How to qualify leads with an AI SDR: BANT and MEDDIC prompts

Qualify with an AI SDR by turning a framework into one or two questions per conversation, asked in the prospect's own context, not as a form. BANT (budget, authority, need, timeline) suits short cycles, and MEDDIC (metrics, economic buyer, decision criteria, decision process, identified pain, champion) suits longer ones. The agent asks and records. A person judges, and handles price and contract.

Qualification frameworks were written for people who sit across a table from a buyer for an hour. An AI SDR works in short messages over days. The framework still helps, but you have to translate it from a checklist into a conversation.

What are BANT and MEDDIC?

BANT has four letters:

• Budget. Can they pay for it?
• Authority. Are they the person who decides, or can they reach that person?
• Need. Do they have the problem you solve?
• Timeline. Is there a reason to act soon?

MEDDIC has six:

• Metrics. What number do they want to move?
• Economic buyer. Who holds the budget?
• Decision criteria. What will they compare?
• Decision process. What steps and people stand between a yes and a signature?
• Identify pain. What hurts now?
• Champion. Who inside will argue for you?

BANT is quick and works when the deal is small and one person decides. MEDDIC is slower and fits a purchase with several people and a process. Use the one that matches the way your customers buy, and do not use both.

Why not ask everything at once?

Because a person who replies to a cold message has given you a moment of attention, not an interview. Our data supports a light touch: 80% of conversions on our own LinkedIn account came from a reply to the first message, and the booking link typically went out after three prospect messages. The conversation is short, so every question has to earn its place.

The rule we use: ask one question per message, make it relevant to what the person just said, and never ask a question whose answer you already have in the lead data.

How do you turn a framework into prompts?

Write each letter as an instruction to the agent, with when to ask and when not to.

Need (BANT) or Identify pain (MEDDIC): "Ask what the person does today about [problem]. Ask it in their words, after they have replied once. Do not describe our product before they answer."

Timeline: "If the person shows interest, ask whether this is something they are looking at this quarter or later. Accept 'later' without pushing."

Authority or Economic buyer: "If the person is not the one who decides, ask who is involved in choosing a tool like this. Ask for a name or a role, never for a forward."

Budget: "Do not ask about budget. If the person raises price, say what the published plans are and hand the conversation to a person for anything beyond them."

Decision process: "Ask only after a clear yes to a call: 'Is anyone else joining, so I can send them the same details?'"

Champion: "Never ask directly. Watch for a person who offers to forward the thread, and record it."

Put these in the persona or the rules the agent follows. The help page on writing rules your agent follows shows how, and the goals library holds ready qualify and convert goals to start from.

What does qualifying by asking look like in practice?

The best example we have is not a sales campaign. In the Shelly installer programme, the agent asked four numbered questions in the body of an email, in the installer's language, with nothing to click. 360 installers replied, 225 answered the survey and 172 of the 210 complete surveys came back in one message. From the answers, 64 big project installers were surfaced for a person to follow up.

The questions were few, numbered and easy to answer in one reply, and a person at Shelly took over pricing, support, partnership topics and the big projects. That is the division of labour to copy: the agent asks and sorts, the person handles what costs money. Read the case study and the post on asking 3,940 installers four questions.

The Hotel Pirin reservations agent shows qualification in a different form: it asked for the dates and the party size before it quoted. Where those were missing, the first reply asked for them, and once it had them the agent quoted from the rate card. Dates and headcount are the hotel's version of timeline and need.

What should stay with a person?

Four things.

1. Budget and price beyond the published plans. Let a person decide how to answer.
2. Contract and legal terms. The agent should not discuss them.
3. Judging a complex fit. In the Lamothermic study engineers asked what the foundry could cast. The agent could only answer precisely after the president wrote down what the company does and does not pour. When the question goes beyond the written facts, the agent should stop.
4. Anything where a wrong yes costs money. Write it down as a rule.

Sellinger's agent stops on its own when it cannot answer, for example a price it has no answer for, and a person can take over any conversation. The post on what an AI SDR does after the first reply describes the flow.

How do you record qualification?

Keep it simple: a few fields on the lead, filled from the conversation. For example, "need confirmed", "timing: this quarter", "decision maker: yes or no", "other people involved". The point is that a person opening the thread before a call sees the answers without reading everything.

Review a sample each week. Read twenty qualified threads and ask: did the agent record what the person said, or what it hoped? Fix the prompts where it hoped.

What do the first twenty threads teach you?

Read them with three questions. First, did the agent ask a question the person had already answered? That means the prompt needs a rule about reading the thread. Second, did it ask something that sounded like a form? Rewrite the prompt so the question follows from the last reply. Third, did it push after a no? A refusal should close the thread.

Our own data shows why the early messages matter. Follow ups added 45% more replies on our own account, from 26.5% at the first message alone to 38.4%, but the share of first replies that were negative rose from 0.9% at the first message to 8.5% at follow up 1. Each extra question is a touch with a cost, so ask only what you will use.

A checklist

• [ ] I picked BANT or MEDDIC, not both.
• [ ] Each letter is written as an instruction with when to ask and when not to.
• [ ] The agent asks one question per message, tied to the last reply.
• [ ] Budget, price beyond the plans and contract go to a person.
• [ ] Answers are recorded in fields a person can read before the call.
• [ ] I review twenty qualified threads every week.