Identify your ideal client with AI: method and prompts

🎯 Introduction: without an ideal customer, AI amplifies the noise

90% of outbound campaigns fail for a single reason: the target is too broad. "SMEs in France", "marketing directors", "SaaS startups"... These are not targets, they are oceans. And when you give a vague target to AI, it generates vagueness at scale — thousands of leads that will never convert.

The alternative? Build a precise ICP (Ideal Customer Profile), then use AI to identify the companies and people that match exactly. A well-defined ICP transforms a 2% conversion rate into 15-25%.

In this article, we look at how to build your ICP with AI in 4 steps, with ready-to-use prompts and concrete tools to go from theory to a list of prospects.

📋 Table of Contents

1. What is an ICP (and why it's crucial)
2. The 4 components of an ICP in 2026
3. 5 AI prompts to build your ICP
4. Identify intent signals with AI
5. Transform the ICP into a prospect list
6. Score and prioritize with AI
7. Common mistakes
8. The key takeaways
9. Recommended tools
10. FAQ

1. What is an ICP (and why it's crucial)

Definition

The ICP (Ideal Customer Profile) is a precise and data-driven description of your ideal customer. Not a marketing persona with a name and a stock photo — an operational description that guides your prospecting actions.

ICP vs Persona: the difference

• ICP: description of the ideal company (industry, size, revenue, tech stack)
• Persona: description of the ideal person (role, motivations, frustrations, buying journey)
• You need both: the ICP to target the right companies, the persona to talk to the right people

Impact on conversion

💡 Pareto's Law applied: 20% of your customers generate 80% of your revenue. Your ICP is the description of those 20%.

2. The 4 components of an ICP in 2026

In 2026, an ICP based solely on firmographics (industry + size) is no longer enough. Here are the 4 components of a modern ICP:

1. Firmographics

The company's characteristics:
- Industry (NAF code or SIC code)
- Size: headcount, revenue, number of locations
- Location: country, region, city
- Status: startup, SME, mid-sized enterprise, large enterprise
- Company age

2. Technographics

The tools and technologies used:
- Tech stack (for B2B SaaS)
- Marketing tools (HubSpot, Mailchimp, Salesforce...)
- Cloud infrastructure (AWS, GCP, Azure)
- CMS (WordPress, Shopify, custom...)

💡 Why this matters: if you sell an AWS → GCP migration tool, targeting companies on AWS is 10x more effective than targeting "tech SMEs".

3. Buying behavior

The signals that indicate a need:
- Dedicated budget for your product category
- Buying process (who decides, how long it takes)
- Purchasing frequency (subscription, one-off, renewal)
- Price sensitivity vs. value sensitivity

4. Intent signals

The behaviors that indicate they are currently in-market:
- Hiring in a key department
- Recently raised funding
- Blog article or LinkedIn post on a topic related to your offer
- Visit to your website (via tools like Clearbit, 6sense)
- Searching for keywords related to your solution

⚠️ Intent signal is the most powerful component in 2026. A company hiring a "Head of Data" is likely building a data infrastructure — that's the best time to reach out if you sell data services.

3. 5 AI prompts to build your ICP

Here are 5 prompts to use in Claude or your favorite AI assistant to build your ICP step by step.

Prompt 1: Analyze your best existing customers

I am going to give you the list of my 10 best customers (those who generate the most revenue and are the most satisfied). For each one, I will give you: industry, size (headcount and revenue), contact's job title, problem solved, price paid.

Based on this data, identify the common patterns and propose an ICP in 3 parts:
1. Firmographics (mandatory criteria + nice-to-have)
2. Decision-maker persona (job title, motivations, frustrations)
3. Intent signals to watch for

My customers:
- Customer A: B2B SaaS, 45 employees, €3M revenue, CTO, cloud migration problem, €15,000/month
- Customer B: FinTech, 120 employees, €12M revenue, Head of Data, broken data pipeline, €25,000/month
- Customer C: B2B SaaS, 30 employees, €2M revenue, CEO, lack of visibility on metrics, €8,000/month

Why it works: your best existing customers are the best source of data for your ICP. AI identifies patterns you don't see.

Prompt 2: Define qualification criteria

My offer: implementation of complete data infrastructures (ETL pipeline, data warehouse, dashboards)
Price: starting at €8,000/month
Sales cycle: 3 to 6 weeks

Define the BANT criteria (Budget, Authority, Need, Timeline) to qualify a prospect.
For each criterion, provide:
- The question to ask
- The ideal answer (green flag)
- The acceptable answer (yellow flag)
- The disqualifying answer (red flag)

Why it works: the BANT framework is classic but AI adapts it precisely to your offer, instead of remaining generic.

