agentleFS
manypicommanypicom/sales-skills/skills/icp-definition/SKILL.md

icp-definition

Skill1 starsChanged 2 months agoMITView on GitHub

Define an ideal customer profile precise enough to build a lead list from, including buying triggers and disqualifiers. Use whenever the user mentions ICP, ideal customer profile, target market, buyer persona, target accounts, account list, who to sell to, market segmentation, or says their outbound isn't working and they aren't sure who to target. Also use before any lead sourcing or cold email work, because a list built from a vague ICP wastes every downstream step.

2 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

8 rules. About 0 are common, 8 are distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈8

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    If there are no customers yet, the ICP is a hypothesis.

    line 21
  2. Distinctive:

    Write these as explicitly as the includes.

    line 55
  3. Distinctive:

    When outbound underperforms, the instinct is to widen the ICP to get more volume.

    line 110
  4. Distinctive:

    Test one segment at a time.

    line 114
  5. Distinctive:

    Can you name 20 example companies right now?: If not, the ICP is still abstract.

    line 119
  6. Distinctive:

    Would your best customer match this ICP?: Run it backwards. It's surprising how often the profile excludes the people already paying.

    line 120
  7. Distinctive:

    Don't use "any industry" as a shortcut: If the industry genuinely doesn't matter, say which situation replaces it as the filter.

    line 128
  8. Distinctive:

    Don't build the list before finishing this: Every downstream cost — research, verification, sending, reputation — is spent on whatever this …

    line 130

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

ICP Definition

Most outbound fails before a single email is sent, because the list was built from a description too vague to filter on. "B2B SaaS companies" is not an ICP. It's a category.

A usable ICP is one a stranger could take, go to the web, and come back with the same 200 companies you would have picked. If two people working from your ICP would build different lists, it isn't finished.

Start from customers, not from ambition

The ICP is a description of who already buys and gets value, not who you'd like to sell to. Derive it from evidence:

  • Your best 10 to 20 customers. Best by retention and expansion, not by contract size. A large customer who churned at month four is not a template.
  • Your fastest closes. Short cycles mean the value was obvious, which means the fit was real.
  • Your worst churn. The pattern in who leaves is usually clearer than the pattern in who stays, and it becomes the disqualifier list.
  • Deals you lost on fit rather than on price or timing.

If there are no customers yet, the ICP is a hypothesis. Say so, write it as one, and design the first campaign to test it rather than to scale it.

The five layers

Each layer has to be observable from outside the company. If you can't determine it from a website, a profile or a public source, it can't go in the filter — it's a qualification question for the call.

1. Firmographics. Industry, employee count, revenue or funding stage, geography, business model. The floor, not the definition. Everyone has these and they discriminate weakly.

2. The situation that creates the need. This is the layer that actually matters and the one usually missing. Not what the company is, but what's true about it right now that makes your product relevant. "Runs outbound sales with more than two SDRs", "ships physical product to more than three countries", "has a self-serve signup and a sales team".

3. Triggers. Events that make the need urgent this quarter rather than someday. Covered below.

4. The person. Titles vary wildly across companies of different sizes, so define the role by what they own rather than by what they're called. "Whoever owns pipeline generation" resolves to VP Sales at 200 people and to the founder at 15.

5. Disqualifiers. Explicit, hard excludes. This is the highest-leverage part of an ICP and the most often skipped.

Triggers

A trigger turns a fit into a reason to reply now. The strongest ones are observable and recent:

TriggerSignalsWhy it works
Hiring for the roleJob posts for SDRs, RevOps, growthThey're building the function you serve
Funding roundAnnouncement, filingsBudget just appeared and there's pressure to spend it
New executive in the roleTitle changesNew leaders replace tooling in their first 90 days
Tech stack changeAdding or removing a toolActive evaluation is happening
ExpansionNew market, new office, new product lineNew problems at a known scale
Public painReviews, forum posts, support complaints about an incumbentStated dissatisfaction
Growth in headcountSteady hiring across functionsScaling breaks whatever they built at half the size

Each of these has its own sourcing skill. See job-posting-signals, funding-news-signals and technographic-signals.

