Keep your phone number in your own name and forward it to the vendor rather than porting it, get a written data export of customers, bookings and recordings, choose month-to-month terms until the tool has proved itself, and write down how to switch back before you go live. Then test the exit once during the trial.
Lists of restaurant AI startups sell the tools. Nobody lists what happens when one is acquired, changes its pricing or disappears. For a restaurant the biggest exposure is the phone number, because customers have it in their phones and your Google profile points to it. The twelve questions to ask before you sign up deal with whether a tool works. This checklist deals with whether you can leave.
Why young vendors need a different checklist
A large POS company has lawyers, a status page and a support department. A two-year-old startup may have a founder answering emails. Neither is bad, and many good tools come from small teams. But the failure modes differ. Young vendors get acquired and the product is folded into another, change pricing after a funding round, drop features, or close, sometimes with short notice.
Closures happen to real, useful products. Bench, an accounting service, shut down on 27 December 2024, and Clockwise, a calendar tool, shut down on 27 March 2026 with customer data deleted, according to public notices. Neither is a restaurant tool, but they show the pattern: the product works, then it stops, and what you get back depends on what you arranged before. Treat this as ordinary supplier risk, like a produce supplier who might go out of business, and manage it the same way, with a second source and a plan.
The phone number: four setups, four levels of risk
If the tool answers calls, how the number connects decides almost everything. There are four common arrangements.
| Setup | How it works | Risk | Exit effort |
|---|---|---|---|
| Call forwarding | Your number stays with your phone provider; calls forward to a number the vendor gives you | Low. Switch the forward off and calls ring you again | Minutes |
| Forward on no answer or busy | Rings your staff first; only unanswered calls go to the AI | Lowest. The tool is a backup, not the front door | Minutes |
| Port the number to the vendor | The number is moved to the vendor's phone provider and the vendor becomes the account holder or carrier | High. You depend on them to move it back | Days to weeks, needs vendor cooperation |
| New number from the vendor | Customers are told a different number | Medium. Every printed menu and listing carries the wrong one | You reprint and update everywhere |
My default advice: start with forward-on-no-answer or busy. The vendor gets your overflow, which is where the missed-call value is, and you keep control. Move to full forwarding once you trust the tool. Port a number only if the feature you need requires it (some vendors say outbound calls or text messages need it), and then only with the terms below.
If you do have to port, ask these questions first, and get the answers in an email:
- Whose name is the number registered in during service, yours or the vendor's?
- If I cancel, how long does it take to port the number back out, and is there a fee?
- If the company closes, who can authorise a port-out, and is there a documented process?
- Will the vendor give me the account details and any PIN the carrier requires, on request, without a support ticket queue?
A vendor who says "we never port numbers" is making the simplest answer for you. A vendor who says "you own the number and here is the process" and can send that in a sentence is also fine. A vague answer is the red flag.
Customer data: what you hold, what they hold
An AI phone or messaging tool collects things you did not have before: call recordings, transcripts, caller numbers, booking details, sometimes dietary notes. Ask for an inventory. A sample question to send:
Subject: Data and exit questions before we sign
Hi [first name],
Before we start a trial, could you answer these in writing?
1. What customer data do you store for us (recordings, transcripts, names, numbers, bookings, notes)?
2. In what format and how quickly can we export all of it, and is there a fee?
3. How long do you keep recordings and transcripts, and can we set it ourselves?
4. If we cancel, when do you delete our data and do you confirm in writing?
5. Do you use our calls or customer data to train your models or anyone else's?
6. If the company is acquired or closes, what notice do we get and what happens to our data?
Thanks, [your name]An illustrative reply that is good enough: "Recordings and transcripts are stored for 90 days by default, configurable. You can export everything as CSV and audio files from the dashboard at no charge. On cancellation we delete within 30 days and send written confirmation. We do not train on customer calls. In an acquisition, customers get 60 days' notice." An illustrative reply that is not: "Your data is secure and you can contact support for exports." That says nothing about format, time or cost. Ask again, specifically, or treat it as an answer.
