This third article in a series on AI will build on previous articles by focusing on the types of predictions and related example use cases while drawing a contrast to casino problems AI is not yet able to solve.
All of a sudden, Artificial Intelligence is everywhere. Ask Siri for directions and a route appears, type an idea for a baroque novel and a story is generated with twists and elaborate subplots. The results can sometimes be dazzling.
Many casino operators feel that AI should be important but are unclear how. In this, the third article in a series of AI uses in the casino industry, we will explore realistic applications of AI. In previous articles we explored the importance of data to AI applications and introduced AI’s role as well as related myths.
In exploring the application of AI, it is helpful to organize AI approaches into two types: pattern matching (what AI pundits call supervised learning) versus pattern discovery (often called unsupervised learning). In pattern-matching we have examples of known behaviors and demographics and the corresponding value (labels) we are trying to predict.
As an example, we recently supported a property in developing a tool which used pattern matching to predict if a new player is a likely high-potential player. For instance, we know that slot players who live in a particular area with an average bet above $3 will likely have a high annual spend (predicted label).
Alternatively, in pattern discovery we simply have behaviors but not a specific value we are trying to predict (unlabeled). The most common application of pattern matching is identifying naturally occurring groups of players who share similar preferences or connections.
A third bucket of AI predictions is emerging: generative AI. Generative AI includes applications like large-language models. Watch this space for a subsequent article devoted entirely to generative AI.
We always recommend to properties that AI (at least today) simply provides a more accurate assumption within a larger problem. Let’s take a churn example:
Before AI we often structured a churn campaign based on average assumptions. For instance, based on analysis we know that 90% of patrons who haven’t visited in the past 365 days will never come back. Armed with this insight, we then constructed a churn program where any player missing for over 365 days falls into a churn campaign. The problems with this approach are legion:
Example of assumptions better calibrated via AI
Mass personalization for marketing
Mass personalization is the most obvious and fundamental use case for AI in the casino industry. Simply, we train a model on patrons we know a lot about and then use this model to guess at preferences for players we know less about.
In the churn example outlined above, “pattern matching” behaviors predicted by AI include the churn likelihood which is being used to identify an audience of potential defectors. Pattern matching AI is also calibrating elements of the offer which are optimized to profitably deliver the change in behavior we are seeking. 
Sometimes predicting future behaviors derived via pattern matching AI can be foundational, meaning the prediction is used as an input to other models while in other cases predictions fit a very narrow use case.
Foundational future behavior predictions may include player valuation, promotional sensitivity and typical visitation patterns. Narrow cases might be calculating offer values, predicting response rates, curating a special event list or even approving credit requests.
Predicting future behaviors via pattern matching AI allows us to personalize values but often we want to create unique experiences or identify communities active on the floor. AI further gives us a rigorous framework to personalize unique experiences by finding similar players who have known preferences. Using the churn example, we might find that similar patrons who are loyal all engage on non-gaming amenities. Thus, our churn campaign/reactivation offer features a hotel and F&B component to this segment. Pattern discovery has both foundational applications like identifying communities of players, identifying key influencers, mapping patron journeys, and also narrow cases like setting preferences for elements in a marketing offer.
Operational applications of AI
While marketing is an obvious low-hanging fruit, AI is helpful in calibrating decisions across the business.
Identifying unusual behaviors
During game play, if a player exhibits odd behaviors and is preternaturally lucky, this warrants at least a pause. We can further correlate the DNA of a patron’s play patterns or deviations from the expected to known scams. For instance, a Chinese patron playing baccarat who encounters four consecutive banker outcomes will bet on banker more than 80% of the time. If, instead, this patron bets on Lucky 6 and wins, the situation is surprising but not necessarily suspicious.
If the corresponding Lucky 6 count is high, the bet was late in shoe, and other players behave similarly, then the behavior lapses to suspicious. Our concluding article in this series will explore AI applications within the domain of integrity, exploring the issues of abuse, leakage and outright fraud.
Forecasting
A few years ago, one of the authors worked at a property which made a mistake in forecasting a seemingly unimportant holiday in China which led to under-spreading the floor significantly. We estimated that this forecasting foible reduced EBITDA on the day by ~30%: a product of foregone revenues and increased over time. Forecasting, long the dismal science of the casino industry, is being reshaped by more accessible algorithms and better-quality data (like the data from smart tables). These improvements in forecasting (an example of pattern matching) hold promise for optimizing pricing, yielding casino hotel room usage, staffing and the like.
Cost right sizing
We can use pattern discovery for identifying journeys and the related touch points which shape the experience for patrons who constitute most of the revenues. These critical touch points are more precious to the success of the organization and should be right sized with care.
Robotics and automation
Robotics brings AI to life. When we think of robotics and automation, the image that often comes to mind is a robot delivering food in a sushi restaurant or a robot vacuum we pass at an airport terminal – novel, but somewhat limited in scope. However, robotics and automation go far beyond these examples and are ultimately about solving practical problems.
One of the authors encourages companies adopting robotic automation to approach utilization like casino productivity analysis. For instance, just as productivity in table games depends on factors like deal rates and pricing, turning rooms efficiently while at the same time personalizing service is also fundamental, albeit a major challenge at a large property. In this example, integrating AI insights from CRM systems with housekeeping robot assistants can create seamless service.
What is a bridge too far today?
At least today, AI has specific limitations. While it can fill-in the blanks on an offer, for example, AI cannot yet make that creative leap to construct the offer itself. In this example, offer creation today requires an understanding of resources, player objectives, as well as style. Generative AI may close this gap in the future and this is the subject of our next article.
