Ideal Customer Profile

Ideal Customer Profile

How This Works

An ICP built from opinion is a guess with a template around it. This skill builds one from evidence — survey responses, interview transcripts, usage data, win/loss notes — synthesized across four layers: who they are (demographics), how they decide (behaviors), what they're hiring your product to do (Jobs to Be Done), and what's actually broken for them (needs). It deepens your brain's existing Section 2 in place rather than creating a second ICP file — one source of truth stays intact for every downstream skill.

Step 0 — Load brain Section 2 (current ICP, however thin) and any guardrails from /context/meta-patterns.md.

Step 1 — Gather and segment: collect available research, then identify which customer cohort actually has the value — highest retention, fastest time-to-value, best expansion — since that's who the profile should describe, not just "whoever we have."

Step 2 — Profile across the four layers: Demographics, Behaviors, JTBD, Needs.

Step 3 — Deepen brain Section 2 with the profile (on user confirmation).

Step 4 — Learning Close: log the session to /context/skill-sessions.md.

Trigger

• When: Defining or sharpening who you sell to from research data — PMF surveys, interview transcripts, usage analytics, win/loss notes — or when brain Section 2 is firmographic-only with no behavioral or JTBD layer.
• Not for: Choosing among several candidate segments once the ICP is profiled → beachhead-segment, run after. Turning a confirmed ICP into positioning or messaging → positioning-messaging, run after. Full brain build from zero → product-marketing-context, run first if no brain exists at all.
• Example prompts:
  • "Build my ICP from our PMF survey data"
  • "Here are 8 customer interviews — define our ideal customer"
  • "Who are our best-fit customers? I have usage and churn data."
  • "Our ICP is just company size and industry. Go deeper."

Inputs

• Args: Research data — PMF survey responses, interview transcripts, usage or expansion data, win/loss notes, support tickets. If none is provided, ask for it; this skill synthesizes evidence, it doesn't generate a profile from assumption alone.
• Defaults: If brain Section 2 already has demographics but no behavioral, JTBD, or needs layer, treat this as additive — profile the missing layers, don't re-litigate settled firmographics.
• Context keys:
  • /foundation/brain.md — required. Section 2 (ICP).
  • /context/meta-patterns.md — optional; guardrails from prior sessions.
  • Brain contract: Reads Section 2. Writes Section 2 only — the four profile layers, appended on explicit confirmation. Never creates a separate ICP file.

Pre-flight

• Load /foundation/brain.md Section 2 if it exists.
• Load /context/meta-patterns.md if present; surface any guardrail fired 2+ times in prior ICP sessions.
• Hard block: brain entirely absent → stop, direct to product-marketing-context first. A thin Section 2 is not a block — that's what this skill exists to fix.
• Soft block: no research data provided and none referenced from brain → ask for it before profiling. A profile built on assumption alone should say so explicitly, not pass as evidence-based.

Steps

Step 0 — Load Context

Load brain Section 2 as it exists and any fired guardrails.

Gate check: brain absent → surface: "Brain not found. Run product-marketing-context first — even a thin Section 2 gives this skill something real to build on."

Step 1 — Gather and Segment by Value

Ask what research is available if not already provided:

"What do you have to work with? Survey responses, interview transcripts, usage data, win/loss notes, churn analysis — any of these work, and more than one is better."

If multiple customers or segments are represented in the data, identify which cohort the profile should actually describe — the highest-value one, not an average across everyone:

• Highest retention or lowest churn
• Fastest time-to-value
• Strongest expansion or upsell pattern
• Best reference or case-study potential

State which cohort you're profiling and why before moving to Step 2. A profile that averages your best customers with your worst describes nobody.

Step 2 — Profile Across Four Layers

Every claim traces to a source. Mark anything without one [A] (assumption) — same convention beachhead-segment uses. A profile that hides its assumptions behind confident prose is more dangerous than one that states them.

Demographics — who they are on paper: company size, industry/vertical (specific — "SaaS" is not a vertical), geography, job title and department, org structure and budget ownership.

Behaviors — how they decide and adopt: how they discover and evaluate solutions, buying process and timeline, solo decision vs. committee, technical adoption speed, tool-switching frequency, peer influence.