Prompt 3: Identify specific intent signals

For a company selling AI automation services to B2B SaaS SMBs of 20-200 employees, what are the 10 most predictive intent signals of an imminent purchase?

Rank them by relevance (1 = strongest) and for each one:
- Where to detect it (LinkedIn, website, directory, press...)
- How to verify it automatically
- Recommended action timeframe (upon detection, within 24h, within 1 week)

Why it works: AI cross-references thousands of B2B sales patterns to identify the signals most correlated with a purchase.

Prompt 4: Create the decision-maker persona

Based on the following ICP: B2B SaaS SMBs of 20-200 employees in the Paris region, using HubSpot and AWS, with a revenue of €2M to €15M.

Create a detailed persona of the decision-maker including:
- Typical profile (job title, age, career path)
- Their 3 biggest frustrations at work
- What they read (blogs, newsletters, LinkedIn)
- How they make purchasing decisions
- The most common objections and how to address them
- A "day in the life" of this persona

Why it works: a detailed persona allows you to write messages that truly resonate, instead of speaking to a generic profile.

Prompt 5: Validate and refine the ICP

Here is my current ICP: B2B SaaS SMBs of 20-200 employees in the Paris region, €2-15M revenue, using HubSpot and AWS.
Here are my 5 last lost prospects:
- 12-person startup, no in-house data engineer
- 80-employee e-commerce company, Magento + OVH stack
- 300-employee SaaS company, purchasing process via a 5-person committee
- 25-employee web agency, maximum budget €2,000/month
- 50-employee FinTech, one-off need with no recurrence

Analyze why these prospects didn't convert and propose:
1. Which criteria in my ICP were poorly calibrated
2. Which criteria were missing
3. The revised ICP in version 2

Why it works: an ICP is not set in stone. Analyzing losses is the best way to continuously refine it.

4. Identifying intent signals with AI

Sources of intent signals

Automating detection with Clay.com

Clay.com is the most powerful tool for automating intent signal detection. Specifically, you import a list of target accounts (1,000+ companies), and the platform performs automatic enrichment by retrieving the company's latest LinkedIn posts, recent hiring, funding via Crunchbase, tech stack via BuiltWith, and recent news. Next, an AI engine analyzes this enriched data to assign an intent score of 1 to 100 to each company, based on the presence of your pre-defined intent signals. You then simply filter the "hot leads" (score > 70) to prioritize your prospecting towards companies currently in the market.

Scoring prompt for Clay

You are a sales analyst. Evaluate the purchase intent score (1-100) for this company.

Data:
- Company: DataFlow
- Sector: B2B SaaS
- Headcount: 65
- Recent hires: 1 Data Engineer, 1 Product Manager
- Latest news: Launch of a new AI feature
- Tech stack: AWS, HubSpot, Segment
- Latest LinkedIn post: "We are hiring our first Head of Data to structure our data"

Signals to look for:
- Recent data hiring
- Mention of data/AI projects
- Use of tools compatible with our offer

Respond in JSON: {"score": int, "signals_found": [str], "recommendation": str}

"Hot" threshold = 70+

5. Turn the ICP into a prospect list

Once the ICP is defined, you need to find the matching companies and people.

Step 1: list target accounts

Use your ICP as a filter on the following sources:

• LinkedIn Sales Navigator: advanced filters (industry, size, location, growth)
• Apollo.io: database of 275M+ contacts with detailed filters
• Crunchbase: for startups (funding, growth stage)
• Google Maps (via Apify): for local businesses

Step 2: find decision-makers

For each targeted account, identify the right people:

• Hunter.io: Domain Search → list of emails by company
• Prospeo.io: LinkedIn Scraper → profiles matching your persona
• Phantombuster: automated extraction from LinkedIn Sales Navigator

Step 3: enrich and score

Once your raw prospects are extracted, the scoring logic consists of assigning a score of 0 to 10 based on how well they match your ICP. We recommend distributing the points as follows: 4 points for firmographics (2 if the industry matches, 2 if the size is within the target range), 3 points for technographics (1 point per tool from the required tech stack, capped at 3), and 3 points for location (if the region is within your scope). By applying this framework, you get three segments: hot leads (score ≥ 7/10) to contact as a priority, warm leads (4 to 6/10) to nurture, and cold leads (< 4) to discard.

