icp-agent

Skill: ICP Agent

Purpose

Define, refine, and validate an Ideal Customer Profile using the Science of Scaling framework. Two modes: Quick Start (structured analysis from inputs) or Deep Dive (CRM data analysis with scoring model).

This skill combines Mark Roberge's Red/Yellow/Green framework with ICP expansion methodology and a data-driven analysis approach.

Inputs

• Company name (required)
• Mode: Quick Start or Deep Dive
• For Quick Start: Product description, top 5-10 best customers, why they chose you, common objections from poor-fit prospects
• For Deep Dive: CSV of closed-won/lost deals (12-24 months), customer success metrics if available

Process

Quick Start Mode (10-15 minutes)

Step 1: Gather Context

Ask the user about:

• What their product does and who it's for
• Top 5-10 best customers (company name, industry, size)
• Why those customers chose them over alternatives
• Common objections from poor-fit prospects

Step 2: Build the ICP Profile

Produce a structured ICP with these components:

1. FIRMOGRAPHICS: Industry, company size (employees + revenue), geography, growth stage
2. TECHNOGRAPHICS: Tech stack signals, tools they use, infrastructure indicators
3. BUYING TRIGGERS: Events or conditions that create urgency to buy
4. PAIN POINTS: Specific problems the product solves for this profile
5. DISQUALIFIERS: Red flags that indicate a poor fit
6. BUYER PERSONAS: Key titles involved in the buying process

For each component, provide specific, observable criteria, not vague descriptions.

Step 3: Apply Red / Yellow / Green Classification

Using the Science of Scaling framework, classify each attribute:

Attribute	GREEN (Target)	YELLOW (Inbound Only)	RED (Disqualify)
Employee Count	[ideal range]	[adjacent range]	[outside range]
Industry	[core verticals]	[adjacent verticals]	[excluded]
Geography	[primary markets]	[secondary markets]	[not served]
Tech Stack	[ideal tools]	[acceptable tools]	[incompatible]
Key Roles	[must-have roles]	[nice-to-have]	[missing critical]

Include a Change History section: document the date, what changed, and which customers moved as a result.

Step 4: Phase-Appropriate Recommendations

Based on the company's scaling phase (PMF / GTM Fit / Growth):

• PMF: Keep ICP tight. Validate hypothesis by talking to 20-40 customers (50% bullseye, 50% periphery). Measure success via retention, not close rate.
• GTM Fit: 90% of resources on proven ICP. 10% on 2-3 expansion experiments. Each experiment = mini-startup with its own PMF journey.
• Growth: Segment matrix (Product x Market x Channel) classified as Scale / Experiment / Ignore.

Deep Dive Mode (45-60 minutes)

Step 1: Upload and Analyze CRM Data

Load the CSV with pandas. For each attribute, calculate:

• Distribution (% of customers in each category)
• Correlation with deal size (do larger deals come from specific segments?)
• Correlation with sales cycle length (do certain segments close faster?)
• Correlation with retention (if available)

Step 2: Identify Best Customer Profile

Segment by ACV (top 25% = "Best Customers"). Compare Best vs Rest across all attributes. Identify hero signals: the 1-3 attributes that most powerfully separate best from rest. A hero signal must be observable from the outside (so it can drive prospecting) and must discriminate strongly (a large gap in prevalence between Best and Rest, not a few percentage points).

Step 3: Build Scoring Model

Create a weighted scoring model based on the analysis:

• Layer 1 (Account Fit, 0-100): Firmographic/technographic signals weighted by discrimination power
• Layer 2 (Engagement, 0-100): Behavioral signals if available
• Layer 3 (ACV Potential): Bucketed from actual data distribution

Weight each signal by how strongly it separates best customers from the rest in the data, not by intuition. Define tier routing: Tier 1 / Tier 2 / Tier 3 / Disqualify.

Step 4: Produce ICP Document

Output format:

1. Primary ICP: Highest LTV + fastest cycles + best retention
2. Secondary ICP: Worth pursuing with modified approach
3. Disqualification Criteria: Consistently underperforming segments
4. Red/Yellow/Green Scorecard: Observable attributes with classifications
5. Scoring Model: Weighted criteria with decision tree for account qualification
6. ICP Expansion Roadmap: 90/10 allocation plan with experiment hypotheses

Key Principles (from Science of Scaling)

• LTV over CAC: Define ICP by maximum lifetime value, not minimum acquisition cost
• Tight then expand: Constrained ICP for 0-5-20M, then experiment with 10% of resources
• Retention is the measure: ICP fitness = customer success, not close rate
• Hold the line: If a rep brings a $1M deal from a red company, tear it up
• Distractions vs opportunities: If it doesn't require 40%+ of dev resources to serve = opportunity. Otherwise = distraction.
• Division hopping: For enterprise expansion, start with a small deal to get the logo and security clearance, then expand across divisions

Output

Save to: [company]-icp-definition.md in your working directory.