Photo by Ron Lach · CC Pexels License undefined

12 buyer questions to ask before you buy an AI receptionist

September 3, 2026 · 8 min read

A no-fluff buyer’s checklist to evaluate AI receptionists fast — booking safety, fees, voice, transcripts, storage, and real test steps.

You need a receptionist that books jobs, not one that creates messes

You’re fielding calls between jobs. You don’t have time to babysit a bot. Use this checklist to push vendors past the demo and into the details that affect your calendar, your costs, and your liability.

1) Will it double‑book? Ask for the exact booking flow and test it live

Say this on the call: “Show me your live calendar write.” You’re looking for an immediate, atomic create event — not a delayed batch job.

What good looks like (vendor must demo on your test calendar):

• AI queries live availability via the calendar API.
• AI offers slots, caller picks a time.
• AI immediately makes a create‑event call and only confirms to the caller after the write succeeds.
• If the write fails (slot taken), it restarts selection.

Why this matters: calendar “sync” can mean many things. If they poll on a schedule or do delayed writes, you’re exposed to race conditions. [Google Search Central] doesn’t cover this, but the Google Workspace Calendar API sync guide explains incremental sync and change tokens — if they rely on periodic sync rather than a confirmed create on your primary calendar, conflicts can slip through.

How to test: while the vendor is mid‑booking on your calendar, you create a conflicting event for the same slot. The AI should detect the conflict and recover without confirming the original time.

Red flags: “We confirm later by text,” “We hold the time then write it after the call,” or “We check availability every 5 minutes.”

2) What happens if a slot disappears mid‑call?

Race conditions happen. Ask the vendor to demonstrate an intentional conflict:

• Step 1: AI offers 2:00pm.
• Step 2: You (or a staff member) book 2:00pm manually on the same calendar.
• Step 3: The AI attempts the write and gets a 409/conflict.
• Expected: AI apologizes, re-queries availability, and offers the next time. No “we’ll email you a confirmation later.”

This is the difference between “we check availability” and “we perform an atomic write.” The latter avoids double-booking.

3) Can it respect buffers, travel time, resources, and multi‑staff calendars?

For home‑service or clinic workflows, you need buffers between jobs and resource constraints (truck, room, device). Ask for a live demo that shows:

• A 30‑minute job with a 15‑minute buffer enforced.
• A two‑technician or chair+provider constraint.
• A travel time gap between jobs.

Ask where those constraints live (your calendar vs vendor logic). If constraints live only in the vendor, they can drift from reality. If they live in your calendar system, the AI should read them and still perform an immediate write.

4) Where are recordings and transcripts stored — and how do I delete them?

Get a written data map: storage location, retention period, deletion path, and subprocessors.

Ask them to show you the exact UI/API steps to delete a recording and confirm the deletion. Telephony platforms like Twilio document external storage and data‑retention/deletion controls, as well as consent considerations. See Twilio’s help on recordings and data retention. If the vendor uses a provider like Twilio under the hood, their capabilities will mirror those controls — you should be able to retain in the vendor’s cloud, export to your storage (e.g., S3), set retention, and purge on request.

If they can’t show a deletion path with proof, assume the data sticks around.

5) Who processes the audio and transcripts? Do you sign BAAs? What consent lines do you play?

You’re often on the hook for consent and compliance. Vendors must name their subprocessors (STT, TTS, LLM, telephony). Twilio warns buyers to review consent rules (one‑party vs two‑party jurisdictions) and disclose recordings; see Twilio’s legal considerations for recording. If you need HIPAA‑eligible workflows, ask whether the vendor and each subprocessor will sign BAAs and whether transcripts/recordings are kept in HIPAA‑eligible services. Twilio’s architecture guide discusses HIPAA design patterns and third‑party processing; see Twilio’s Architecting for HIPAA.

Ask to hear the exact disclosure line the AI plays. Confirm it covers recording and SMS follow‑ups if they’ll text callers.

6) What are the hidden costs and minute math on your plan?

Have them walk one sample month on a screen share. Industry write‑ups call out gotchas: setup fees, per‑minute or per‑call rates, overages, SMS and transfer charges, holiday surcharges, minute‑bundle expirations, and seat licenses. See the breakdown in Ringeden’s virtual receptionist pricing guide.

Concrete request: “Please show a worked example invoice for 200 incoming calls and 50 transfers.” That specific scenario is recommended as a buyer test in Ringeden’s guide. Make them itemize:

• Setup fee and first month’s base charge.
• Minute counting rules: billed vs answered, rounding, whether ring time counts, whether transfer time counts.
• Per‑transfer and per‑SMS fees.
• Overage rates and whether bundles expire.
• Contract term, auto‑renewal, and cancellation fee.

If they won’t produce a sample invoice, assume the fine print hurts.

