Key takeaways
- Cost, fit, skills and integration are the top barriers operators report, so ask vendors to prove fit and integration on your own calls.
- Ask to see the exact handoff your staff receive, not a slide about it.
- Cover recording consent, texting consent and data use in writing before go-live.
- Prefer short terms, clean data export and a pilot with agreed success measures.
Why buyers hesitate, and what to do about it
Restaurant operators are not short of technology pitches. They are short of time to check them. In November 2026 National Restaurant Association research, the top obstacles to adopting or expanding technology were cost (48.8%), business fit and value (43.8%), workforce and expertise gaps (41.3%) and integration challenges (38.8%)[1]. The same research found training and education top the list of support operators say would help them adopt technology and AI[1].
Every one of those barriers can be tested before you sign. The questions below are grouped by topic. Send them to each vendor in writing, ask for a live demonstration on the call types that matter to you, and score the answers the same way for everyone.
Before you talk to any vendor
List your top call types
From two weeks of call logs or voicemails, list the five most common reasons people call: same-night bookings, takeout status, hours, catering, large parties.
Decide what the system should finish
For each call type, write what a good outcome is: a complete reservation request, an ordering link sent, a catering lead with guest count.
Write your never list
Allergy answers, refunds, promises about tables, anything about a guest's complaint. These become handoff triggers.
Pick success measures
Calls answered, requests captured, handoffs completed, wrong answers found in review, staff time saved.
Name an internal owner
One manager owns setup, weekly review and the vendor relationship. Without an owner, the pilot drifts.
Reservations, ordering and POS
- Which reservation platforms do you connect to today, and can you show a live booking or request in ours?
- Is the connection live now, or planned? If planned, what is the date and what happens until then?
- How do you handle held tables, combined tables, pacing limits and large-party rules?
- For takeout: do you send an ordering link, capture the order for staff, or create a ticket in our POS? Which POS systems are supported today?
- How are menu changes, 86'd items and modifiers kept in sync?
- What happens when the reservation or POS system is down?
Ask every vendor to label each capability as live or planned. A feature on a roadmap is not a feature you can use on Friday.
Staff handoff
- Show me the exact message my host receives after a reservation call.
- Which events trigger a handoff, and can we change them without a support ticket?
- Can you warm-transfer to a person, and what happens if nobody picks up?
- How does a manager receive complaint and allergy calls, and is there a backup path?
- Where do call summaries and transcripts land, and who is notified?
- How do staff correct a wrong answer so it does not happen again?
Recording, consent and texting
An AI phone system usually records or transcribes calls and often sends texts. Both are regulated. Federal law permits recording with one party's consent, but some states, such as California, require the consent of all parties[2]. The Telephone Consumer Protection Act lets consumers sue for actual losses or $500 per violation, whichever is greater[3], so texting consent and opt-outs need to work from day one.
- What disclosure does the system play, and can we edit the wording?
- Is the disclosure captured in the recording itself?
- Can recording be switched off for a caller who objects?
- Which number do texts come from, and is it registered for business messaging?
- How are opt-out replies such as STOP handled, and how fast is a number suppressed?
- Are marketing texts kept separate from service texts?
Data use and privacy
Guests are wary of how AI uses their information. About six in ten Americans say they would like more control over how AI is used in their own lives[4]. Your contract should give you, and through you your guests, that control.
- Who owns the recordings, transcripts and guest details?
- Is our call data used to train models for other customers? Can we opt out?
- Where is data stored, and how long is it kept by default?
- Can we delete a guest's data on request, including the vendor's copies?
- Which staff at the vendor can access our calls, and is access logged?
- What happens to our data when the contract ends?
Contract and support terms
- Can we start with a pilot at one location, with agreed success measures and an exit if they are not met?
- What is the minimum term, and how is renewal handled?
- What is included in onboarding, and who writes the house information and scripts?
- What are support hours, especially Friday and Saturday evenings?
- How much notice is given before pricing or terms change?
- Can we export our call history and settings if we leave?
