Why gambling businesses are betting on machine learning

Online gambling businesses have access to an impressive volume, variety and velocity of data. This data has a wide range of functions within the industry, maximising revenue and identifying the players who are most likely to spend money, to preventing addiction, provide recommendations, and hone their customer targeting and brand focus.

The key to unlocking the value of this data, however, lies firmly in the hands of machine learning. It is this that turns the reams of unstructured data gathered every day into usable, structured information to inform decisions across the company.

Mobile First Strategy

The vast majority of online gambling businesses now takes place via mobile and tablet. As such, the mobile-first or mobile-only approach is central to the gaming and betting industry. Machine learning has helped considerably with helping to optimise mobile experience for consumers and data for gambling businesses.

Deloitte predicts that over 300 million smartphones (or more than one-fifth of the units sold in 2017) will have on-board neural network machine learning capability.

It is precisely this on-board ML that will help gambling apps to optimise the experience they offer to players. As Deloitte puts it:

“True product differentiation can be achieved by identifying new game formats or features that are specifically developed for mobile and that take into account the way in which consumers use and interact with their mobile devices.

For instance, this could include developing apps that use location-specific data in tailoring content to players, or new games that are based on using the swiping motion instead of traditional cursor clicking.”

One issue with the preponderance of mobile in the gambling industry, however, is the issue of customer loyalty. One player could have multiple apps on their phone from different brands. Customer loyalty is notoriously low. Ensuring that the quality of experience is enough to keep players on one single app is a challenge that can only be solved by coordinating product offering with information gleaned from datasets translated by a machine learning algorithm.

Categories of Algorithms

The machine learning models necessary to perform these vital tasks largely comprise three distinct categories: the Classification model, the Regression model, and the Clustering model. It is through a brief description of these three types of model that we can begin to understand how machine learning provides data insight.

Classification

The Classification-type algorithm identifies which class a data observation belongs to, segmenting the data out of a set of predefined classes. These classes attribute a set of actions undertaken by the player to predict likely scenarios. For example, the algorithm can identify activities that tend to mean that a player is expected to deregister. By sending an alert for a human, or another algorithm, to perform a remedial action to avoid this eventuality (e.g. perhaps by offering a bonus sum to the customer’s online profile to encourage them to play on).

The algorithm can also use the data to identify any gambling bots being used. These sort of gambling bots are a kind of AI-equivalent to card-counting (a practice obviously banned in offline casinos, and therefore equally illegal online).

Crucially, a Classification-type algorithm is key to identifying addiction through player behaviour. Online gambling is strictly regulated; online gambling businesses must prove that they take reasonable steps to curb extreme use by moderating and discouraging addictive behaviour. Along with machine learning techniques like ‘random forests’ (which has been found to be 87% accurate at predicting unhealthy behaviour patterns), businesses can ensure they rigorously adhere to regulation.

A good example is the UK firm, Featurespace, which grew out of the University of Cambridge’s engineering department. By using machine learning to help them combat fraud in online gambling, the team were also able to ascertain what constitutes normal behaviour:

“We decided since we’re harvesting so much data for our fraud solution work,” explains co-founder and CTO, David Excell. “How can we use some of that to try to understand the player from a corporate social responsibility point of view, to understand “is that player in control?” and so on.”

Beyond social responsibility, the use of Classification-type machine learning models to identify problem behaviour can also benefit the gambling company themselves. Once a player recognises a gambling addiction themselves, they may take the action of ‘self-exclusion’ (a legally-required option across the gambling industry for gambling addicts to request that the company stops allowing them to access the company’s service). When this occurs, the customer ceases to become a source of revenue altogether. By regulating their use of the platform, the company can keep them spending but remain within regulatory guidelines.

Regression

A Regression-type algorithm will be used to identify relationships between two or more variables presented within the data. From this, it can ascertain or predict a numeric value. The applications of this are to help the business understand the most popular engagement times, such as how many players are online at a particular time of day, or how much a player might spend throughout their membership. This insight that can be monitored over time to optimise user experience and maximise revenue.

Clustering

The Clustering-type algorithm is similar to the Classification algorithm, in that it groups identical instances into clusters. This is the type of algorithm used to make recommendations, such as those you see on Netflix or Amazon.

Not only does the Clustering-type algorithm allow online gambling businesses to make recommendations, but it is also a useful tool for data exploration. The algorithm highlights commonalities within specific groups of players. In this way, it can be as helpful as the Classification model in identifying fraudulent or extreme behaviour by flagging anomalies from the norm within the cluster group. Human employees are then able to intervene at an appropriate point.

While segmentation of customer data has been a critical element of businesses across industries for some time, the task has been relatively manual up until now. The issue of human error means that segmentation data is not always as accurate as it could be. Machine learning reduces error exponentially, thus providing more precise feedback which is crucial to businesses and online gambling in particular.

When armed with a combination of using big data and machine learning algorithms to decipher it, online gambling companies stand to gain dramatically. Both gaming and online gambling are highly automated industries themselves, so taking the power of that automation even further is, most certainly, a safe bet.

If you work for a gaming and betting business and would like to discuss ways in which machine learning can help your business, please do not hesitate to contact me.

[email protected]

@DanAtLogikk