Industries / Media, arts and recreation / Media and entertainment / Movie theaters
Showtimes and screens
Programming decides which film plays on which screen at what times each week, and the number and shape of a cinema's auditoriums set the limits of that choice.
7
Milestones achieved, from 1905 to 2017.
2
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1905
Short shows on a loop
On 19 June 1905 Harry Davis and John P. Harris opened the Nickelodeon in a storefront on Smithfield Street in Pittsburgh. It sold 15- to 20-minute programmes of short films for a nickel each from 8 a.m. to midnight, usually with a pianist playing along; 450 people came the first day and more than 1,500 the next. The Theatre Historical Society of America calls it the first stand-alone movie theatre in the US and gives an estimate of 8,000 nickelodeons in the country by 1908.
Credit
Harry Davis and John P. Harris
1933
Cars become the seats
Richard M. Hollingshead Jr. of Riverton, New Jersey, was granted US patent 1,909,537 for a "drive-in theater" on 16 May 1933. Rows of ramps tilt each car upward so the people inside see the screen over the car in front, and he wrote that it would let people, "particularly in rural sections," watch a film without leaving the car.
Credit
Richard M. Hollingshead Jr
1963
Two screens, one lobby
Durwood Theatres, later AMC, opened Parkway Theatres One & Two in the Ward Parkway Shopping Center in Kansas City on 12 July 1963, with 450 seats in all and The Great Escape on both screens. AMC presents it as the first multiplex; contributors to the Cinema Treasures database point out that Cinema I & II in Manhattan opened the year before and that General Cinema had twin cinemas by mid-1963.
Credit
Stanley Durwood and Durwood Theatres
1995
Megaplexes, then bankruptcies
AMC opened a 24-screen cinema in Dallas in 1995, and chains raced to build megaplexes of 14 to 30 screens with stadium seating, rows stepped steeply so no head blocks the view, at about $1 million per screen. US screens rose from about 24,000 in 1990 to about 37,000 in 2000, growing about 8% a year while attendance grew 2.6% a year, by Prudential Securities analyst James Sullivan's figures. Six of the ten largest US chains filed for bankruptcy in 2000 and 2001, by the American Bankruptcy Institute Journal's count.
LessonBuilding screens faster than audiences grew left the chains unable to pay for them, and they used bankruptcy to walk away from theatre leases.
Credit
AMC Entertainment
2009
Schedules from a demand model
Jehoshua Eliashberg, Charles B. Weinberg, Berend Wierenga and five co-authors published a system that forecasts the visitors for every possible start time of every film, then searches for a near-best weekly schedule of films, screens and times. It was tested in a multiplex in Amsterdam for fourteen weeks, and the authors report that its schedules drew more visitors than the cinema's "intuition-based" ones.
Credit
Jehoshua Eliashberg, Quintus Hegie, Jason Ho, Dennis Huisman, Steven J. Miller, Sanjeev Swami, Charles B. Weinberg and Berend Wierenga, International Journal of Research in Marketing
c. 2016
Software drafts the schedule
Vue, a UK-based cinema chain, began using its own algorithm-based programming tool, AI scheduling (AIS), more than ten years before April 2026. It helps a central team schedule about 30,000 sessions a week across Vue's UK cinemas, was extended to Vue's Italian cinemas about three years before that date and is due in Poland, Screen Daily reported. Vue says an average eight-screen Vue now shows as wide a range of films as a 14- or 15-screen cinema run by a rival.
Credit
Vue International
2017
China builds the most screens
In June 2017 Zhang Hongsen, deputy director of China's film and broadcasting regulator (SAPPRFT), said China had more than 45,000 cinema screens, more than any other market, after adding an average of 19 a day since 2012, as TheWrap reported. China had 40,917 screens on 20 December 2016, when North America had 43,531.
LessonThe Associated Press found new small-town cinemas outside Beijing playing to rows of empty seats in January 2017; screens can be built faster than audiences come.
