharmacies):

curl -s -X POST https://api.exa.ai/search \
  -H "x-api-key: $EXA_API_KEY" -H "Content-Type: application/json" \
  -d '{"query":"VP Sales, Head of Sales, or CRO at [company]","category":"people","numResults":8,"contents":{"highlights":true}}'

Use it for discovery; cite the profile URL. It won't give a verified email — that's a separate step.

Contact selection logic:

• SDR/ops imbalance → VP/Head of Sales (they feel the pain daily)
• Tech stack gaps → RevOps lead if exists, else VP Sales
• Company <50 people → CEO/Founder
• Growth/scaling strain → VP Sales or Head of Growth

SEQUENCING GUARD — pick the contact LAST, not first (WHO after WHAT). The heuristics above produce a candidate, not a target. Before treating anyone as the person to reach out to, you must have (1) verified the company's actual operating model and (2) a validated angle — then pick the contact who owns the pain that angle creates. Selecting a contact on a size/title heuristic before the angle exists is how you end up emailing the wrong person, or the right person with the wrong premise.

• Domain-expert founders are a trap. "Company <50 → founder" often points at someone who built the thing you're about to explain to them (e.g. an attorney-founder of a title company). To them, your "insight" is their daily reality — an instant first-order-to-expert miss. Verify who the person actually is (background/credentials, not just title) before drafting.
• The pain-owner ≠ the most senior person. On bespoke sends the buyer is usually the person who feels the gap (often a newly-hired GTM leader), not the founder who designed the model.

Outreach DM Draft

Write a message that follows these rules:

1. Open with a specific, verifiable observation. Not "I noticed you're hiring." Instead: "Your SDR JDs include CRM management and process documentation as core responsibilities — that's ops work in a seller's job description."

2. Connect to a consequence they can feel. Reference your own experience building GTM infrastructure. "In my experience building [specific systems], that pattern means [specific consequence]." If .agents/gtm-context.md exists, pull your credentials from there.

3. Offer the scorecard as value, not a pitch. "I ran your company through a 7-dimension GTM diagnostic I built — happy to share the full output if useful."

4. Credential through specificity. Reference your company, specific infrastructure types you've built (enrichment pipelines, routing automation, CRM cleanup), and concrete results. Generic "I help companies" = ignored. Specific "I built the enrichment pipeline that does X" = credible.

5. Low-pressure close. "Either way — cool product" or "figured the diagnostic might be useful." Not "let's hop on a call."

Prospect Output Format

Append after the standard scorecard:

  ─── OUTREACH BRIEF ───

  Target: [Name], [Title]
  Why them: [1 sentence — why this person feels the pain]
  Angle: Lead with Red Flag #[X] — [dimension name]

  ─────────────────
  Hey [Name] — [DM draft, 4-6 sentences max]
  ─────────────────

  Adapt for:
  • LinkedIn DM (trim to 3 sentences)
  • Email (add subject line)
  • Warm intro (softer open, reference mutual connection)

DM Quality Check

Before outputting, verify:

• Would the recipient think "this person has seen inside my company"?
• Is the observation specific enough it couldn't be sent to any random company?
• Does it feel like insight, not spam?

If NO to any → rewrite.

---

Command: /gtm-scorecard:batch

Score multiple companies from a text file (one domain per line).

Output a ranked summary table first, then individual scorecards:

═══════════════════════════════════════════════════════════════
  GTM HEALTH SCORECARD — BATCH RESULTS
  Scored: [N] companies | Generated: [date]
═══════════════════════════════════════════════════════════════

  RANKED BY INFRASTRUCTURE HEALTH (lowest = most opportunity)

  #1  🔴 3/10  warmly.ai    — 6 revenue roles, 0 ops, SDRs doing CRM admin
  #2  🟡 5/10  attio.com    — AEs building playbook, no ops layer, Series B
  #3  🟢 8/10  ramp.com     — Mature stack, RevOps in place, healthy ratios

  🔴 CRITICAL (1-3): [N] — immediate consulting opportunities
  🟡 WATCH (4-6): [N] — 60-90 day strain window
  🟢 HEALTHY (7-10): [N] — not a fit right now
═══════════════════════════════════════════════════════════════

---

Error Handling

• No career page found: Score from tech stack + training knowledge. Flag JD-dependent dimensions as (no career data — inferred).
• No tech stack detected: Common with SPAs. Don't auto-score 1. Use (not detected ≠ absent).
• Tiny company (<20 people): Note that dedicated ops may be premature. Adjust expectations.
• Enterprise (1000+ people): Expect mature infrastructure. Low scores here mean something is genuinely broken, not just "early stage."

---

Framework Origin

Built from direct GTM Engineering experience — building enrichment workflows, AI qualification systems, and multi-stage automation pipelines for enterprise clients. The dimensions come from watching dozens of companies go through the exact moment when their manual GTM workflows stop scaling. The "What most people miss" insights come from being the SDR who suffered without infrastructure AND the engineer who built it after.

---

Related Skills

• gtm-context — Your company context calibrates the diagnosis and outreach to your offering
• email-writer — Turn scorecard insights into full email sequences
• prospect-finder — Find and verify contacts identified by :prospect mode
• web-scraping — Deep-dive research on flagged companies