Racing Victoria (RV) had begun a journey to automate the Victorian Thoroughbred race dates fixturing process (meetings) & the programming of races for each of the meetings. The system would provide instantaneous feedback on proposed changes and undertake predictive analysis to determine future outcomes based on known events and/or assumptions and historical data. The application would require the accurate and optimal scheduling of 550 race meetings across 72 venues per year while adhering to business rules, and take in to account high-dimensional variables such as race type, field size, gender and age of horses among many others.

PAG developed a web-based application that schedules 550 annual thoroughbred meetings in near real time, reducing the vast bulk of the manual scheduling work currently undertaken and allowing RV to focus on schedule refinement. As the calendar maximises key variables including turnover and wagering (while respecting certain constraints) a large set of machine learning and optimisation algorithms have been deployed in the system to allow for accurate forecasting and scenario analysis.

PAG's software increased operational efficiency by reducing the time it takes to create a fixture from 6 months to less than 5 minutes. At the conclusion of each meeting, a series of ML algorithms allow the organisation to isolate and quantify the factors that have contributed most heavily to the success (or failure) of a meeting, ensuring a policy of continual improvement.

Furthermore, the accuracy of RV's forecasts of key variables such as Turnover and Gross Revenue increased substantially (less than a 2.5% error rate from forecast compared to actual).