Scenario: Elevator company's factory, sales and installation teams from across the country meet monthly to decide which customers' elevators to manufacture in the coming months. Because of various reasons, the elevators that actually ship the following months is not exactly as scheduled. The accuracy of the scheduling depends on how many months away the delivery is scheduled, but historical rates have been in the 60s and 70s for delivery scheduled two months away.
AI: by training on customer's seven years' worth of historical data, we built a model to predict whether deliveries would be delayed or not. Our overall accuracy for a year and a half of test data improved over the customer's historical rates slightly.