AI in Your Industry
Hospitality

AI for restaurants: where thin margins meet real automation

The loudest restaurant AI, the drive-thru voice bot, is the part that publicly failed. The quiet back-office tools are where the measured money is. Here is what survived contact with real kitchens, for independents in Vancouver, Canada, and the US.

By Adi Huric, founder of Most AI Labs·July 2026·10 min read

McDonald’s spent three years testing AI drive-thru ordering with IBM and shut it down. The SEC charged another drive-thru AI company for claiming 95% automation while humans in the Philippines quietly handled most orders. Meanwhile, the unglamorous stuff, reservation reminders, waste tracking, review responses, keeps producing measurable returns. This article is about that gap.

Written for independent operators and small groups in Vancouver, anywhere in Canada, or the US. Every claim is cited. Vendor numbers are labeled as vendor numbers, and where the evidence is thin we say so.

Why this matters now, especially in Canada

Restaurants Canada reported in late 2025 that 44% of Canadian restaurants are unprofitable or merely breaking even, against 12% in 2019, and its 2026 forecast projects real foodservice sales declining 1.1%. In BC, the general minimum wage rose to $18.25 on June 1, 2026, the highest provincial rate in Canada. US benchmarks say the median full-service restaurant keeps 2.8 cents of every dollar pre-tax; limited-service keeps about 4. Food and labor have each climbed roughly 35% since 2019.

On margins like that, you cannot market your way out. The realistic play is clawing back the money already leaking: empty tables that were booked, food bought and binned, calls that ring out on a Friday night. That is exactly where the evidence for AI and automation is strongest.

The adoption gap nobody advertises

The National Restaurant Association’s 2026 State of the Industry found 26% of US operators using AI, with marketing as the top use, about 10% on admin tasks, and only 6% using AI for customer orders. Sentiment runs far ahead of deployment: Toast found 86% of decision-makers comfortable with AI, TouchBistro’s 2026 surveys put positive sentiment at 85% in the US and 79% in Canada, and Deloitte found 82% of executives planning to increase AI investment while only about one in five has any governance for it.

Read those numbers together and the picture is clear: operators like AI in the abstract, deploy it in the back office, and keep it away from guests. The Canadian data adds a useful detail: 48% of operators say automation has saved them time, and operators using online ordering report an average 18% sales lift from it (TouchBistro’s 2026 Canadian report, run with Harris Poll, including Vancouver operators).

The famous failures, and what they teach

  • ·McDonald’s and IBM, 2024. After 100+ test stores, McDonald’s ended its automated drive-thru ordering partnership. Reported accuracy was 80 to 85% against the roughly 90% a human crew hits, and the failures went viral: bacon on ice cream, hundreds of dollars of unwanted nuggets.
  • ·Presto Automation, 2025. The SEC charged the drive-thru AI vendor for claiming its system automated 95% of orders when roughly 70% needed human agents offshore to complete. The company’s own filings later admitted it.
  • ·Taco Bell, 2025. After 18,000 water cups were ordered as a prank, the chain publicly rethought where voice AI belongs.

The lesson is not that voice AI never works. Wendy’s reports its FreshAI now handles 86% of drive-thru orders without human help (a company number), and White Castle reports 90%+ completion with SoundHound. The lesson is that guest-facing voice AI is the hardest problem in the room, and chains with nine-figure technology budgets spent years and public embarrassment getting it half-right. An independent should not buy the hardest problem first. The chains already paid that tuition for you.

Watch out for this

Your guests have opinions too: 64% of diners worry about AI eliminating jobs, 87% say staff connection is critical to hospitality, and about three quarters accept automation mainly where it fixes short-staffing rather than replacing people (Reputation.com, HungerRush, and Square surveys, 2025). Automate the back office and the phone queue. Keep the floor human.

Phones, reservations, and no-shows: the recoverable revenue

A Harris Poll survey of about 2,000 US adults (commissioned by a phone-AI vendor, so label it accordingly) found 63% of Americans still prefer contacting restaurants by phone. Popmenu’s guest research found 42% of guests will simply book elsewhere if a reservation call goes to voicemail. Every unanswered ring during service is a table quietly walking to a competitor.

No-shows are the other half. OpenTable found 28% of Americans admit to no-showing in the past year, and unmanaged no-show rates run 10 to 20%. The mitigation evidence is solid and cheap: SMS reminders cut no-shows by roughly a quarter to a third (vendor-reported range), and requiring a card on file drops rates to around 3% (TheFork). At an average $45 per head, a 40-seat room eating 10 to 20% no-shows is losing roughly $450 to $900 a night, which is why six no-shows can erase an entire evening’s profit at median margins.

