This restaurant voice AI buyer checklist is for the moment after the demo. Every demo sounds good: the vendor picks the menu, the vendor places the call, and nothing goes wrong. The real decision is made on the questions nobody put in the demo script. Take these into any sales call — ours included — to separate a tool that will work on a Friday rush from one that only works on a Tuesday afternoon.
1. Can I test it on my own menu, with my own questions?
A generic sample restaurant proves very little. Ask to try the system on your menu: the items with odd names, the combos, the modifiers, the half-and-half orders. Then call it the way a rushed customer would — interrupt it, change your mind, ask something off-script. If a vendor won't let you do this before you sign, that tells you something.
2. What happens when it doesn't know?
Every system hits a question it can't answer. What matters is the fallback. Ask:
- Can it forward the call to a person, and under what rules you control?
- If nobody picks up, does the caller get a dead end or a way to leave a message?
- Does it say “I'm not sure” or does it guess? A confident wrong answer about an allergen is worse than a handoff.
Depla's approach is covered on the human handoff page, and the same questions are worth putting to anyone you compare.
3. How does the order actually reach my kitchen?
Taking an order on the phone is only half the job. Ask where it lands: a dashboard, a printed ticket, your POS? Ask whether staff re-key anything. If a POS integration is claimed, ask which one is live todayversus planned. “Integrates with everything” usually means “integrates with one, and has a roadmap.” Our POS integration explainer goes deeper on what to verify.
4. Which languages are live, not “coming”?
If a meaningful share of your callers speak something other than English, ask for a live call in that language, not a slide. Ask about mixed speech too, such as Hindi and English in the same sentence. See our languages page for what Depla supports now.
5. What does it really cost?
Get the full picture in writing: monthly fee, setup fee, per-call or per-minute charges, overage rules, contract length, and how cancellation works. Then model your own call volume against it. Per-minute pricing that looks cheap at 200 calls can look different at 2,000. Depla's plans are published on the pricing page; ask every vendor for the equivalent.
6. Who sees the call data, and how long is it kept?
Calls contain names, phone numbers and sometimes addresses. Ask what is recorded, how long recordings and transcripts are retained, who can access them, and whether you can delete them. Also ask how the vendor handles caller consent and disclosure on your calls, and check with your own counsel; nothing here is legal advice.
7. Can I see what it did?
You are handing your phone line to software. You should be able to read the transcript of any call, see the order it created, and spot where it went wrong. If you can't audit it, you can't improve it, and you can't resolve a customer dispute.
8. How long until it's live, and who does the setup?
Ask what you have to do. Do you upload a menu or does someone build it with you? Who tunes the prompts when it mishandles something? A vendor with a thin onboarding story often means you become the implementation team.
9. How do I leave?
Ask whether you keep your phone number, whether you can export your call and order data, and whether there is a notice period. A confident vendor answers this quickly.
10. Is AI even the right tool for my calls?
Not every restaurant needs voice AI, and a human answering service is a legitimate choice, especially for low volume or highly relationship-driven calls. We lay out the trade-offs in AI vs a human answering service. Run the checklist against both options so the comparison is fair.
A simple way to score vendors with this buyer checklist
Give each question a pass, partial or fail, and weight the ones that matter most to your operation. A busy takeout counter should weigh questions 1, 3 and 5 heavily; a place with many non-English callers should weigh 4. Any outright fail on 2 (fallback) or 6 (data) is worth treating as a deal-breaker. You'll know more from one hour of testing than from any brochure.
Try the questions on Depla
We'd rather you test us than trust us. Try the live demo with your own website and menu, call it like a customer would, and hold us to this list.