“Moneyball: The Art of Winning an Unfair Game” is a book written by Michael Lewis about the Oakland Athletics baseball team and how they successfully used analytical, evidence-based decisions to select a competitive team with a very restrictive budget. Both Michael and the general manager at the time, Billy Beane, gave us an entertaining insight into the world of sport and mathematics at last year’s IBM Information On Demand conference in Las Vegas.
So naturally, when we set the agenda for this year’s Australia & New Zealand Business Analytics Forum we’d hoped to enlist the star of the recently launched Moneyball movie, Brad Pitt, to attend.
Clearly, we were dreaming.
When that fell through, we asked ourselves would this ability to apply analytical, evidence-based decision making work in other sports? Held the same week as the Australian Grand Prix in Melbourne, we decided instead of presenting our usual widget sales demonstration to our keynote audience, we would put IBM’s predictive analytics platform, SPSS, to the test and see if we could accurately predict the outcome of the race.
Using statistics and analytics is not new to the world of motor sport. In fact, the elite racing teams have groups of analysts dedicated to evaluating the information collected from sensors spread throughout the vehicle in order to understand performance through every straight, around every corner and every pit stop. They then use this insight to find areas for improvement and competitive advantage.
Using historical information about past races, drivers and team performance, we were able to configure a predictive model to uncover the hidden trends and patterns in the data. In prior years, we would have placed our bets based on gut feel, who won last year, and other “soft” indicators of success. However, with the help of IBM predictive analytics, we uncovered a greater level of understanding about the influence of qualifying times and the teams that drivers race for.
For example, our results indicated that of those drivers that qualified in the top four positions, that went on to finish the race, and drove for a specific sub-set of teams, 75% of them were likely to finish on the podium. The model then used these rules to predict the likelihood of each driver finishing on the podium in the Australian Grand Prix. Unfortunately, at the time of our keynote, the qualifiers had yet to occur so we had to simulate results in order to get a prediction. After the event, for our own curiosity, we re-ran the prediction using the actual qualifying results.
For the Australian Grand Prix, the top three drivers that were predicted to finish on the podium, did in fact finish in the top three positions. The Malaysian Grand Prix was a little trickier – with severe weather conditions that led to the early termination of the race, our model still managed to accurately predict one of the podium drivers for the race.
For the Chinese Grand Prix….well, you’ll just have to wait and see!
Whilst we embarked on this project for entertainment, it did in fact illustrate that using predictive analytics to support evidence-based decisions led to us having a better outcome – in this case a winning bet.
Now if only my husband would stop asking me to evaluate his Supercoach team selections!