BRENNEN

Industries / Retail, hospitality and personal services / Retail and e-commerce / Convenience stores and fuel retail

Food service

Food service here means food and drink prepared or finished at the store, from coffee and hot dogs to pizza, sandwiches, rice balls and boxed meals.

6

Milestones achieved, from 1984 to 2025.

1

To watch for: announced, not yet confirmed outside.

2

Open deliverables. Nobody has shipped them.

The road

AchievedWatch forOpen

Achieved

  1. 1984

    Pizza at the fuel stop

    Casey's General Stores, a chain whose stores sell fuel and are mostly in towns of fewer than 20,000 people, began selling made-from-scratch pizza in 1984. By April 2002 pizza was sold in 1,290 of its 1,334 stores, and by April 2026 in almost all of its 2,944 stores, where prepared food and dispensed drinks kept 58.6 cents of every sales dollar as gross profit in the year to April 2026.

    sec.govsec.gov

    Credit

    Casey's General Stores

  2. 1993

    Touch screens take orders

    At the Sheetz chain, store manager Earl Springer began making sandwiches to order in 1986 at the store in Williamsport, Maryland. Sheetz installed its first touch-screen ordering menu in Altoona, Pennsylvania, in April 1993 and had them in all stores by 1996, according to the Altoona Mirror.

    altoonamirror.com

    Credit

    Earl Springer and Sheetz

  3. 2009

    Discount ban ruled unfair

    On 22 June 2009 the Japan Fair Trade Commission ordered Seven-Eleven Japan to stop pressing franchisees not to discount perishable food such as boxed lunches, rice balls and sandwiches. Under the chain's system the franchisee bore the cost of food that expired unsold, and head-office staff had suggested that owners who kept discounting could lose their contracts. The company then agreed to pay 15% of franchisees' costs for expired perishable food from July 2009.

    LessonWho pays for unsold food, and who may mark it down, shapes how much is thrown away.

    amt-law.comausfoodnews.com.au

    Credit

    Japan Fair Trade Commission

  4. 2009

    Waste reports become yearly

    From fiscal 2009, Japan's Food Recycling Law has required every food business that generated 100 tonnes or more of food waste in the previous year to report its waste and recycling to the government each year. The reports give a national count for convenience stores: in fiscal 2024 the reporting convenience-store businesses generated about 196,000 tonnes of food waste and recycled about 78,000 tonnes, and their rate under the law, which also credits waste avoided, was 63%.

    maff.go.jpmaff.go.jp

    Credit

    Ministry of Agriculture, Forestry and Fisheries (MAFF)

  5. 2024

    AI suggests the markdown

    From March 2024 Lawson rolled out nationwide an upgrade to the ordering system used in about 14,500 of its stores in Japan. A new AI model picks which short-life items, such as boxed lunches, rice balls and baked bread, to discount, by how much and when, and staff apply the stickers. In June 2026 the news site Impress Watch reported that near-expiry items in Lawson stores carry yellow stickers showing the AI-recommended discount.

    newswitch.jpwatch.impress.co.jp

    Credit

    Lawson

  6. 2025

    Food leads in-store sales

    NACS defines food service as prepared food, commissary food (made in a central kitchen) and dispensed hot, cold and frozen drinks. In 2025 it made up 28.5% of US convenience stores' in-store sales, the largest category, ahead of packaged drinks at 18.7%, and 38.9% of in-store gross profit dollars. In 2005 its share of in-store sales was 11.9%. Prepared food was 73.9% of food service sales, up from 66.4% in 2021.

    convenience.org

    Credit

    NACS

Watch for

  1. A robot fries the chicken

    In a Lawson store that the chain opened with the telecoms company KDDI at Takanawa Gateway City, Tokyo, on 23 June 2025, a robot cooks Karaage-kun, Lawson's fried chicken bites: it checks the oil temperature, lowers the frozen pieces into the oil, shakes the basket and drains it. Lawson's chief executive, Sadanobu Takemasu, set a goal of 30% less labor and 30% more sales for such stores; no results have been published.

    Achieved whenfrying robots are in routine use in many stores, and an outside party reports labor time and food waste against stores without them.

    gourmet.watch.impress.co.jpbcnretail.com

    Claimed by

    Lawson and KDDI, June 2025

Open

  1. 01

    Count what hot cases waste

    A published method and a first open dataset of hot food thrown away at convenience stores: each item cooked, sold and discarded, by hour, logged for eight weeks in at least five stores that agree to take part. Then use the log to test an open forecasting tool against each store's usual cooking plan.

    The brief
    Credit · unclaimed
  2. 02

    Hot holding in inspection records

    An open table of how often stores that sell fuel are cited for hot food kept too cool, compared with other food stores, from New York State's public inspection records. An AI model reads each deficiency note and answers fixed questions: is it about holding hot food, was a temperature taken, which food. A person checks a sample. Stores are matched to OpenStreetMap fuel stations by location.

    The brief
    Credit · unclaimed

Open 01 · the brief

Count what hot cases waste

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A published method and a first open dataset of hot food thrown away at convenience stores: each item cooked, sold and discarded, by hour, logged for eight weeks in at least five stores that agree to take part. Then use the log to test an open forecasting tool against each store's usual cooking plan.

Why it is still open

Public figures are totals. Japan's reports give convenience-store businesses' waste in tonnes for the whole country; Casey's says its prepared-food margin rose "driven primarily by improved waste" without giving a waste rate; and an open hot-food forecasting tool on GitHub, last updated in October 2026, is tested only on synthetic data.

maff.go.jpsec.govgithub.com

Built with what exists

A paper or tablet log at the hot case; a kitchen scale; the stores' own sales records; the open hotfood-forecast package (Python and scikit-learn); optionally a fixed camera over the discard bin with an image-recognition model to count items, checked against the log.

  1. First step

    Write a one-page logging sheet for one hot case and run it for a week in one cooperating store.

  2. Done looks like

    For each store and item type, items discarded as a share of items cooked, with a confidence interval; and a backtest showing how many fewer items the forecaster would have discarded, and how many more sell-outs it would have caused, over the eight weeks.

Open 02 · the brief

Hot holding in inspection records

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

An open table of how often stores that sell fuel are cited for hot food kept too cool, compared with other food stores, from New York State's public inspection records. An AI model reads each deficiency note and answers fixed questions: is it about holding hot food, was a temperature taken, which food. A person checks a sample. Stores are matched to OpenStreetMap fuel stations by location.

Why it is still open

The best-known ranking, which put hot holding sixth among problems in more than 2,000 convenience-store audits in 2018, came from the private audit firm Steritech's own data. New York's open dataset of the Department of Agriculture and Markets holds 157,743 deficiency records from inspections in the last 24 months, written as free text; a word search for hot holding or hot temperatures returns 1,553 rows, many about other matters, such as roller-grill tongs that were not tethered, so no simple count exists.

food-safety.comdata.ny.gov

Built with what exists

The data.ny.gov API; OpenStreetMap amenity=fuel points; Python; an AI model that answers fixed-choice questions about each record, such as Gemini Decision 1 (see the Note "A thousand questions, one pass").

  1. First step

    Download the deficiencies that mention "hot" and hand-label 200 of them as a test set.

  2. Done looks like

    Hot-holding temperature citations per 100 inspections for stores that sell fuel and for other stores, with the AI labels at least 95% correct on 300 hand-checked rows, the code and labels public, and the table rerun every quarter.