How to build an ICP from scratch

Building an ICP (Ideal Customer Profile) from scratch takes 30 minutes if you approach it correctly. The goal is not a perfect ICP — it's a specific enough ICP to generate a list and write a message. You sharpen it later with real data from outbound.

Step 1: Start with your best customers (or best hypothesis)

If you have customers: List your top 5 accounts by one or more of these criteria — fastest to close, highest retention, most expansion, most positive about the product, least support burden. What do they have in common?

If you have no customers: Think about who has the most acute version of the problem you solve, can afford to pay for a solution, and has a decision-maker with authority to buy. Write that down.

Step 2: What company signals actually predict fit?

For each of your best-fit accounts, pull six data points: industry or vertical, employee count or revenue range, geography, business model (B2B, B2C, SaaS, services), funding stage, and any tech stack signals that show up consistently. The signal that matters most is whichever one shows up in four out of five of those accounts without you looking for it.

The resulting company profile might look like: "B2B SaaS companies, 20–200 employees, Series A to Series C, US-based, using HubSpot or Salesforce."

Step 3: Who actually signed the contract?

Go back to your best accounts and find the person who signed. Not who attended the demo — who signed. What was their title, what were they responsible for, and what pushed them to buy when they did? That last question is the one most people skip. The answer to "why now" tells you more about your real ICP than any firmographic filter.

The resulting persona might look like: "VP Sales or Head of Revenue, responsible for pipeline generation and SDR productivity, frustrated that outbound is inconsistent and institutional knowledge walks out with every hire."

Step 4: Write the one-sentence ICP

Combine the company criteria and buyer persona into one sentence: "We sell to [buyer title] at [company type], [company size], who are experiencing [specific problem]."

Example: "We sell to VP Sales and Head of Revenue roles at B2B SaaS companies with 20–200 employees who are running founder-led sales and trying to build a repeatable outbound motion."

If you can't write this in one sentence, your ICP is too broad.

Step 5: Can you find 50 companies that match right now?

Search Apollo, LinkedIn Sales Navigator, or Clutch using your ICP criteria. Can you find 50–100 companies that match? If yes, the ICP is specific enough to generate a list. If you get thousands of results with no clear boundary, narrow it further.

Step 6: What does outbound data tell you to change?

The ICP is a hypothesis until outbound data confirms or disproves it. After 50–100 outreach messages, you'll see patterns: which sub-segments reply more, which titles ignore you, which company sizes convert fastest. Use that data to narrow or adjust.

In an AI GTM system, ICP refinements get written as coaching notes to the SDR agent — for example, "ignore companies under 30 headcount, they churn in 60 days" or "VP of Revenue converts 3x faster than VP Sales at this size." Those notes become memory blocks that filter every future prospecting cycle without you having to re-explain the criteria each time.