6. Score and prioritize with AI

The scoring model

A good score combines 3 dimensions:

Multi-dimensional scoring prompt

Rate this prospect on 3 dimensions (1-10 each):

Prospect: Marie Dupont, CTO at DataFlow

Target ICP: SaaS SMBs of 20-200 employees in Île-de-France, revenue 2-15M€, using HubSpot and AWS.

Available data:
- Headcount: 65
- Estimated revenue: 5M€
- Tech stack: AWS, HubSpot, Segment
- Recent hires: 1 Data Engineer
- Visited our website: yes (3 pages, including the pricing page)
- Opened our emails: 2 emails opened out of 4 sent

Respond in JSON:
{
"fit_score": int,
"intent_score": int,
"engagement_score": int,
"total_score": int,
"priority": "hot", "warm" or "cold",
"next_action": str
}

Hot = total >= 24, Warm = total >= 15, Cold = total < 15

Automating scoring with Clay

Clay.com allows you to automatically score every prospect on import. You configure your scoring model once, and every new entry is evaluated automatically. Result: your pipeline sorts itself, and you spend your time on hot leads.

7. Common mistakes

❌ Targeting too broadly

"SMEs in France" = 4 million companies. "SaaS SMBs with 20-50 employees in Île-de-France using HubSpot" = a few thousand. The more precise the ICP, the higher your conversion rate.

❌ Relying solely on your assumptions

"My ideal customer is a tech startup CEO" — really? Do your data confirm this? Always validate the ICP with data from your existing customers (see Prompt 1).

❌ Ignoring intent signals

A static ICP (firmographics alone) tells you WHO to target. Intent signals tell you WHEN to contact them. Without signals, you are contacting companies that have no current need — and you are wasting your time.

❌ Not refining the ICP regularly

Your customers evolve, your offer evolves, the market evolves. An ICP set once and for all loses its relevance in 3-6 months. Revise it quarterly.

❌ Confusing ICP and TAM (Total Addressable Market)

The ICP is not your total market. It is your priority target. You can expand later, but start with a narrow and well-defined target.

The key takeaways

• A precise ICP increases the conversion rate from 2% to 15-25% by eliminating out-of-target leads.
• In 2026, a modern ICP goes beyond firmographics: it integrates technographics, buying behavior, and intent signals.
• Your best existing customers are the best source of data to build your ICP (use Prompt 1).
• Intent signals (key hiring, funding, tech stack changes) tell you WHEN to contact a prospect.
• An ICP score out of 10 (4 pts firmographics, 3 pts technographics, 3 pts location) is enough to prioritize your prospects into hot/warm/cold.
• Revise your ICP every quarter with data from your lost prospects.

Recommended tools

FAQ

How long does it take to build a solid ICP?
With the 5 prompts in this article and your existing customer data, allow 2 to 4 hours for a first actionable ICP. Refinement is then done on an ongoing basis.

Do I need a different ICP for each offer?
Yes. If you sell AI automation services AND data engineering services, the targets probably differ. One ICP = one offer.

Can I use ChatGPT instead of Claude for these prompts?
Yes, the prompts work with any recent model. Claude is recommended for its ability to follow complex instructions, but GPT-4 or Gemini will do the job.

What should I do if I have fewer than 10 customers to use Prompt 1?
Use your 3 to 5 best customers, even if it's not many. AI is able to identify patterns from a small sample. You can also supplement with lost customers to refine the criteria.

How often should I revise my ICP?
A quarterly rhythm is recommended. Integrate the analysis of your lost prospects (Prompt 5) into your end-of-quarter routine.

✅ Conclusion

Building an accurate ICP is not an optional step — it's the foundation of any outbound prospecting that converts. AI accelerates every phase: analysis of your existing customers, identification of intent signals, and prospect scoring.

To summarize the process:
1. Prompt 1-2 : build your ICP from your best customers
2. Prompt 3-4 : identify intent signals and create the decision-maker persona
3. Prompt 5 : continuously refine with lost prospects
4. Tools : Hunter.io, Prospeo.io and Phantombuster to build your list
5. Scoring : prioritize with the 3-dimension model (Fit 40% / Intent 40% / Engagement 20%)

Next step : now that you know WHO to target and WHEN to contact them, discover how to automate lead generation and write personalized LinkedIn messages with AI. To monetize these skills, check out our guide to launching your AI freelance business in 2025, or learn how to sell AI automation services to your clients. You can also complete your offering by learning to create and sell AI prompt templates or follow our AI guide for freelancers: automating your prospecting.

Your ICP evolves with your market. Review it every quarter to stay relevant.
```