Small businesses are less experienced with AI than large ones: 17% of US firms with fewer than 20 employees use AI, against 30% of firms with at least 250[5]. If your team is small, weight onboarding and support heavily, because they will decide whether the system is set up well.
Run the demo yourself
A vendor-led demo shows the product at its best. A buyer-led demo shows it on your restaurant. Ask each vendor to load a basic version of your house information, hours and booking rules into a test line, then call it yourself, ideally from a noisy place such as your own dining room during service.
- Book a table for six on a Saturday with a birthday and a high chair, then change the time halfway through the call.
- Ask whether the fried chicken is cooked in the same oil as the shrimp; the right answer is a handoff, not a guess.
- Ask for directions and parking, then ask an unrelated question about gift cards in the same breath.
- Pretend to be upset about a takeout order that arrived cold and ask for a manager.
- Call about a rehearsal dinner for forty and see which questions it asks.
- Speak quickly, mumble a name, and spell it only when asked; check whether it reads the name back.
After each call, look at what arrived on the staff side: the summary, the details, the transcript and who was notified. That output, more than the voice, tells you whether the system will save your hosts time.
Score vendors the same way
| Criterion | Example weight | What good looks like |
|---|---|---|
| Handles your top call types | 25 | Live demo of each, with a complete handoff message |
| Handoff quality | 20 | Summary, details and callback number reach a named person fast |
| Integration | 15 | Live connection to your reservation or ordering tools, or a clear interim process |
| Consent and texting | 15 | Editable disclosure, recorded announcement, working opt-out |
| Data terms | 10 | You own the data, deletion works, no training on your calls without consent |
| Support and onboarding | 10 | Evening and weekend support; vendor helps write setup |
| Contract flexibility | 5 | Pilot, short term, clean export |
Score each vendor out of 100 using the same calls and the same sheet. Two managers scoring independently, then comparing, catches the halo effect of a polished demo.
Red flags
- No live demonstration on your scenarios, only recordings chosen by the vendor.
- Features described as available that turn out to be planned.
- No way to change handoff rules without a support ticket.
- Vague answers about whether your calls train models for others.
- Long minimum terms with no pilot or exit.
- Allergy questions answered by the system in the demo.
Example: choosing between two systems
Example: a fictional two-location group shortlists two vendors. Vendor A has the smoother voice but sends staff a recording and a phone number after each call. Vendor B sends a complete request with party size, occasion and callback number, and its allergy handoff reaches the manager's mobile with a summary. Both score similarly on consent; only Vendor B allows a one-location pilot with a 60-day exit.
Example: the group chooses Vendor B, runs the pilot at the busier location and reviews call summaries weekly. Before go-live, the manager works through a structured test using how to test an AI phone agent before going live and sets triggers from when an AI phone agent should hand off to staff.
Frequently asked questions
What is the most important question to ask an AI phone vendor?
Ask to see the exact message your staff receive after a reservation call and after a handoff. It shows whether the system finishes tasks or just moves work to later.
Should an AI phone system connect to our reservation platform?
It helps, but only if the connection is live and tested on your floor plan. Until then, reservation requests that the host confirms are a safe interim process.
Do we need a pilot before signing?
A one-location pilot with agreed success measures and an exit is the best protection against poor fit, which operators list among their top technology barriers.
Who owns call recordings and transcripts?
Your contract should say you do, explain how long the vendor keeps them, and confirm that deletion requests remove the vendor's copies too.
References
National Restaurant Association (2026). Research Insight: The Future of Restaurant Technology (press release, November 5, 2026) (opens in a new tab) ↑
Legal Information Institute (Cornell Law School), US Code (2026). 18 U.S. Code § 2511 - Interception and disclosure of wire, oral, or electronic communications prohibited (opens in a new tab) ↑
Legal Information Institute (Cornell Law School), US Code (2026). 47 U.S. Code § 227 - Restrictions on use of telephone equipment (opens in a new tab) ↑
Pew Research Center (2025). How Americans View AI and Its Impact on People and Society (opens in a new tab) ↑
Federal Reserve Bank of Minneapolis (analysis of Census Bureau BTOS) (2026). AI adoption in business grows steadily but unevenly (opens in a new tab) ↑