The Voice AI Discovery Call: 21 Questions I Ask Before Quoting an AI Receptionist (2026)

TL;DR: A voice AI discovery call has one job: leave with enough facts to scope, price and prove the agent. That means five things written down before you hang up: how many calls the business gets and misses, what one customer is worth, which call types the agent should handle, which systems it has to touch, and who signs off. I build production voice agents on Retell AI for US clients, wired into n8n, GoHighLevel and Twilio, and we run VoiceDash, the white-label client portal agencies give their clients. Below are the 21 questions I actually ask, grouped in the order I ask them, plus the red flags that tell me to walk away.

Most agencies treat the discovery call as a warm up for the demo. They ask "so, how many calls do you get?", hear "a lot", and jump straight into showing an agent. Then they write a proposal with no numbers in it, quote a price with nothing to anchor it against, and wonder why the prospect goes quiet.

The discovery call is where the deal is actually won. The demo and the proposal only repeat back what you learned here.

Before the call: what to know in advance

Spend ten minutes on the business before you dial. Call their main line at a busy hour and after hours, and note what happens: rings out, voicemail, a human who sounds rushed, a phone tree. Check their website for booking links, hours and service area. Look at their Google reviews for anything about "never answers the phone" or "couldn't get through."

It lets you say "I called on Tuesday evening and got a full voicemail box," which beats any slide.

Part 1: The phone today (questions 1 to 5)

These questions establish the problem in the owner's own words. Do not pitch anything yet.

1. Walk me through what happens when someone calls your main number right now. Who picks up, from which device, and what happens if they do not?
2. Roughly how many calls do you get in a normal week? Owners often guess. That is fine. Ask whether their phone provider or CRM can export a call log, because a real number beats a guess.
3. How many of those do you think go unanswered? Then ask when: lunch, after 5pm, weekends, while the team is on a job or with a patient.
4. What do callers usually want? Get the top five reasons in order. This becomes your list of call types.
5. What happens to a voicemail? If the answer is "someone checks it the next morning," you have already found the cost.

Part 2: The value of a call (questions 6 to 9)

This is the part most agencies skip, and it is the part that sets your price.

6. What is a new customer worth to you, on average? Ask for first job value and lifetime value if they know it.
7. Of the people who call and actually talk to someone, how many book? A rough close rate is enough.
8. When a caller hits voicemail, how many do you think call back? Most owners know the honest answer is "not many."
9. Have you tried to fix this before? An answering service, a receptionist, a missed call text back tool. What did it cost and why did they stop?

Write the answers down exactly as they say them. If an owner tells you they miss "maybe 30 calls a week" and a new customer is "worth around $500," those two numbers carry the whole proposal. I walk through turning them into a monthly figure in voice AI ROI, and how that figure anchors the price in voice AI agency pricing.

Question 9 matters more than it looks. If they cancelled an answering service because it got details wrong, accuracy is the thing you sell. If they cancelled because it was expensive, value is. The comparison I use is in AI receptionist vs answering service.

Part 3: Scope (questions 10 to 15)

Now you define what the agent will and will not do. This is where you protect yourself from a project that grows forever.

10. If the agent could only do two things well, what would they be? Usually: book appointments and answer the same five questions.
11. What should never be said on the phone? Prices that change, medical or legal advice, guarantees, anything the owner wants to quote personally.
12. When should a caller reach a real person? Get the exact triggers: emergencies, angry callers, existing customers with an open job, anyone who asks. Then ask who receives the transfer and during which hours. The patterns I use are in voice AI call transfer to a human.
13. Should the agent answer every call, or only the ones your team misses? Overflow and after hours is an easier first step than replacing the front desk.
14. Do you get calls in other languages? If a meaningful share of callers speak Spanish, that changes the build. See bilingual AI receptionists.
15. Do you also want outbound calls? Reminders, follow ups, reactivation. Usually I park this for phase two and note it as an upsell.

Turn the answers into a short list of call types in the client's vocabulary. "Books new patient cleanings" is a scope line. "Intent routing" is not. That list goes straight into the proposal, as I describe in writing a voice AI agency proposal.