Jobs to Be Done — what they're actually hiring the product to do:

• Functional job (the concrete outcome they need produced)
• Emotional job (how they want to feel afterward)
• Social job (how they want to be perceived by their team)
• What they're trying to stop doing or stop experiencing

Needs and Pain Points — what's actually broken: specific pain points, current workarounds and their limits, cost or time burden, available budget, competing priorities. Rank the top 3–5 by how often they showed up in the source material, not by how compelling they sound.

Exit check — the One-Line Test: can this compress into one sentence a salesperson could use as a 10-second qualification filter?

"[Company type] in [vertical] at [size], where [buyer role] needs to [functional job] because [top pain point]."

If it can't compress, a layer is still too vague — usually Behaviors or JTBD. Return there before Step 3.

Step 3 — Deepen Brain Section 2 (on Confirmation)

Show the exact addition before writing:

"Adding to brain Section 2 — Demographics, Behaviors, Jobs to Be Done, and top Needs/Pain Points, built from [source data]. Existing content stays unchanged. Here's what I'll append: [show exact text]. Save this?"

Append to /foundation/brain.md Section 2 — do not replace existing content, do not create a separate file:

## Section 2: ICP Definition
[existing content unchanged]

### Behavioral & JTBD Profile (added via ideal-customer-profile, [date])
**Source:** [survey / interviews / usage data / win-loss — n=X if known]

**Behaviors:** [how they discover, decide, adopt]

**Jobs to Be Done:**
- Functional: [one sentence]
- Emotional: [one sentence]
- Social: [one sentence]

**Top Pain Points (ranked by frequency):**
1. [Pain — source]
2. [Pain — source]
3. [Pain — source]

### One-Line ICP Statement
[Company type] in [vertical] at [size], where [buyer role] needs to
[functional job] because [top pain point].

Never write without this explicit confirmation.

Step 4 — Learning Close

Append one row to /context/skill-sessions.md (create with header row if absent):

skill: ideal-customer-profile
session_date: [YYYY-MM-DD]
pattern: [one falsifiable statement about this session, or "none"]
source: [surprised / wrong / missing / n.v.t.]

Write directly, no permission needed — separate from the brain write above, which still requires explicit confirmation.

Outputs

• Files written: /foundation/brain.md Section 2 — appended Demographics (if missing), Behaviors, JTBD, and ranked Pain Points (Step 3), only after explicit confirmation. /context/skill-sessions.md — one appended row per session (Step 4).
• Chat output format: Source and cohort stated → four-layer profile → One-Line ICP Statement → confirmation prompt for the brain write.
• External side effects: n.v.t.
• Next skill: After brain Section 2 is deepened, check product-marketing-context/.claude-plugin/next-skill-map.md for "After ideal-customer-profile" and surface that prompt. Do not auto-run — ask.

Verification

• Cohort being profiled is stated and justified, not just "our customers."
• All four layers profiled: Demographics, Behaviors, JTBD, Needs.
• Every claim without a cited source flagged [A].
• Pain points ranked by source frequency, not asserted importance.
• One-Line ICP Test passed before the brain write is proposed.
• Brain Section 2 write shown to the user and confirmed before it happens.
• Existing Section 2 content preserved, not overwritten.
• Session logged to /context/skill-sessions.md.

Operating Rules

• Deepen brain Section 2 — never fork a second ICP file. Every brain-reading skill in this stack expects Section 2 to be the ICP.
• Evidence over assertion. Every claim without a cited source gets [A] and lowers confidence in the result.
• Profile the highest-value cohort, not the average customer. Averaging your best and worst accounts produces a profile that fits neither.
• Rank pain points by frequency in the source data, not by how compelling they sound in the room. The loudest pain isn't always the most common one.
• The One-Line Test is mandatory. If the ICP can't compress to a 10-second qualification sentence, a layer is still too vague.
• Never overwrite existing Section 2 content. Append; don't replace unless the user explicitly asks to revise settled firmographics.
• Brain write requires explicit confirmation. Never append silently.

Quality Gate

Do Not Use For

• product-marketing-context — full brain build from zero, or any section other than Section 2.
• beachhead-segment — choosing among multiple candidate segments once the ICP is profiled; run this skill first for real inputs.
• positioning-messaging — turning a confirmed ICP into a positioning statement or messaging hierarchy; run this skill first, then that one.
• n.v.t.