Industries / Retail, hospitality and personal services / Hospitality and tourism / Restaurants and food service
Ordering and paying
This is how guests choose, order and pay: what the menu must tell them, and the counter, belt, website, app and screen through which the order reaches the kitchen.
6
Milestones achieved, from 1958 to 2024.
1
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1958
Sushi rides a conveyor
Kaiten-zushi, or conveyor-belt sushi, puts small plates on a moving belt that passes every seat, so diners take what they want without ordering from a cook. The first such restaurant, Mawaru Genroku Sushi, opened in 1958, by the account of the Japanese site Nippon.com. Shiraishi Yoshiaki, who ran a standing sushi bar in Osaka, added the belt to cut labour and costs. The Osaka tourism bureau says the founder of Genroku Sangyo got the idea from bottles moving on a conveyor at a beer factory in Suita, and wanted to ease a labour shortage.
Credit
Shiraishi Yoshiaki and Genroku Sangyo
1994
Pizza ordered on the web
On 22 August 1994 the software firm The Santa Cruz Operation (SCO) and Pizza Hut announced PizzaNet, a trial in the Santa Cruz area of California. A customer filled in a menu page on the web. The order went over the internet to a server at Pizza Hut's head office in Wichita, Kansas, then by modem over phone lines to the nearest Pizza Hut. The store could phone first-time customers to check the order, and payment was made at the door. The companies called it the first way to order pizza delivery over the internet; that "first" is their own claim.
Credit
Pizza Hut and The Santa Cruz Operation
2015
Order ahead in the app
Starbucks let customers order and pay in its phone app and collect the drink without queuing. The service began in Portland, Oregon, in December 2014, and Starbucks expected to have it in all 7,400 of its company-run US stores by the end of September 2015, Nation's Restaurant News reported. By January 2026 Starbucks said mobile orders were 32% of its transactions. The app and the bar fell out of step. Howard Schultz, a former chief executive, called the mobile system the company's "biggest Achilles heel" in June 2024, and GeekWire described the morning rush: the app quoted everyone the same seven-minute wait, customers arrived together and drinks piled up at the pickup counter.
LessonAn order that skips the queue still has to be timed against what the kitchen or bar can make, or the queue moves to the pickup counter.
Credit
Starbucks
2018
Calories on the menu board
From 7 May 2018 the US Food and Drug Administration (FDA) has required restaurant chains with 20 or more locations to show the calories of standard items on menus and menu boards. Joshua Petimar and colleagues studied 104 restaurants of one fast food franchise in Louisiana, Texas and Mississippi that added calorie labels in April 2017. Over 49 million transactions, calories bought fell by 60 per transaction, about 4%, and the drop faded over the following year. England has required calories on menus, including online menus, at food businesses with 250 or more employees since April 2022. Alexandra Kalbus and colleagues read Deliveroo and Uber Eats menus across Great Britain and found that the median share of restaurants showing calories fell from 14% in June 2022 to 12% in October 2023, and was 9% in the poorest neighbourhoods against 14% in the richest. The rules cover large chains, not independent restaurants.
LessonA calorie label shifted fast food purchases a little and only for a while, and online it reached a small share of restaurants.
Credit
the US FDA; the UK government; measured by Petimar et al. (BMJ, 2019) and Kalbus et al. (Health & Place, 2026)
2019
Blind diners win ordering case
Guillermo Robles, who uses a screen reader (software that reads a web page aloud), twice failed to order a customized pizza on Domino's website and app. On 15 January 2019 the US Court of Appeals for the Ninth Circuit held that the Americans with Disabilities Act (ADA) covers Domino's website and app, because they serve its restaurants, which the Act lists as places open to the public. The court set no technical standard, but said an order to follow the Web Content Accessibility Guidelines (WCAG) 2.0 was a possible remedy. Domino's added a phone number for screen-reader users only after the suit was filed. The US Supreme Court denied Domino's petition on 7 October 2019.
LessonOnce ordering moved online, a site that blind customers cannot use became a barrier to the restaurant itself.
Credit
Guillermo Robles; the US Court of Appeals for the Ninth Circuit
2024
Sushi chains drop the belt
In early 2023 a video spread of a diner at a Sushiro restaurant licking a soy sauce bottle and a teacup and touching sushi passing on the belt. Sushiro stopped sending unordered plates around for a time and sent only ordered plates to tables. By September 2024, The Japan News reported, Sushiro had stopped using conveyor belts in all its stores and moved to touch-screen ordering, with Digiro, a large screen at each table that shows sushi flowing past as video, in 19 stores. Hamazushi had dropped belts at 90% of its outlets and expected to cut about 1,000 tonnes of food waste a year, since sushi left too long on a belt is thrown away. Kura Sushi kept its belts, says AI forecasts of what to put on them hold waste to about 3%, and has taught the cameras that count its plates to spot tampering.