AI phone agents that answer, take reservations, and answer “are you open on Sunday” questions are the newer layer on top. The strongest public case is a Miami restaurant whose AI handled about 7,700 reservation calls in three months, with a claimed $600,000 revenue effect over eleven months, reported via Forbes but sourced from the vendor. We found no independent study of restaurant phone AI yet, and we would rather tell you that than pretend. The missed-call problem is real; measure your own unanswered calls for two weeks before believing anyone’s revenue claims.

Waste and forecasting: the money already in your bins

Commercial kitchens waste 5 to 15% of the food they purchase. ReFED puts US restaurant food waste at $25 billion a year and estimates about $14 returned for every $1 invested in waste reduction. AI waste tracking (a camera and scale over the bin, learning what gets thrown out and why) is dominated by vendor evidence: Winnow, deployed in 3,000+ kitchens, claims typical waste cuts of 50%+ in year one, worth 2 to 8 points of food cost. The biggest published deployment is chain-scale: IKEA’s kitchens cut waste 54% and saved an estimated $37 million.

Do the derived math for an independent (our arithmetic, not a study): a $1M-revenue restaurant buys roughly $320,000 of food at typical cost ratios. Wasting 5 to 15% of it means $16,000 to $48,000 a year in the bin. Against the median full-service pre-tax profit of $28,000 on that same $1M, reclaiming even half the waste rivals your entire bottom line. This is why demand forecasting and smart prep lists, the least cinematic AI in the industry, are quietly its best-paying.

Reviews: the one place the evidence favors independents

The most rigorous study in this space is Michael Luca’s Harvard work on Yelp: a one-star rating increase drives a 5 to 9% revenue increase, and the effect exists for independents only. Chains see none of it. That asymmetry is rare good news: reputation work pays you specifically because you are not a chain. On a $1M independent, one star is worth roughly $50,000 to $90,000 a year (derived from Luca’s range).

Meanwhile BrightLocal’s consumer survey finds 87%+ of consumers reading reviews for local businesses, and the share expecting a same-day response to their review tripled year over year. This is a perfect AI-plus-human task: software watches every platform and drafts responses in your voice, a person approves and personalizes. Nobody needs a study to tell you the owner does not have time to answer every Tuesday-night Yelp review at 11 pm. Now the guest expects it anyway.

The pattern across every industry we research holds here too: the boring automations with years of evidence out-earn the impressive ones with a demo video. In restaurants the gap is the widest we have seen. The demo-video category literally drew SEC charges.

Scheduling and labor: promising, vendor-proven only

AI scheduling tools promise demand-matched rosters and less unbudgeted overtime, and operators who use them report real relief. But the published numbers are vendor numbers (7shifts and peers), with no independent study we could find. The nearest neutral signal: in the NRA’s 2026 survey, 69% of operators who added technology say their restaurant became more efficient. Treat scheduling AI as a probable win priced like a utility, not a guaranteed one priced like a transformation. With BC labor now at $18.25 a floor hour, even a few recovered scheduling hours a week covers the software.

The honest bottom line

  • ·Start with no-show defense: SMS reminders and card-on-file for bookings. Strongest evidence, near-zero cost, measurable in a month.
  • ·Then review response discipline: AI-drafted, human-approved, same-day. The Harvard math says reputation pays independents specifically.
  • ·If your food cost runs above 32%, pilot waste tracking: vendor-heavy evidence, but the derived math is compelling against 2.8% margins.
  • ·Phones: measure first, then pilot. Two weeks of call logs tells you what unanswered calls cost you. Short contract, verify against your own baseline.
  • ·Skip guest-facing voice ordering. McDonald’s, Presto, and Taco Bell ran that experiment publicly so you do not have to.

If you want this mapped to your restaurant with your own covers, food cost, and call volume, that is what our AI consulting work does. It starts with the free 7-day audit: where your bookings, bins, and phone lines actually leak, with the math shown before you spend anything. Fixed price. You own what gets built.

Wondering what an ad budget buys here? Our free lead calculator runs the verified benchmark math for this industry: restaurant offers on Meta run $3.16 each, the cheapest lead in any industry we track.

Sources

If any of this is your week

Start with the 7-day audit.

7 business days. A real document. Yours to keep — whether you hire us or not.

Start with the 7-day auditOr talk to the founder first

Read next

Dental

AI for dental clinics: diagnostics, no-shows, and the front desk

The same evidence-first treatment for dental practice owners.

Read→
Legal

AI for law firms: what the evidence actually supports

Measured productivity gains, court sanctions, and the billable-hour math.

Read→