Then test it. During the trial, export your data once. Open the file. Check that customer names and numbers are there, that dates make sense, and that recordings play. An export that exists but cannot be opened is worth nothing when you need it.
Call recording also raises consent questions. Rules about telling callers they are recorded, and about automated voices, vary by place and by who is on the call, so ask the vendor what their tool says at the start of a call, and get advice on what you must say where you operate. Do not rely on a vendor's default wording without checking it.
Contract terms: the clauses that matter for a restaurant
You do not need to read forty pages the way a lawyer would. Look for six things and write each one in a line on your vendor sheet.
- Term and renewal. Month-to-month, annual, or longer? Does it auto-renew, and how many days' notice does cancelling need? Notice windows of 30 to 60 days that begin before the renewal date catch owners out.
- Price changes. Can the vendor raise prices during the term? With how much notice? Can you leave without penalty if they do?
- Setup and minimums. One-off fees and minimum commitments. For example, one restaurant phone vendor has quoted a setup fee of about $300 on top of monthly plans of roughly $40 to $200 (check the currency and current figures on its own page); a fee like that is a cost you do not recover if you leave in month two.
- Usage and overage. Calls, minutes or orders included, and what extra ones cost. Some phone tools price per call, so a busy month costs more.
- Service levels. Is there any uptime commitment, and what happens when the AI is down? The answer should be that calls fall back to your staff or voicemail.
- Termination and data return. How you end the contract, what you get back, and by when.
Terms do not have to be perfect. They need to be known. Ask for the ones you dislike to be changed, and note that small vendors often agree to a shorter trial or a month-to-month term when asked, because they want the reference.
The bigger the commitment, the more review you need. For any contract longer than a year, or one that involves taking payments, have a qualified adviser read it. The tutorial on evaluating an AI software vendor with a scorecard gives a general framework that sits alongside this one.
POS integration: the dependency that hides in plain sight
A phone tool that takes orders must send them somewhere. If the connection to your POS is built by the startup on top of your POS vendor's interface, it depends on two companies staying friendly. Ask four things. Which POS versions are supported? Who built and maintains the connection? What happens to orders if the connection fails: do they print, ring in the kitchen, or vanish? And does your POS vendor charge for third-party access? Some POS companies charge for API access or limit which partners may connect, so a tool that worked last month can stop after a policy change.
The test is to place five orders by phone during the trial, with a modifier on each, and watch what arrives in the kitchen. Then break it on purpose: switch off the POS connection and place another. If the failed order is silent, you have found a risk worth more than any feature. See whether Popmenu AI answering takes orders or only sends links for an example of how much that distinction matters, and POS-built AI against add-on tools for the other side of the decision.
What small vendors will usually agree to if you ask
Startups want customers who stay and give references, and many will bend where a large company will not. Four requests are worth making by email before you sign, each of which costs the vendor little:
- A month-to-month term for the first three months, then the annual price if you stay.
- A written promise that the number stays registered in your name, and that they will cooperate with a port-out within a stated number of days.
- A named export format and a commitment to run an export for you within a few working days of a request.
- Thirty days' written notice of any price change, with the right to cancel in that window.
An illustrative email: "We would like to start on a monthly term for 90 days, with the annual rate if we continue. Could you confirm in writing the four points below?" Keep the reply. If a vendor refuses all four, that is information about how they will treat you in month nine.
Payments, deposits and who holds the money
If the tool takes deposits, pre-orders or card details over the phone, the next question is where the money goes. Does it pass through your own payment account, or does the vendor collect it and pay you out? Money held by a vendor is exposed to its failure in a way money paid directly to your account is not. Prefer setups where the payment goes straight to your own processor, and where the vendor never sees full card numbers; ask who is responsible for card data and how the vendor handles it. For anything that touches payments, take advice, and check the vendor's payment terms in the contract, not just on the sales page.
The same logic applies to gift cards and prepaid vouchers. Balances that live only inside a vendor's system are at risk if it closes. Ask whether you can export outstanding balances, and keep a monthly copy.
