Pathé let SilverScreener's math decide which movies keep their screens

Every Monday morning, Pathé Theaters in the Netherlands faced the same decision: which movies currently playing in its cinemas should stay, and which of the available replacements should take their screens. Managers made these calls on judgment, and each individual call looked small — but together the keep-or-kill decisions set what the whole multiplex earned.

Researchers from Wharton, Erasmus and elsewhere implemented SilverScreener, a mathematical-programming system built on Swami, Eliashberg and Weinberg's 1999 movie-theater model, for one six-screen Pathé theater. They first developed an attendance-forecasting system from Pathé's historical data, managerial judgment and theater-specific factors; the optimizer then selected retained movies and replacements from the titles actually available.

A fully controlled experiment was not possible, so the test was honest but imperfect: the SilverScreener theater's revenues were compared against two similar multiscreen cinemas still run unaided. The theater using SilverScreener's recommendations earned more than both, and managers' attitudes toward the modeling system improved after they had worked with it.

The takeaways are twofold. First, the value came from attacking the whole weekly allocation at once, not from a better forecast of any single film. Second, adoption followed measured performance — the same managers who were skeptical of a scheduling model warmed to it after the revenue comparison.

Forecast each film, then let math allocate the screens

The dumb alternative is booking by instinct: keep the movies that feel popular and hope replacements pan out, treating each screen in isolation.

Splitting the problem into forecasting attendance, then optimizing keep-replace allocation, mirrors how the decision unfolds: prediction where data helps, optimization where trade-offs live.

Restricting replacements to titles actually available that week made the recommendation executable; a schedule ignoring distribution reality is a suggestion, not a decision.

Comparing against two comparable unaided cinemas gave managers a number to argue with; attitude change followed evidence, not persuasion.

The weekly cadence matters: a modeling investment amortizes over dozens of decisions a year, not one big bet.

Dozens of small weekly choices set a venue's revenue; forecast-plus-optimization beats intuition on repeated allocation calls — and a live comparison against unaided peers earns managers' trust.

The implementation paper appeared in Interfaces in 2001 as the field follow-up to the 1999 SilverScreener model of Swami, Eliashberg and Weinberg, and became one of the standard references on decision support for movie exhibitors, with the model's forecast-then-optimize structure cited in later work on theater management. Pathé remains the Netherlands' largest exhibition chain.