Consumers in the gambling and gaming industry typically roll the dice, but so does the industry itself. Organisations are faced with daily threats to customer safety and satisfaction as well as commercial profitability.
Gamblers often rely on hunches or intuition, while the House prefers hard facts. Ultimately though, they’re
both all about prediction. Fortunately for online gambling and gaming companies, they have access to a wealth of big data created by every click of a customer’s mouse. The key to extracting valuable predictive insights from that data will be sophisticated machine learning.
Why machine learning?
Machine learning means the ability to learn relationships and patterns within data without being explicitly programmed. It requires large datasets and it requires planning. Different companies have different priorities and goals behind developing machine learning algorithms. One may want to harness player data to inform and improve game design, whereas another company may be more interested in maximising revenue and identifying the players most likely to spend money.
Let’s use the example of a business that wants to tackle the problem of addictive gambling behaviour in order to keep customers gambling safely, and within the perimeters of regulation. Machine learning algorithms are a good solution as they ‘learn’ patterns and correlations from vast historical datasets of past player behaviour and can then predict future outcomes. A key example of this would be whether a player is addicted or not.
In the case of spotting addictive behaviour, a gambling company can build a profile of what constitutes normal behaviour for each player and machine learning algorithms will identify deviations to the normal behavioural patterns. This can be used to alert a gambling or gaming company when a player exhibits addictive habits so that the company can potentially intervene and take corrective action.
Fraud detection and credit risk
In online gaming, there is often a large volume of credit card payments. Many companies also offer newly registered users free credit as an incentive.
This means the motivation is there for some players to try and abuse these offers with multiple or fake accounts – a practice known as ‘bonus abuse’ – effectively defrauding the company. At significant enough scale, this could even feed into the company’s credit risk.
Other fraud risks include legitimate accounts being hijacked, or stolen credit card details being used to place large bets.
Fortunately, ‘learn’ patterns can be used to identify and prevent abuse by building a picture of normal account activity and flagging up suspicious, and potentially fraudulent, patterns.
Analytics for Anti-Money Laundering (AML)
There is an increasing amount of regulatory compliance pressure applied to casinos to reduce risk, especially when it comes to money laundering. In fact, casinos are regularly fined millions of dollars for flouting AML laws.
Therefore, much like banks, gaming and casino companies stand to gain a lot from automating their processes for combatting AML. Automated detection software can help to increase the detection rate of suspicious activity, while reducing the investigation time. By aggregating patron and transactional data, compliance staff can more quickly get to the root cause of suspicious activity.
Machine learning can boost automated detection software and, keep up with money launderers as they switch tactics or change their patterns. It would mean businesses stayed one step ahead rather than waiting for the software company to spot the trend and send out an update to address it.
This would cut down laundered funds slipping through the net, and also demonstrate proactivity to the regulators.
Machine learning models can broadly be categorised into three types: clustering, classification and regression. The gambling examples above could be achieved by a classification model, in which the algorithm identifies which class a data observation belongs to out of a set of pre-defined classes. For example, these algorithms can be used to predict whether a customer is addicted or not; whether a player is a bot or a real player; or whether a customer is likely to deregister or not.
Regression models, on the other hand, find relationships between two or more variables and predict a numeric value, such as how many players will be online at 7pm on Friday or how much a player is likely to spend in their lifetime.
Lastly, clustering models identify similar instances and group them into clusters. This is often useful for recommendation algorithms, where it is possible to recommend relevant information to a player based on the similar preferences of those in their cluster. It is also a useful tool for data exploration as it automatically highlights commonalities within certain groups of players. It enables detection of extreme or fraudulent behaviour where the observation is anomalous and falls outside of the cluster groups.
Machine learning can give online gambling and gaming companies a major boost commercially and help them to act responsibly and compliantly by predicting problem behaviour before too much damage is done. It requires significant investment of time and resources but machine learning is a safe bet for those that get it right.