Gains:
- Ability to clarify the ideal customer profile (ICP) and buyer persona with the help of AI
- Ability to qualify leads in a structured way with frameworks such as BANT and MEDDIC
- Ability to apply the discipline of verifying company/person information produced by AI against the source
The most expensive mistake in sales is talking to the wrong person. A representative spends weeks explaining the right product to the wrong customer at the right time and eventually receives the answer "there is no budget" or "I am not the one who decides." Yet this information can largely be extracted before first contact. That's exactly what lead research and qualification does: distinguish early on who could actually be your customer and who is a waste of time. In this unit you will learn how to use AI as the engine of this process; We will see step by step, from creating the Ideal Customer Profile (ICP) and buyer personas to scoring with frameworks such as BANT and MEDDIC. Let's put a warning at the beginning: Company and person information produced by AI is never considered "real" without being verified.
Ideal Customer Profile (ICP) and Buyer Persona
ICP is the answer to the question "which type of company is the ideal customer for us?" Persona is the person you talk to within that company. They are two different things: ICP defines the company (firmographic), persona defines the person.
concept
What defines
Example signals
ICP (firmographic)
Ideal company
Sector, number of employees, turnover, geography
ICP (technographic)
Technology used by the company
CRM, infrastructure, software they use
Persona
Person in the decision process
Title, goals, pain points, KPIs
Firmographic signals answer the question "Is this company suitable for us in terms of size?" Technographic signals answer the question "Are they already using a technology that complements/conflicts with us?" For example, companies that use a certain e-commerce infrastructure may be a ready ground for your integration.
ICP definition prompt: "Below are the characteristics of the top 10 customers we have acquired in the last 12 months. Create an Ideal Customer Profile (ICP) by finding COMMON patterns in them. Make a table with the following dimensions: industry, headcount range, annual turnover range, geography, technologies they use, typical pain point. Only rely on patterns in the data, do not add assumptions outside the data. Top customers: {{customer_list}}"
Persona extraction prompt: "We sell {{product}}. Our ICP is {{icp_summary}}. Define 3 personas that influence the purchasing decision in these companies. For each persona:- Title- Primary goals and KPIs by which they are measured- Top 2 pain points- Value proposition of our product translated into their language- 2 points that may object to purchasing"
Qualification Frameworks: BANT and MEDDIC
We measure whether a lead is "real" with structured questions. Two common frameworks:
TAPE — for fast, lightweight characterization:
- Budget: Are there resources to pay?
- Authority: Can the person I am talking to make decisions?
- Need: Is there a real, urgent need?
- Timeline: When do they plan to receive it?
MEDDIC — deeper for complex, large B2B sales:
- Metrics: Numerical benefit measured by the customer.
- Economic Buyer: The person who actually controls the budget.
- Decision Criteria: Decision criteria.
- Decision Process: Decision process and steps.
- Identify Pain: Concrete pain.
- Champion: The person who defends you inside.
Frame
when
depth
TAPE
Short cycle, low-medium value
quick elimination
MEDDIC
Long cycle, high value, multi-stakeholder
deep, systematic
Tip: For small, quick sales, start with TAPE; Switch to MEDDIC as the deal grows and the number of stakeholders increases. Tell the AI clearly which framework you want to score with, or it may confuse the two.
Research Summary and Attribution Scoring with AI
We use AI for two jobs here: (1) summarize the raw information you have by frame, (2) suggest a score. The score is a "preliminary ranking" for you, not the final decision.
Lead qualification prompt (BANT): "Evaluate the following lead notes with the BANT framework. Give 0-5 points for each dimension and write your justification. If information is missing, say 'no data', DO NOT GUESS. Probe: total score, 'hot/warm/cold' label and next step suggestion. Lead notes:- Company: {{company}}- Interviewee/title: {{person}}- Notes: {{interview_notes}}"
Lead list prioritization prompt: "Sort the following 20 leads according to their suitability to our ICP ({{icp_summary}}). For each lead: suitability score (0-100), strongest fit signal, greatest uncertainty. For those with low scores, briefly write the reason. If company data is missing, mark FITTING, 'must be verified'. Lead list: {{lead_list}}"
Weak Prompt / Strong Prompt
WEAK: "Is this lead good?" (Result: a flimsy, one-word comment.) STRONG: "Score this lead using the BAND framework. Give 0-5 to each dimension, write a justification, mark the missing information as 'no data' and suggest the next step based on the total score. Don't go beyond your notes."
Risk of Fabrication: Source Verification
AI can credibly fabricate information it doesn't know about a company or person: turnover, number of employees, title, even the name of the decision maker. If you come into contact with this information, you will lose your credibility in one fell swoop.
Caution: Verify any firm-specific data the AI returns (turnover, growth, executive name, tech stack) from the primary source: firm's own site, LinkedIn, official newsletters. Do not use unverified information in outreach.
A simple verification workflow:
- Ask the AI to “put a source tag on each claim” for the lead summary.
- Check each item marked "Must be verified" from the source one by one.
- Process the verified information into the CRM; Delete what cannot be verified.
- Only update scoring with verified information.
- Only then move on to outreach.
Mini Case
B2B SaaS representative Kaan gives a lead list of 50 people to AI and has it sorted according to ICP suitability. “Firm A reached 200 employees last quarter,” AI says. Kaan opens the company on LinkedIn; The number of employees is around 60. AI exaggerated. Kaan marks this signal as "must be verified", enters the real number and deducts the score. In return, it confirms the AI's "Company B is opening a new market soon" signal with new postings on the company's career page and prioritizes this lead. Result: AI ranked 50 leads in minutes, Kaan validated the list and focused on the 8 really hot leads.
Common Mistakes
- Starting to investigate without identifying the ICP (targeting everyone, shooting no one).
- Processing the company data provided by AI into the CRM without verifying it.
- Mixing BANT and MEDDIC and doing half scoring.
- Allowing the AI to fill in the missing information with its “guess”.
- Thinking that the score is absolute truth and disabling human judgment.
- Focusing only on the company instead of the persona and skipping "who am I talking to".
Tip: Always justify the AI score by asking “why did you score it that way?” If the justification is weak, the score is also weak. The rationale also indicates the points you need to verify.
In summary
- ICP describes the firm (firmographic + technographic), persona describes the person; Both are necessary.
- BANT is for quick qualifying, MEDDIC is for deep and complex sales.
- AI summarizes research and suggests scores; The final decision belongs to the person.
- The score is a preliminary ranking and must be requested with justification.
- Company/person information produced by AI cannot be used without verification from the primary source.
- Missing information should be marked as "no data" rather than "estimate".
Application task
Make a list of your last 5-10 acquired customers and have the AI find common patterns with the ICP prompt above. Prioritize the 10 leads you have using the resulting ICP. Then verify each company-specific claim the AI makes on LinkedIn and the company site for the 3 highest-scoring leads; Note which claims turn out to be false or incomplete. This exercise teaches both ICP discipline and verification reflex at the same time.