A Week of Showtimes, Laid Out by Hand
A cinema chain runs more than twenty sites with auditoriums of several categories and projection formats. Every week a small team lays out about four thousand showtimes by hand. Times and auditoriums are placed first, prices and formats after. The work takes days every week, plus a revision once the first day of a film's run is in.
The scheduling rules live in the planners' heads. What prime time is measured from, how far apart neighboring auditoriums must start, which film plays where: none of it is in a document. Prices follow a weekday and weekend grid and rarely move, and once a week someone checks what nearby cinemas charge and notes it down.
Limits of an Attendance Forecast
- A forecast scores a finished schedule. It answers how many people will come to a showtime, not which showtimes to place. The planner still lays out the week by hand.
- Showtimes are optimized one at a time. Auditoriums in one cinema compete with each other, and the best choice per showtime does not add up to the best day.
- Room and contract limits are missing. A 2D room cannot run a 3D show, and a distributor's contract sets a minimum number of showtimes per format. A schedule that ignores them cannot be run.
- Accuracy is measured at the wrong level. Weekly accuracy looks convincing, while the error per showtime, where the decision is made, is several times larger.
- Sales history carries outside events. Cancelled showtimes, force majeure and weather have to be labeled, or the model learns from failures the schedule never caused.
Schedule and Price From One Forecast
The solution is designed as two loops computed together: a draft schedule for the coming release week and the price of every showtime. Both rest on one demand forecast and one set of constraints, so the schedule never proposes a showtime that cannot be priced.
- Data. Sales history, pre-sales and outside signals such as weather, search interest and competitor prices.
- Forecast. Demand per showtime and per seat. Premieres with no history of their own use general-purpose time-series models.
- Draft. A schedule for the whole chain: time, auditorium, format and the number of showtimes per film. Operating hours, contractual minimums per format, what each room can project, gaps between start times and family content before 18:00 are hard constraints, and a draft that breaks them is never produced.
- Price. The model estimates how sensitive demand for a showtime is to price and proposes a change inside a corridor.
- Planner. A planner edits the draft and approves the prices, in one action across the chain.
Planners can describe a new scheduling rule in chat, in their own words. It is translated into a formal constraint, checked by the team, and used in every draft after that.
Technology Stack
Fair Prices and Firm Limits
Ticket prices and the schedule carry the chain's revenue and its obligations to distributors. The system prepares the draft and the price list; a planner reviews both and publishes them.
- The whole visit counts. The target is the revenue of the visit, ticket plus concession stand, rather than the ticket price on its own.
- Price depends on seat and time, never on the viewer. Two people buying adjacent seats for the same showtime pay the same.
- Price corridors. Each showtime's price has its own lower and upper bound and a cap on how far it can move in one step, and the base price card can be restored with one click.
- Elasticity from experiments. Price sensitivity is estimated with causal inference and price experiments, not with correlations.
Planning Figures
The numbers in this section set the scale the system was designed around.
- About 4,000 showtimes a week drafted across the whole chain.
- No draft in breach of a hard constraint.
- No price change without a planner's approval.
- Schedule changes no later than a day before the showtime, and no automatic changes at the weekend.
- A daily check of competitors' ticket prices.
Venues That Face the Same Puzzle
Any venue that shares limited time and rooms between titles, with prices tied to the seat and the hour, meets the same questions:
- the schedule is built by hand and takes days, not hours;
- a handful of planners know how it is built;
- prices follow a weekday and weekend grid while demand moves daily;
- some constraints come from contracts with content suppliers.
Concert and theater venues, sports arenas and amusement parks plan this way too. For another operator, scheduling and pricing can also be built one at a time, with the level of automation set together. The pricing side follows the same logic as our AI pricing engine for an equipment distributor, where a person also signs off every change. Human-in-the-loop in our glossary and our predictive analytics consulting page describe the approach.
showtimes a week the solution is designed to draft across the chain
sites covered by one draft schedule