Simulation Based Decision Making
By Tony Gruebner, GM Analytics, Insights & Modelling, Sportsbet
As the trade deadline approached for the 2017 NBA season, Lebron James knew his team needed some fine-tuning if he was going to take it to his eighth consecutive NBA finals appearance.
James is one of the smartest tacticians in the game, both on and off the court, despite a recent Donald Trump tweet to the contrary. As well as being one of the greatest players of all time, he is also the unofficial Head Coach and GM of any team he plays for. And often when asked by reporters how he makes decisions about his team, his revelation is surprising – in part, by frequently playing NBA 2k, the market leading basketball video game, “I mix and match a lot of line-up changes and things of that nature to see how we can be really good”.
Either wittingly or unwittingly, James uses simulations, an increasingly popular method of analytics that enables both humans and machines to fine-tune their decision making through virtual models to derive better outcomes. simulations are a technique that we use extensively at Sportsbet across a range of different business purposes such as digital marketing, outbound marketing, scheduling & forecasting, pricing and risk practices to name a few.
The flagship Sportsbet example of generating business value through simulation modelling is our Same Game Multi Product. By way of background, a Multibet, also referred to as a ‘Parlay’ or ‘Accumulator’ is a very popular bet type where the customer combines two or more single bets together. If each of the individual bets wins then the punter is paid accordingly. However, historically in the global wagering industry, customers have not been able to place these types of bets on events that occur within a single game as calculating the payout of a traditional Multibet is based on multiplying the odds of independent events together. For example, if a customer places a Multibet that consists of two independent events both with a 50 percent chance of occurring, the probability that both will occur is 25 percent (50 x 50).
Simulations play a key role at businesses and the trend is likely to grow in the future as artificial intelligence techniques expand the capability to digitally model the real world in a cost-efficient fashion that allows for improved decision making
In 2017 we (Sportsbet) became the first bookmaker in the world to offer extensive Same Game Multis with an almost infinite number of combinations available to our customers. And, to achieve this and calculate the odds, my team performance-tuned the models to instantaneously run 10,000 simulations of a game. Using the example above, we look for the number of times in those simulations where the conditions of both Steph Curry scoring more than 40 points and his team winning are met. In the year since its release it has become one of our customers favorite bet types which is amazing for a product that is essentially only possible due to the power of simulation modelling.
Data Products are models where machines make decisions that impact the real world without human input. Other more mainstream examples exist in the form of Self-driving cars, AlphaGo and Amazon’s Warehouse robots, where machines simulate the ultimate result of each decision to ensure that on every occasion they are making the optimal selection.
However, simulations aren’t just for the realm of our machine overlords. A good simulation can also help a human be more informed and therefore make better decisions such as the Lebron James for example. simulating complex scenarios that require multiple decisions with a range of different internal and external variables can help decision makers by allowing them to tweak parameters and perform what-if scenario based analyses.
Often in the modeling there is a trade-off between a model’s accuracy and the ability for a human to understand the model. However, simulations can clearly show their outputs in detail to users and can often allow users to tweak model parameters to provide what-if scenario based analyses allowing them to be more easily verified, communicated and understood. This can be true even when the models that underpin the simulation are extremely complex or black box. In fact, websites like FiveThirtyEight and the Upshot by the New York Times prosper by taking complex simulation models that predict complex questions such as who is going to win an election or which teams will make the NFL playoffs and then applies clever interface that allow the models to be easily consumed.
Simulations already play a key role at businesses like Sportsbet and many others. And this trend is likely to grow in the future as improving technology and artificial intelligence techniques expand the capability to digitally model the real world in a cost-efficient fashion that allows for improved decision making. It’s also worthwhile noting that as a society, we are also getting more comfortable with “trusting” computers, and it’s this acceptance together with the advancement in technology that will see simulations gain exponential mainstream usage.