The agent defines which companies to sell to, sharply enough that a BDR could build a list from it tomorrow and that it excludes most of the market. Marketing uses it to target, sales to qualify, product to prioritise. The ICP describes the account; the persona describes the human buyer inside it.
Done looks like: all eight dimensions defined (firmographics, tech stack, buyer persona, job to be done, existing alternatives, trigger events, budget authority, reachability) with specifics; written exclusions; an addressable list size; a weighted rubric that sorts real accounts into strong / workable / marginal / disqualify; icp_scorer.py --mode validate exits 0.
The ICP feeds project-management/gtm/gtm-strategy.
Before defining the ICP, settle the inputs below. When one is missing or vague and would change the result, ask about it; when it wouldn't, use a sensible default and say which one you used.
Ask about the two or three inputs that matter most for this request, not the whole list. If the user wants a result straight away, produce it and state your assumptions at the top.
# Complete ICP definition: exit 0
python3 project-management/gtm/ideal-customer-profile/scripts/icp_scorer.py \
--input project-management/gtm/ideal-customer-profile/assets/sample_icp.json --mode validate --format markdown
# Thin ICP (4 of 8 dimensions, no exclusions): exit 1
python3 project-management/gtm/ideal-customer-profile/scripts/icp_scorer.py \
--input project-management/gtm/ideal-customer-profile/assets/sample_icp_flawed.json --mode validate --format markdown
# Rank four accounts: Northwind 95 strong, Contoso 68 workable, Fabrikam 45 marginal, Tailspin 20 disqualify
python3 project-management/gtm/ideal-customer-profile/scripts/icp_scorer.py \
--input project-management/gtm/ideal-customer-profile/assets/sample_accounts.json --mode score --format markdown
# Input schema and flags
python3 project-management/gtm/ideal-customer-profile/scripts/icp_scorer.py --help
icp_scorer.py is the quality gate for this skill. What the scorer checks: validate mode counts the eight dimensions marked defined (fail under six), warns when firmographics are not specific, triggers are not listed, the addressable list is missing or under 200, or there are no disqualification criteria. Score mode normalises the weights to 100, converts each 0-3 rating to a share of its weight, and bands at 80 / 60 / 40. The defined and specific flags are self-reported: read the definitions too.
Exit codes: 0 no fail findings · 1 validate: fewer than six dimensions defined; score: no accounts · 2 unreadable file, invalid JSON or a top level that is not an object. --mode is required. JSON by default, --format markdown, --output FILE.
references/icp-dimensions-deep.md.--help; assets/sample_icp.json) and run validate mode. It must pass with no fail findings; then verify that each warning is fixed.--help). Run score mode on a real list and check the bands match the sales team's gut on accounts they know; if not, fix the weights.Mistake: "B2B companies with 100-10,000 employees."
Why it happens: Narrowing feels like losing deals.
Instead: Write exclusions first. A real ICP rules out most of the market.
Mistake: The ICP describes the customers the team wants, not the ones who buy, stay and expand.
Why it happens: Aspirational logos are more exciting than the current base.
Instead: Derive it from the top 20 customers and test it against closed-lost deals.
Mistake: A precise profile nobody can find through any channel the company has.
Why it happens: Reachability is treated as marketing's problem later.
Instead: Size the addressable list and name the channels before adopting the ICP.
More: references/icp-refinement-from-data.md.
In Scope:
--mode validate audits an ICP definition for completeness and disqualification criteria, and --mode score scores accounts with weighted criteria (icp_scorer.py)Out of Scope:
project-management/gtm/gtm-strategy/)product-team/ux-researcher-designer/)business-growth/revenue-operations/)business-growth/competitive-teardown/)Important Caveats:
defined or specific flag is entered by hand. The tool checks and weights them but does not verify them.| File | Purpose |
|---|---|
scripts/icp_scorer.py | ICP validation and account scoring |
assets/sample_icp.json | Complete eight-dimension ICP; validate exits 0 |
assets/sample_icp_flawed.json | Thin ICP; validate exits 1 |
assets/sample_accounts.json | Four accounts across all four bands |
assets/icp_template.md | Fill-in template |
project-management/gtm/gtm-strategy — uses ICP as inputproject-management/strategy-frameworks/business-model-canvas — segments blockproject-management/discovery/customer-interview-script — interview-based ICP discoverybusiness-growth/competitive-teardown — competitive contextc-level-advisor/cro-advisor — sales contextc-level-advisor/cmo-advisor — marketing context