7) What’s the failover path if the AI gets lost or the call quality tanks?

You need a deterministic fallback. Ask them to wire this during trial and prove it live:

• Rule: If the AI asks the same question twice or hears “representative,” transfer to your team or your on‑call partner.
• If transfer fails, drop to voicemail and text you the transcript link.
• After‑hours rule: book to next‑day holds and send a summary SMS.

Then place three test calls that trigger each path. Don’t settle for “we can configure it later.”

8) How accurate are transcripts and are they editable?

Transcripts feed your CRM and keywords in call search. There’s no universal accuracy guarantee, so ask for either word‑error‑rate (WER) or a real sample transcript from a call in your niche (accents, truck noise, or medical terms). Twilio’s HIPAA architecture paper touches transcription/processing choices and where data flows; use it as a prompt to get specific about tooling and accuracy claims; see Twilio’s Architecting for HIPAA.

Must‑haves:

• You can edit transcripts and call notes after the call.
• Bookings auto‑attach the call summary to the calendar event or ticket.
• You can export transcripts to your CRM without manual copy/paste.

9) How do we measure voice quality before we sign?

Perceived voice quality has a standard: MOS (Mean Opinion Score), a 1–5 scale used by telcos and derived from ITU‑T P.800 methods. Ask the vendor for their MOS under load and any objective estimators (PESQ, POLQA). See Wikipedia’s page on Mean Opinion Score for the baseline.

Run your own quick check:

• Place five calls from a mobile on weak LTE and five from a desk phone.
• Note clipping, delay, and whether the AI speaks over you.
• If they claim a MOS, ask how they measured it and on what network conditions.

10) How does the AI confirm contact details and qualify leads?

Lead quality matters more than call length. Ask them to show a configurable intake:

• Call‑back number with read‑back confirmation.
• Email with spell‑back.
• Address with unit/apartment capture.
• Job type + urgency, and a custom “do not book these” rule.

Then verify where this data lands: calendar event fields, your CRM, and a daily summary email. If you use forms already, ask the AI to post into the same pipeline.

11) Can I bring my calendar, phone number, and CRM — and leave later without data lock‑in?

You should be able to:

• Use your existing Google/Microsoft calendar, not an isolated vendor calendar.
• Keep your phone number, or at least port it out later.
• Export recordings, transcripts, and contacts on demand.

Have them demonstrate an export of transcripts and contacts. For recordings/transcripts deletion and retention settings, ask for the exact path as shown in Twilio’s retention guide, or the equivalent if they use another platform.

12) What’s your 60‑minute live pilot plan — this week?

Give them an hour and hold them to it:

• 10 minutes: connect your real calendar and a test phone number.
• 10 minutes: set consent disclosure, after‑hours rule, and failover routing.
• 20 minutes: run conflict tests to prove no double‑booking (see #1–2 and use the Google Calendar API sync framing to ask about immediate writes vs polling).
• 10 minutes: capture a full intake, create a booking, and show the event with attached notes.
• 10 minutes: show the deletion flow for one recording/transcript and export a CSV of leads.

If they can’t do that in a guided session, expect a bumpy rollout.

Practical extras you’ll be glad you asked

• Do you log reason codes when the AI fails to book? Helps you improve scripts and policies.
• Can the AI read service‑area rules by ZIP and reject out‑of‑area jobs?
• Are holidays and PTO read from the same calendar your team uses?
• Can I approve the exact scripts and refusal rules before go‑live?

Where Fluxaro fits

If you want the website, bookings, AI answering, and analytics under one roof so the data actually lines up, Fluxaro builds custom, conversion‑focused sites and software with a 24/7 AI receptionist and a live analytics dashboard. See what we build on our Services page, or skim some live work on Examples.

Field test pack (copy/paste for vendor calls)

• Show me your immediate create‑event write on my calendar and how you prevent conflicts under load.
• Run a mid‑call conflict and recover without confirming the taken slot.
• Play your consent line. Tell me exactly where you store audio/transcripts, default retention, and the delete‑now steps. Provide subprocessors and BAAs where needed. See Twilio’s legal considerations.
• Provide a worked invoice for 200 calls and 50 transfers with setup, per‑minute, overage, SMS, transfer, and cancellation terms per Ringeden’s pricing guide.
• Provide transcript accuracy metrics (WER) or a real sample from my niche; confirm I can edit and export. Reference Twilio’s HIPAA architecture for where data flows.
• Provide MOS voice quality numbers and how measured. We’ll place 10 calls from mixed networks; see Mean Opinion Score.

Next step

Block 30 minutes, run the 60‑minute pilot outline above with your top vendor, and insist on the sample invoice and data‑deletion demo on the same call. If you want this wired into a site and booking system that’s built for conversions, book a quick call with Fluxaro here: Book a call.