Part 4: Systems (questions 16 to 18)

The integration is where builds slip, so find out exactly what you are connecting to.

16. Where do bookings live today? Google Calendar, GoHighLevel, a practice management system, a field service tool, or a paper diary. Ask whether it has an API or whether staff will need to confirm bookings manually.
17. Where should call details go after the call? A CRM, a text to the owner, an email, a Slack channel. Ask who reads it.
18. What phone setup do you have, and can you forward or port the number? Carrier, VoIP provider, whether they want to keep the existing number. My setup notes are in voice AI phone number setup.

If the booking system has no way in, I say so on the call and scope a "collect details and text the team" flow instead of promising live booking. Being honest here saves the relationship later. The booking side is covered in voice AI appointment booking.

Part 5: Decision and proof (questions 19 to 21)

19. Who else needs to say yes? A partner, an office manager, a compliance person. If they are not on the next call, the deal stalls.
20. What would make you say this worked after 30 days? Get a concrete answer: a number of booked appointments, no more full voicemail box, fewer callbacks. This becomes your pilot success criteria, structured the way I describe in the voice AI pilot program.
21. How would you want to check what the agent is saying to your customers? This is my favourite question, because almost every owner says some version of "I'd want to listen to the calls."

Why question 21 sells the portal

That last answer is the moment I mention reporting. Owners are nervous about letting software talk to their customers. The fix is not reassurance, it is visibility: every call recorded, transcribed and summarised in a place they can open any time.

That is why we built VoiceDash. You connect your Retell account, create the client, and they get a portal with your logo on your domain, showing only their own calls, recordings, transcripts and usage. They never see Retell or any other client. You can add their office manager or partner as team members, and bill a monthly fee, per minute or per call through your own Stripe account, with VoiceDash taking no cut. It is live in under 10 minutes with no code, on Starter, Growth or Ultimate plans from $19/mo with a 7-day free trial. VAPI and Bland support are coming soon.

Saying "you will be able to hear every call from day one, in a dashboard with my branding" turns the trust objection into a selling point. What I put in front of clients each month is in voice AI client reporting.

Red flags that tell me not to take the deal

Not every discovery call should end in a proposal. I pass, or at least slow down, when I hear:

• "We answer every call already." There is no missed call cost to recover, so the agent becomes a nice-to-have.
• "Just make it do everything." No scope means no finish line. Push for the two jobs that matter.
• "Can it promise callers a price?" If the business needs exact quotes on the phone and cannot give you a fixed price list, the agent will either refuse constantly or get it wrong.
• "I don't want to be involved." You need an approved answer sheet and someone to review test calls. An owner who will not give an hour before launch will blame you for every bad call after it.
• No decision maker on the call and no path to one.

What to send after the call

Within 24 hours, send a short recap email that repeats their numbers back to them: calls per week, missed calls, value per customer, the two call types, the systems, and the 30 day success criteria. Ask them to correct anything wrong.

It proves you listened and gets written agreement on the facts your price depends on. Then build a short demo on their actual call types, which I cover in how to build a voice AI demo, and send the proposal.

The discovery call checklist

1. Called their line before the meeting and noted what happened
2. Weekly call volume, ideally from a real log
3. Missed calls, and when they happen
4. Top five reasons people call
5. Value of a new customer and rough close rate
6. What they tried before and why it stopped
7. The two jobs the agent must do well
8. What the agent must never say
9. Transfer triggers, recipient and hours
10. Booking system, CRM and phone setup confirmed
11. Every decision maker identified
12. 30 day success criteria agreed in their words
13. How they want to review calls (your cue for the portal)
14. Recap email sent within 24 hours

The bottom line

A good voice AI discovery call is mostly listening. Get the call volume, the missed calls, the value of a customer, the scope, the systems and the success criteria, and the proposal almost writes itself. Skip it and you end up competing on price against every other agency with a demo. If you want the bigger picture of the sales process this sits inside, start with how to sell AI receptionists.

Want to show prospects on the discovery call exactly where they will hear every call? Start free on VoiceDash or book a demo and I will walk you through the portal I give every client.