Case file 305 · Ideal Customer Profile · free · MIT

Know who to sell to, and who to turn away.

Hand your assistant the customer research you already have. It returns a profile of the customer most likely to get value, stay and expand: firmographics, behaviors, jobs to be done, pain points, the people who decide, and a clear list of who is not a fit.

  • Firmographics
  • Behaviors
  • Jobs to be done
  • 5-7 pains
  • Impact metrics
  • Decision makers
  • Journey
  • GTM messaging
  • Not a fit
  • Ideal-of-the-ideal
FIT

Ideal-of-the-ideal

Size
25-40 people
Industry
Design or software studio
Location
Portland delivery area
Office days
3-5 in office
Facilities
No kitchen
Buyer
Office manager ordering for 30+

NOT FITUnder 15 staff · fewer than 3 office days · staffed kitchen · daily hot meals a must · outside delivery area

Excerpt from a saved run on “Tidewell Desk”, an invented lunch-kit startup. Wording shortened; the run itself marked its data as thin.
What it is

A synthesis method, not a guess

One instruction file that walks a capable AI assistant through six steps: gather the data, segment customers by value, profile their firmographics, map how they buy, define their jobs to be done, and document their pains and needs.

It builds on Jobs to Be Done theory (Clayton Christensen) and asks the assistant to keep both the ideal profile and the acceptable secondary segments in view.

Who it is for
  • Founders deciding where to focus scarce sales time.
  • Product managers turning interviews or PMF survey data into a target.
  • Marketing and sales leads writing messaging and qualification rules.
  • Researchers who need a structured readout of what they collected.
FIT

Works well for

  • Defining an ICP from product-market fit survey data
  • Finding patterns in churn, expansion and best accounts
  • Prioritizing sales and marketing effort
  • Checking a new opportunity against the profile
  • Refreshing the target market each quarter
NOT FIT

Not designed for

  • Collecting research for you; it synthesizes what you supply
  • Live market or firmographic data lookups
  • Replacing customer interviews (the Skill itself suggests 10+)
  • Credit, legal or compliance screening of real people
Intake and return
You provide
  • Interviews, survey responses or transcripts
  • Customer success, churn or retention figures
  • Product usage analytics
  • Sales activity and win/loss notes
  • Your customer list and competitive notes
You receive
  • Firmographic and behavioral profile
  • Functional, emotional and social jobs
  • Top 5-7 pain points with impact figures
  • Decision process, stakeholders and journey
  • GTM messaging, disqualifiers and the ideal-of-the-ideal

Where it runs

Any AI assistant that can read a Skill file, including Claude Code and Codex. Download the free plaintext package from Finch, then install it with your assistant’s Skill method or ask the assistant to read SKILL.md.

No extra API key, script or service. You still need access to an assistant model, which may have its own subscription or usage cost. The Skill does no web research of its own; it works from the material you give it.

Limits and licence

A profile is only as sound as the research behind it. With few interviews, treat the result as a hypothesis and check every item the assistant marks as unknown or assumed. Share only customer data you are allowed to give your assistant.

MIT licence, Copyright (c) 2026 Pawel Huryn. Source: phuryn/pm-skills at commit 2a12295a7717, path pm-go-to-market/skills/ideal-customer-profile. The package adds a README, NOTICE, provenance note and a cover; the instructions are unchanged.

How it was checked

One Claude Code run on a fictional company produced all ten parts the Skill promises, with seven pain points and five disqualification criteria. It said plainly that six interviews fall short of the ten the Skill suggests, and marked 18 places as UNKNOWN or ASSUMPTION. Two parts, impact figures and journey timeline, could only be partly filled from the data. The package downloaded back from Finch matched the audited files byte for byte. Native installation in each client, repeat runs and real business outcomes were not tested. See Benchmark for the record.