AI for Sales and LinkedIn outreach · Lesson 1
ICP and segmentation
How AI helps you figure out who to sell to, faster.
Why use AI for ICP
Building an ICP used to be a manual grind: look at past customers, hunt for patterns, write it up. AI speeds this up dramatically — it spots patterns in data quickly and proposes hypotheses you can then test.
Steps
- Gather data on current customers: company size, industry, decision-maker role, average deal size, retention.
- Hand Claude/GPT this data and ask it to find patterns: "which segments drive the most LTV, and which ones churn".
- Write the ICP down on a single page.
- Break it into sub-ICPs with distinct pains.
What matters
- Don't trust patterns built on 20 observations. That's noise.
- Validate hypotheses with interviews.
- Refresh the ICP once a quarter.
The trick
Give the model your customer data AND the list of reasons deals were lost ("closed-lost reasons"). Loss patterns often say more than win patterns.
Practical exercise
What to do after this lesson
Export a CSV of your customers and closed-lost deals. Hand it to Claude. Get ICP hypotheses. Validate them with 3 interviews.
Ready-to-use prompt
Template for this lesson
Copy and adapt to your context. Text in angle brackets should be replaced.
Help me build an ICP. Customer data (CSV): <…> Closed-lost data: <…> Give me: 1. 3 ICP hypotheses with the reasoning behind each. 2. For each one, the typical decision-maker profile. 3. Their main "pain" and the triggers. 4. How to validate this within 1 week.
Common mistakes
What people get wrong
- Trusting patterns drawn from a tiny sample.
- Not validating through interviews.
- An "ICP for everyone" — which is not an ICP at all.
Pro tips
What works but no one documents
- Closed-lost reasons are often more useful than wins.
- Refresh the ICP once a quarter.
- Build sub-ICPs with distinct pains.
When to use
Ongoing sales work.
When not to use
Too early a stage — you don't have customers yet.
Квиз — 2 вопроса
1.What is more useful for an ICP?
2.What's wrong with an "ICP for everyone"?
Отвечено: 0 из 2