Designing Discovery Questions for AI Lead Qualification

Discovery questions are the core of qualification — the questions that surface whether a caller is a fit, how urgent they are, and what they need. In traditional sales, SDRs and AEs spend years developing intuition for which questions to ask in which order. For voice AI, that intuition has to be encoded in prompts and rubrics — and getting it right is the difference between AI qualification that generates pipeline and AI qualification that wastes calls.

TL;DR

• Good discovery questions open the caller up, don't close them down.
• Order matters: warm questions first, sensitive questions later, optional when possible.
• Use open-ended questions for rich signal; closed-ended for specific data.
• The LLM can rephrase; you own the intent, not the wording.
• Test questions against real calls; iterate.

The question archetypes

Open-ended. "What brought you to us today?" "Tell me about the problem you're solving."

Closed-ended / specific. "Are you the decision-maker?" "What's your timeline?"

Situational. "How is the team handling this today?" "What's your team size roughly?"

Consequences. "What happens if this doesn't get solved?" "What does success look like?"

Use a mix. Open for texture, closed for structure.

The opening — warmth first

Bad: "What is your budget?"

Better: "What brought you to us today?"

Start warm. Capture context. Specific questions come after trust is built. Even 30 seconds of rapport dramatically changes disclosure rates.

The sequence

Roughly:

1. Intent. "What brought you to us?"
2. Pain. "What's driving that?"
3. Current state. "How is your team handling it today?"
4. Fit. "Tell me about your company size / tech stack / use case."
5. Authority. "Who else is involved in this?"
6. Timeline. "What's your rough timeline?"
7. Budget. Soft probe, often late or implicit.
8. Next step. Book meeting, send resources, or polite exit.

This arc moves from general to specific, from easy to harder questions.

Open vs closed strategy

• Open to surface qualitative signals (pain, use case, motivation).
• Closed for specific data that fits structured fields (size, timeline, role).

Example:

• Open: "Tell me about what you're trying to solve."
• Closed: "Are you the decision-maker, or is this a team decision?"

LLM rephrasing

Don't script word-for-word. Give the LLM the intent:

System prompt:
"Ask about timeline. Phrase it naturally — 'when are you 
hoping to get this going?' or 'how soon do you need 
something in place?' or similar. Don't ask 'what is 
your timeline' verbatim — feels robotic."

LLMs rephrase well when given the goal.

Sensitive questions

Budget and authority are the sensitive ones. Handle carefully:

• Don't lead with budget. Triggers defensive posture.
• Ask about scale as a proxy. "Roughly what size is your team?"
• Listen for signals. Caller says "we're a Series B startup" — informs budget posture without asking.
• Ask about process, not person. "Who else is involved in choosing something like this?" is better than "Are you the decision-maker?"

Dynamic branching

Not every question applies to every caller:

• Skip "company size" if it's a consumer.
• Skip "timeline" if they clearly said "I'm just exploring."
• Skip "authority" if they self-identify as founder/CEO.

LLMs can handle this natively with a good system prompt.

Follow-up probing

When the caller gives a short answer, probe for depth:

Caller: "We need better reporting."

Agent: "Can you tell me more about what's missing in 
your current setup?"

Caller: "We're using a bunch of spreadsheets and it's 
slow to get insights."

Probing turns shallow answers into qualified signal.

Avoiding interrogation feel

Balance questions with reflection:

Agent: "Got it — reporting is the main pain, team of 
about 40 analysts, and you want something in place by 
Q3. That sounds like a solid fit for our platform. 
Any specific integrations you're hoping for?"

Summarize back. Let the caller feel heard, not just interrogated.

What NOT to ask

• "Is this a real project?" (Insulting.)
• "Can you afford it?" (Insulting.)
• Anything protected (age, gender, ethnicity).
• Anything you could infer from context.
• 15 structured questions in a row.

Testing questions

• Sample real calls. Which questions produce rich answers?
• A/B test phrasings. "What brought you" vs "How can I help."
• Measure drop-off. If callers bail at question 4, move that question later.
• Iterate monthly.

See how to A/B test voice agent prompts.

Question templates by vertical

Enterprise SaaS:

• "What drove this evaluation?"
• "What tools are you currently using?"
• "Who's the executive sponsor?"
• "What's the timeline for a decision?"

Consumer services (home services, fitness, etc.):

• "What are you looking for?"
• "How soon would you like to get started?"
• "Where are you located?"

Healthcare / regulated:

• "What brought you to us?"
• "Are there specific compliance needs?"
• "How urgent is this for you?"

Financial services:

• "What are you looking to set up?"
• "Are you looking for personal or business?"

Customize for your domain.

Measuring question quality

• Answer richness. How often do callers give detail vs single-word?
• Qualification completion. % of callers who provide enough to qualify.
• Call length. Should be 3–5 minutes for qualification.
• AE feedback. "Was the briefing useful?"

Related reading

• Inbound Lead Qualification with Voice Agents
• Inbound Voice for Trade Shows and Events
• How AI Agents Should Handle Pricing Questions on Inbound Calls
• Lead Qualification for High-Volume Marketing Channels
• How AI Agents Handle "Send Me an Email Instead"

FAQ

How many discovery questions is too many? Past 7 structured questions, callers bail. Aim for 5–7.

What if the caller doesn't want to answer? Respect it. Move on. Qualification can be partial.

Can we ask for specific metrics (e.g., revenue)? Usually avoid. Infer from context or company.

How do we handle talkative callers? Let them talk — rich signal. Redirect if off-topic: "Got it — can I ask about your timeline?"

Should discovery questions differ for inbound vs outbound? Yes — inbound has self-selected interest; outbound needs to earn engagement first.