Replacing gut feelings: Using AI to sharpen the ICP

Most sales teams in SMEs know their customers - or so they think. If you ask, the ideal customer profile often sounds like this: "medium-sized industry, something with production, 50 employees or more". That's not a profile. It's a feeling. And you don't build a plannable pipeline on feelings.
The problem is not laziness, but a lack of data. A sharp Ideal Customer Profile (ICP) is created by analyzing your own best customers - and nobody does this on a day-to-day basis because it is laborious. This is exactly where AI shifts the economy: what used to take a day of workshops becomes a 30-minute task.

Step 1: Your own existing customers as raw material

Before any prompt is written, material is needed. Export your 15-20 most profitable and loyal customers from the CRM - not the biggest ones, but the ones with whom cooperation runs smoothly and the margin is right. Make a note of each one: industry, number of employees, region, reason for the initial purchase and - if known - which specific problem you solved.
This last point is the most important and the most often overlooked. A good ICP does not describe what a customer looks like, but what situation they are in when they need you.

Step 2: Work out patterns with AI

With this list, a language model can be used as an analyst. The following prompt finds commonalities that get lost in day-to-day business:

You're an experienced B2B sales analyst. I give you 
a list of our best existing customers with industry, 
size, region and the problem we solved.

Analyze the list and work out:
1. which 3-5 characteristics appear more often than average 
often?
2. what common trigger event did these customers have before 
did these customers have before they bought?
3. which industries or sizes do NOT seem to be a good 
a good fit?

At the end, formulate a 5-sentence ideal customer profile in 
plain text.

Here is the data: [INSERT LIST]

The result is deliberately not yet a finished profile, but a hypothesis - but one that is based on your real data rather than on the memory of your longest-serving colleague.

Step 3: Test the hypothesis against reality

A common mistake is to immediately take the AI profile at face value. It is better to have the assumptions disclosed and check them against two or three customers who were not on the initial list. Does the profile also apply to them? Only this cross-check turns a hypothesis into a reliable tool.

Conclusion: A sharp ICP is not a strategic luxury, but a prerequisite for ensuring that every subsequent hour of research and contact does not go to waste. AI makes analysis, which used to be too expensive, routine. In the next two articles, we will make the profile researchable - and define who it excludes.