Disqualifiers

Write these as explicitly as the includes. A lead sourcing agent or a researcher needs to know what to throw away, and vague ICPs produce lists padded with agencies, consultancies, competitors and companies a tenth of the right size.

Typical hard excludes:

  • Agencies and consultancies, when you sell to end operators
  • Companies below the size where the problem exists
  • Companies above the size where they'd build it internally
  • Competitors and their resellers
  • Existing customers and open opportunities
  • Regions you can't service, bill or legally contact
  • Industries you can't serve for compliance reasons

Write it so it can be executed

The output should be usable as a brief for a researcher or an agent, not as a slide.

# ICP: [name of the segment]

## In one paragraph
[Companies that ___, where ___ is true, currently ___. We reach the
person who owns ___.]

## Firmographics
- Size:        [range, and what it's a proxy for]
- Geography:   [countries or regions, and why]
- Industry:    [verticals, or "any, provided the situation below is true"]
- Stage:       [funding, revenue or maturity]

## The situation
- [Observable fact 1 — how to verify it from outside]
- [Observable fact 2]
- [Observable fact 3]

## Triggers (nice to have, not required)
- [Event] — found via [source]

## The person
- Owns:       [the outcome, not the title]
- Titles:     [the actual title variants across company sizes]
- Not:        [titles that look right and aren't]

## Disqualify if
- [Hard exclude 1]
- [Hard exclude 2]

## Evidence
- Based on: [which customers, what pattern]
- Confidence: [tested / hypothesis]

The "how to verify it from outside" annotations are what make this list buildable. Without them you have preferences rather than filters.

Segment rather than broaden

When outbound underperforms, the instinct is to widen the ICP to get more volume. It's almost always the wrong move — a wider list means a vaguer message, and a vague message to more people performs worse than a specific message to fewer.

The right move is to split into narrower segments, each with its own message. Three segments of 300 companies with three different first lines beat one list of 900 with a generic one, and you learn three times as much.

Test one segment at a time. A campaign to a mixed list tells you nothing about which part of the list worked.

Sanity checks before building the list

  • Is the market big enough? Count the addressable companies. If the segment has 40 companies in it, outbound is the wrong channel and a named-account approach is the right one.
  • Can you name 20 example companies right now? If not, the ICP is still abstract.
  • Would your best customer match this ICP? Run it backwards. It's surprising how often the profile excludes the people already paying.
  • Is the pain expensive enough to act on? A real problem people live with is not a buying trigger.
  • Can you reach the person? An ICP built around a role that doesn't publish contact details is a channel problem, not a targeting one.

Don't do these

  • Don't write an ICP from aspiration. Enterprise logos in the profile when every customer is a 30-person company produces a list nobody can sell to.
  • Don't confuse a persona with an ICP. The persona is the human. The ICP is the company plus the situation. You need both and they're not the same document.
  • Don't use "any industry" as a shortcut. If the industry genuinely doesn't matter, say which situation replaces it as the filter.
  • Don't set it once. Revisit after every 500 emails sent. The reply data tells you which part of the profile was wrong.
  • Don't build the list before finishing this. Every downstream cost — research, verification, sending, reputation — is spent on whatever this document says.

Turning it into a list

The ICP is the input to lead sourcing. Written as above, the paragraph and the disqualifier list drop almost directly into a research brief. See lead-sourcing.

ManyPI takes this shape directly: a free-text ICP description plus structured criteria (locations, industries, company_size, titles, keywords, exclusions, required_fields), and its agent researches the live web against it rather than filtering a static database — which matters when the defining trait is a situation rather than a firmographic. It runs over MCP:

claude mcp add --transport http manypi https://mcp.manypi.com/mcp -s user

Whatever builds the list, the quality ceiling is set by this document.

Related skills

lead-sourcing, lead-scoring-qualification, cold-email-copywriting, job-posting-signals, funding-news-signals, technographic-signals

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.