Is the vendor likely to be around? Signs you can check in ten minutes
You cannot predict a closure, but you can read signals. None is proof, and big names fail too.
- Public pricing. A price page that states numbers, and has not changed three times in a year, suggests a stable model. Pages that say "contact us" for everything are not a fault but tell you less.
- A changelog or status page. Recent entries show a team still shipping and a place to check for outages.
- Who answers support. Send a pre-sales question and time the reply. Slow answers before you pay will not improve afterwards.
- Customer count claims. Treat them as marketing unless named references will talk to you. Ask for two restaurants of your size that have used it for six months, and call them.
- Funding and ownership news. A funding round is good news for the vendor and a pricing risk for you; an acquisition changes terms. Set a news alert on the vendor name.
Worked example: a 60-seat restaurant trials phone AI
An illustration with invented numbers. A 60-seat restaurant loses about 10 calls a week during Friday and Saturday service. A startup offers phone AI at $150 a month on monthly billing, or $85 a month on annual billing with a 12-month term. The annual saving is $780, and the risk is 12 months of commitment to a young company.
Step 1: choose the number setup. The owner sets "forward when busy or no answer" on her existing line. Her number stays with her carrier. Exit effort: switching off a forward.
Step 2: ask the six data questions. The reply promises a CSV export and 30-day deletion, but says nothing about training use. She asks again and gets a clear "no". She keeps the email.
Step 3: choose monthly billing. Over a year, monthly costs $1,800 against $1,020 annual, a difference of $780. She runs a three-month trial on monthly ($450), then decides. If the tool earns its keep she can ask for the annual rate with a cancellation clause. If it does not, her maximum loss is $450 plus any setup fee, not $1,020 for a service she dislikes.
Step 4: test the export. In week two she downloads recordings and the caller list. The CSV opens, but call times are in a different time zone from hers. That is a minor defect she notes for the export notes.
Step 5: write the exit sheet. One page: how to switch off the forward, who to tell, where the data lives, the date of the last export. She files it with the contract.
The decision comes after the trial, on the numbers: calls answered, bookings made, and how many staff interruptions it saved.
A one-page vendor risk sheet, filled in
| Item | Answer | Risk |
|---|---|---|
| Number setup | Forward on busy and no answer; stays with our carrier | Low |
| Data export | CSV and audio, tested on [date], opens fine | Low |
| Training on our calls | "No", in email dated [date] | Low |
| Term | Monthly, 30 days' notice | Low |
| Price change clause | 30 days' notice, can cancel | Medium |
| POS connection | Built by vendor; failed orders print as "unsent" | Medium |
| Vendor age and signals | Two years, public pricing, no status page | Medium |
| Fallback if down | Calls ring staff phone | Low |
Anything marked high on the number or the data is a reason not to sign. Medium is a reason to set a review date and keep the term short.
The exit drill: practise leaving before you need to
- Write the switch-back steps. Who turns off the forward, on which account, with what login.
- Export once a quarter. Keep the files on your own storage with a date.
- Keep customer-facing listings pointing at your number. Google profile, website, delivery apps. Never at a vendor's number.
- Keep a backup plan for the job. A voicemail script, a booking link, or an agreed callback routine. Missing calls for a week is survivable if the process is written down.
- Tell the manager. The person on shift when the tool fails should know what to do without phoning the owner.
When a vendor gives notice of a closure or a big change, move quickly: export everything, start the port-back if you ported, switch the forward off, and tell customers if the number is changing. The first 48 hours are the cheapest time to act.
What to do if you are already signed up
Many owners read this after signing. Do not panic; run the same checks now. Find out whose name the number is in, run an export this week, and read the cancellation clause. If you ported a number and cannot find the answer, email the vendor and ask for the process in writing. If the answer is poor, consider a quiet move back to forwarding while you still have a good relationship. For the costs side of the same decision, what phone answering costs and pays back and the vendor comparison help you judge whether to stay. If you would rather have someone go through your contract and plan the switch with you, that is what my AI implementation consultation covers.