Credit
China's cinema operators; figures from SAPPRFT
Watch for
Forecasts sold to cinemas
Vista Group, a New Zealand software firm that estimates its own share at 46% of large cinema circuits outside China and India, says its Box Office Forecasting tool predicts a film's domestic opening weekend "to an accuracy level of +/-15%", and that its Assisted Scheduling tool, built with the chain Pathé, cuts the time spent scheduling by up to 50%.
Achieved whenAn outside party publishes a test of the forecasts on a full year of releases, or a chain publishes logged scheduling hours before and after adopting the tool.
Claimed by
Vista Group, June 2025 and April 2026
A manager's own scheduling agent
Sergio Chavez, manager of the two-screen municipal cinema in Tynset, Norway, built his own scheduling and marketing agents with Anthropic's Claude, using national cinema data and the cinema's own sales. Admissions rose 21% in a year to 14,000, against average growth of 2.5% for Norway in 2025, and he credits much of the rise to the agents.
Achieved whenThe gain holds for a second year in national admissions statistics and is repeated at other cinemas that use the same method.
Claimed by
Sergio Chavez, Tynset Kino, April 2026
Open
01
An open show-level benchmark
A public dataset of single screenings, giving film, cinema, auditorium size, start time and tickets sold, donated by cooperating cinemas with all customer data removed, and a benchmark that scores forecasts of each show's attendance one week ahead.
The briefCredit · unclaimed02
Alternate AI and human schedules
A controlled trial at a few cinemas. In alternate weeks the published schedule is either drafted by an AI agent from past attendance and approved by the programmer, or written by the programmer alone; admissions are compared week by week, adjusted for which films were on release, and programmer hours are logged.
The briefCredit · unclaimed
Open 01 · the brief
An open show-level benchmark
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public dataset of single screenings, giving film, cinema, auditorium size, start time and tickets sold, donated by cooperating cinemas with all customer data removed, and a benchmark that scores forecasts of each show's attendance one week ahead.
Why it is still open
Published studies of attendance per show use one chain's private data: 179,103 shows from a Polish cinema network (Paweł Baranowski, 2020) and five years of transactions from an Indian multiplex chain (Sundararaman Venkataramani and co-authors, 2019). This research found no released dataset from either. The one public set found, Cinema-Tickets on Kaggle and OpenML (142,524 rows from about eight months of 2018, licensed for non-commercial use only), replaces every film and cinema with a code and hides the locations, so a forecast cannot use what the film is or where it plays, and it comes with no fixed test period. Vista's forecast accuracy claim has no public test.
Built with what exists
Exports from the cinemas' own ticketing systems; public film details from The Movie Database (TMDB) API; baselines from Baranowski's regression models, gradient-boosted trees in LightGBM, and a small model fine-tuned on one graphics chip such as Liquid AI's LFM3 (see the Note "One GPU to fine-tune").
First step
Ask three independent cinemas for one year of anonymised show-level sales exports and agree on a shared column format.
Done looks like
At least 10 cinemas in at least 3 countries and at least 100,000 shows over 12 months, a fixed held-out test period, and the mean absolute error in tickets per show published for at least three baseline models.
Open 02 · the brief
Alternate AI and human schedules
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A controlled trial at a few cinemas. In alternate weeks the published schedule is either drafted by an AI agent from past attendance and approved by the programmer, or written by the programmer alone; admissions are compared week by week, adjusted for which films were on release, and programmer hours are logged.
Why it is still open
The 2009 Amsterdam test ran fourteen weeks at one multiplex. Vista's 50% time saving and Tynset's 21% admissions rise are the users' and makers' own reports, with no comparison against weeks scheduled by hand.
Built with what exists
An AI model such as Claude Fable 5.2 (see the Note "Three working devices") to draft schedules and explain them; Google OR-Tools, an open-source optimiser, to fit shows to screens, running times and cleaning gaps; the cinema's own ticketing exports.
First step
With one independent cinema, draft AI schedules in parallel with the real ones for four weeks without publishing them, and log how far they differ and how long each took.
Done looks like
At least 3 cinemas running 20 alternating weeks each, with the analysis plan published before the trial starts; the difference in admissions reported with a confidence interval, plus programmer hours per week under each method.