A prospect asks "how much does it cost?" and gets back: "What's your name? Which area are you in? What's your budget? When are you planning to buy? Are you the decision-maker?" By the third question they close the chat and message whoever answered first. That is what qualification looks like when it is copied straight from a sales checklist, and nobody remembers there is a person on the other side with a question of their own.
This article covers how to design lead qualification for an AI agent so your sales team gets the data it needs without the prospect feeling cross-examined: which fields really matter, which ones you can infer without asking, what order to ask in, when to stop, and how to audit an existing bot against your own conversations. The rules apply to most service businesses; for a property-specific walkthrough see our piece on qualifying real estate enquiries.
Why qualification turns into an interrogation
The problem is rarely the AI. It is the brief. A manager takes every field on the CRM record, turns each into a question and tells the bot to "collect everything before handing over". The result is a form stretched across ten messages.
Signs your flow has already crossed the line:
- the bot asks a question before answering the prospect's question;
- a single message contains two or three questions;
- it asks for things the prospect already wrote, or that the channel already shows (a WhatsApp name, a phone number from the form);
- the questions always come in the same order, whatever the person says;
- it is unclear why a question is asked and what the answer would change.
If two or more of these apply, rewording will not fix it. The set of questions itself needs to change.
What your sales team actually needs
Qualification answers one question for the salesperson: "Is this worth my time right now, and where do I start?" Three or four fields are almost always enough. Everything else can wait until there is a live conversation and the prospect has a reason to share.
| Field | Why sales needs it | Can you get it without asking? |
|---|---|---|
| Need (product, problem) | Is this a fit, and who should take it | Often: from the first message, the landing page or the ad they came from |
| Timing | Separate "this week" from "just looking" | Sometimes: "urgent", "before we move in", "for next term" |
| Scope or budget | Pitch the right option, skip the wrong ones | Rarely directly; indirectly from the plan, quantity or size they mention |
| Preferred contact method | Don't call someone who asked to text | Yes: the channel they are writing in is the answer |
Questions such as "who signs off on this?" or "who else are you comparing?" are useful, but not at first contact. They belong to the salesperson's conversation, or to the agent only if the prospect raises them.
Seven rules for an AI agent's questions
- Answer first, then ask. If an exact answer depends on details, give the range of what it depends on: "It mostly depends on floor area. Roughly how large is the space?"
- One question per message. Ask two and people answer the last one.
- Say why you are asking. "So I don't suggest something that won't fit, could you tell me…" lands very differently from a bare "Budget?".
- Offer options instead of open questions. "Within the next two weeks, or later?" is easier to answer on a phone than "When are you planning?".
- Never ask what you already know. The agent should read the first message, the lead source and the chat history, then ask only about empty fields.
- Budget last, and gently. By then the prospect has received something useful and understands the question. Ranges, or a question about scope, often work better than asking for a figure.
- Set a question budget. Decide in advance how many follow-ups the agent may ask before handing over. If the data isn't complete by then, hand over anyway and flag what is missing.
Example: the same enquiry, two ways
A hypothetical air-conditioning installer; the prospect messages on WhatsApp in the evening.
The interrogation
Prospect: Hi, how much is it to install an AC unit?
Bot: Hello! May I have your name? Please share your area and phone number.
Prospect: Sam, Marina. So how much?
Bot: Thanks, Sam! What's your budget for installation? When are you planning?
The conversation
Prospect: Hi, how much is it to install an AC unit?
Bot: Hi! It mainly depends on the pipe run and the floor: a standard install is one price, longer runs or work at height cost more. Do you already have the unit, or do you need help choosing one?
Prospect: Already bought it, just need it fitted.
Bot: Got it. Which floor and area? That lets the technician tell straight away whether access equipment is needed and quote an exact price.
Prospect: 7th floor, Marina.
Bot: Thanks. Is this week convenient, or later? I'll pass this to our team and they'll reply here with the price and available slots.
Both versions ask three questions. In the second, each one follows an answer, comes with a reason and is easy to reply to. The name comes from the profile; the phone number is never requested because the reply goes to the same chat. The bot does not invent a price. A person quotes it from the price list.
When to stop and hand over
| Situation | What the AI does | What stays with a person |
|---|---|---|
| Prospect is ready: wants a visit, a quote or an invoice | Hands over immediately, without waiting for other fields | Get in touch fast |
| Prospect sidestepped a question twice | Stops asking, answers helpfully, flags the gap | Clarify in a live conversation |
| Unusual request, discount, complaint | Summarises and hands over | The decision and the reply |
| Prospect is annoyed or asks for a human | Apologises and hands over with no further questions | The conversation |
The handover should be a short summary, not the full transcript: need, timing, scope, contact channel and what is still unknown. Good qualification is wasted if the lead then sits for half a day, which is why first response time matters just as much.
Auditing your bot: a checklist
Pull the last 20–30 conversations where the prospect went quiet and check:
- After which bot message did they stop replying? If it is usually the same question, rewrite it, move it or drop it.
- Did the prospect get an answer to their first question before being asked anything?
- Were there messages with two or more questions?
- Did the bot ask for something already in the first message or the form?
- Were there cases where the prospect was ready and the bot kept asking?
- Does the handover summary flag missing data, or are empty fields simply empty?
If you serve customers in more than one language, check each one separately. A question translated word for word sounds as robotic as one written in corporate jargon.
FAQ
Do we have to ask about budget at all?
Often not. A question about scope (size, quantity, volume) usually gives sales enough, with a note that money wasn't discussed. If you can't recommend anything without it, ask at the end and offer ranges.
How many questions is too many?
There is no universal number; it depends on the market and the channel. Your own conversations are the benchmark: if people drop off after a particular question, you have passed the limit. Two or three follow-ups are usually enough for a first contact in a messaging app.
Won't the AI ask questions at the wrong moment?
It can if you give it a rigid script. Describe the goal instead: which fields are needed and why, not the order of questions. The agent can then skip what it already knows. Spot-check conversations for the first few weeks regardless.
What about leads that don't qualify?
Don't delete them. Record the reason in your CRM ("not now", "not a fit", "no reply") and decide who to follow up with later.
To see how an agent like this would fit into your enquiry flow, visit our AI agents page or browse more articles in Sales. Cost depends on scope and data, and the first conversation is there to work it out: get in touch.