LessonThe open belt that defined the format wasted food and left plates open to strangers, and the largest chains turned it into an order screen and a delivery track.
Credit
Akindo Sushiro; Zensho's Hamazushi; Kura Sushi
Watch for
Mobile pickups spread by algorithm
Starbucks says a "Smart Q" algorithm now spaces out mobile order pickups instead of sending everyone to the counter at once, and that peak service times are under its four-minute target. It says the 650 stores that took its new staffing and service model first are outperforming the rest of the chain by about 2 percentage points in comparable sales.
Achieved whenAn outside party, such as a mystery-shop firm or a published study, measures order-to-pickup times and pickup counter crowding in Starbucks stores and finds the four-minute figure holds at peak.
Claimed by
Starbucks chief executive Brian Niccol, January 2026, as reported by GeekWire
Open
01
Menu options test set
A public set of restaurant menu photos with a checked answer key, and a scorer. The key lists every item and price and, above all, every option group: sizes, choice of side, add-ons and their prices, and how many choices are required or allowed. These options are what a delivery app or an ordering bot must get right to take an order. The scorer compares a model's structured output with the key, item by item and option by option.
The briefCredit · unclaimed02
Screen reader ordering test
A repeatable test of whether blind customers can order on restaurant websites and apps. Testers who use screen readers build one set order on each site, an item with two changes such as a size and a removed topping, and stop at the payment page, so no order is placed and no money is spent. Log whether they reach the payment page, how long it takes and every step that blocks them.
The briefCredit · unclaimed
Open 01 · the brief
Menu options test set
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public set of restaurant menu photos with a checked answer key, and a scorer. The key lists every item and price and, above all, every option group: sizes, choice of side, add-ons and their prices, and how many choices are required or allowed. These options are what a delivery app or an ordering bot must get right to take an order. The scorer compares a model's structured output with the key, item by item and option by option.
Why it is still open
DoorDash wrote in March 2025 that it now turns menu photos sent by restaurants into structured menus with text recognition and a language model, and sends doubtful results to people, because human checks found errors in "a reasonable proportion" of menus. It published no accuracy figures and no data. The public menu benchmark found, MOTBench (2025), asks models to read each dish with its price and unit, and its abstract does not cover option groups. GitHub searches for menu option or modifier datasets found only demo ordering apps.
Built with what exists
A phone camera; an open labeling tool such as Label Studio; a JSON schema for menus with option groups and minimum and maximum choices; vision-language models, which read images and text, such as Claude Fable 5.2 (see the Note "Three working devices") and an open model such as MiniCPM-V; Python for scoring.
First step
Ask 20 local restaurants for permission, photograph their menus, write the schema and have two people key five menus separately.
Done looks like
At least 300 menu photos from at least 100 restaurants in at least three languages, holding at least 5,000 items and 1,000 option groups, keyed separately by two people, published under an open license with permission from each restaurant, and a table of item-level and option-level accuracy for at least three models.
Open 02 · the brief
Screen reader ordering test
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A repeatable test of whether blind customers can order on restaurant websites and apps. Testers who use screen readers build one set order on each site, an item with two changes such as a size and a removed topping, and stop at the payment page, so no order is placed and no money is spent. Log whether they reach the payment page, how long it takes and every step that blocks them.
Why it is still open
The 2019 Robles ruling held that the ADA covers Domino's ordering site and app after Robles could not order a customized pizza, but it set no technical standard. Guides for restaurant sites name item options, pickup time pickers, cart updates and checkout as the usual barriers, and give no measured completion rates. AccessMenu, a tool tested with 10 blind users at the W4A accessibility conference in 2025, measured how well they could read and question restaurant menus, not whether they could place an order. A blind writer on the blog of UsableNet, an accessibility firm, said in May 2026 that item options "break in predictable ways across almost every chain I have tried". A search found no published test of how often screen-reader users complete restaurant orders.
cdn.ca9.uscourts.govwpslash.comws-dl.blogspot.comblog.usablenet.com
Built with what exists
The free NVDA screen reader on Windows, VoiceOver on iPhone and TalkBack on Android; the axe-core accessibility checker; Playwright to record each flow; an AI browser agent such as Browser Use 1.0 to repeat the scripted order between human sessions (see the Note "Deleting 80% to start over"); paid testers recruited through blind people's organizations; a public spreadsheet.
First step
Write the set order and the definition of a blocking fault, and run it yourself with NVDA on ten chains' sites, logging every fault.
Done looks like
The ordering flows of 50 large chains and 50 common ordering platforms each tried by at least five screen-reader users, with a public table of completion rate, median minutes and blocking faults per flow, repeated six months later.