Vendor-risk questions from restaurant owners
Should I ever forward my main number to an AI vendor?
Forwarding is the safest start, because the number stays with your phone provider and you can switch the forward off in minutes. Port the number to the vendor only if you must, and only once you have the porting and return terms in writing.
Is month-to-month always better than a yearly contract?
Usually, for a young vendor, because it lets you leave if service slips. A yearly term can be worth it for a lower price once the tool has run for a quarter and you have tested the export. Never accept a long term plus an auto-renewal you cannot easily cancel.
What data should I be able to take with me?
At least customer names and contact details, booking history, call recordings and transcripts if you want them, menu and settings, and your phone number. Ask for the export format and test it during the trial.
How do I know if a startup is likely to fold?
You cannot know. Look at signs: how long it has operated, whether pricing is public, whether it publishes a status page and changelog, and who answers support. Then plan as if it might, so a closure costs you a bad week rather than your number.
Further reads
- Questions to Ask an AI Vendor Before You Sign Anything — General questions for any AI vendor.
- How Much Does AI Cost a Small Restaurant Each Month? — A monthly AI budget for a small restaurant.
- How to Buy AI Software for a Shop or Restaurant: Pre-Purchase Checks — Pre-purchase checks for shops and restaurants.
- What an AI Implementation Plan Looks Like for a Small Restaurant — An implementation plan for a small restaurant.
- What If Your AI Vendor Shuts Down? Checks Before You Commit — The general shutdown plan for any AI supplier.
- AI Vendor Lock-In: How to Keep Your Data and Prompts Portable — Keeping data and prompts portable across tools.
- What Uptime and Support Should an AI Vendor Promise You? — What uptime and support promises to ask for.
- How to Check an AI Software Vendor's Customer References — How to call references so they tell you something.
- AI Consulting for Restaurants: What to Ask and What to Expect — A sample scope of work, the questions to ask, and the deliverables a restaurant should expect from an AI consultant in the first month.
- Is Yelp Host Worth It for a Small Restaurant? — An independent cost-and-listing verdict on Yelp Host: the price, the break-even sum, and what your Yelp page and POS must have in place.
- Maple AI for Restaurants: Voice or Pro Plan, and Who Needs Which — Maple Voice answers and transfers calls; Pro adds orders and reservations. Compare the monthly cost, break-even orders and the test to run first.
- Owner.com Pricing for Restaurants: Flat Rate or Per-Order Fee? — Owner.com's $249 plan with a 5% order fee against the $499 flat plan: the break-even in sales and orders, with worked monthly numbers.
- Pizza Phone Ordering System: What AI Costs per Month — Per-minute, flat-fee and per-customer AI phone pricing for a pizza shop, with a missed-call payback sum and the extras that never show on the price page.
- SevenRooms Voice AI: What It Does and Who It Suits — What SevenRooms Voice AI does, why you must ask for a quote, and how it compares with Yelp Host, Slang and Popmenu.
- Restaurant AI Voice Agent: 12 Test Calls to Run Before Go-Live — Twelve scripted calls with pass marks, blocker or fix labels and a go or no-go rule, so you find the failures before a guest does.
- Square AI Voice Ordering for Restaurants: Set Up and Test (Beta) — The Dashboard steps, the plan you need and a seven-day test-call schedule to run before Square's AI answers your restaurant phone.
- AI Drive-Thru Ordering: What a Fast-Food Franchisee Controls — Who really decides on AI drive-thru ordering in a franchise: three models, the clauses to read, questions to put in writing and a lane-level sum.
- AI Robots in a Restaurant: Are They Worth It for a Small Place? — Serving robots versus handheld and QR ordering for a small restaurant: what each removes, a walk audit with sums, and the lease-or-buy terms to demand.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: vendor pricing pages and earlier price checks for Maple, Shepherdline and Yelp Host (27 October 2026); public shutdown notices for Bench (December 2024) and Clockwise (March 2026); your own contract and each vendor's terms of service.