It’s one of technology’s sexiest topics—machine learning—conjuring up images of Watson playing Jeopardy!, self-driving cars, and “I’m sorry Dave, I can’t do that.”
But these examples feel—even to this author—a bit futuristic and inaccessible. What can machine learning do for me now?
Actually, most people interact with machine learning every day. Netflix recommended I watch Sherlock. Amazon recommended a Michigan Snuggie.
A friend wrote an email about a restaurant, and Gmail offered some pre-populated one-click responses:
These are all examples of machine learning in action. “Alexa, what’s machine learning?” Indeed—Alexa, Siri, and Cortana all use machine learning to mimic human conversation. Voice-to-text? Machines use large training sets to learn how people speak. The text they produce is a prediction of what we’re saying.
At its core, machine learning is processing data, “learning” from the data, and making predictions based on the data. Netflix, Amazon, Gmail and Alexa use data to predict what I want to watch, buy, type and hear. But we have slot machines. Dice. Cards. How can we use machine learning?
To answer this, it helps to understand the types of problems machine learning solves: classification and clustering. A classifier might aim to predict whether a picture is of a cat or a camel, or whether an email is spam or “ham”—yes, that’s really what we call non-spam emails. The classifier might learn that misspellings, words like “buy,” and certain pharmaceuticals are good predictors of spam. A clustering algorithm might learn that I probably want to watch the latest season of Sherlock, or that people who buy diapers at the grocery store also buy beer—something to which this new dad can attest!
In the gaming world, a classifier might…
- Figure out that certain advantage players often have PO boxes or out-of-state addresses, a large theo-to-actual disparity, and play certain game types.
- Predict that a slot’s bill validator will break next week.
- Learn that a slot does $300 theo per day, and flag the $2,000 theo it did in the last 90 minutes, despite losing money, so that a tech can investigate and find that we erroneously reset the mini-progressive too high.
- Predict that a patron won’t play for the next three months, allowing marketers to confidently send a more aggressive offer.
Meanwhile, a clustering algorithm might…
- Identify a set of 15 Aristocrat games, located all across the gaming floor, that tend to be played by the same set of people (diapers–beer example above). This may encourage us to bank those games together.
- Tell us that a player really likes five of those games but hasn’t yet played the other 10. We can include recommendations and images of those machines in her next email correspondence.
Especially in marketing and on slot floors, where there’s an abundance of data and where the revenue and cost numbers are large, many companies have begun developing internal teams or working with vendors to turn their data into intelligence, optimizing both their top and bottom lines. Machine learning algorithms can even help analyze guest satisfaction surveys. This is what machine learning is doing for the industry now.
But, machine learning is an evolving tool in this process, and there’s a lot farther to go to leverage this technology to its fullest. It is not easy—competition from the tech industry makes it difficult to find talent, and market rates for the skill set are high. However, since machine learning is all about prediction, I will offer one of my own: machine learning will be transformational in gaming. And while we aren’t to the level of true 1-to-1 marketing and in-session bonusing yet (we are close), there are many routine areas of operations where machine learning can make an immediate EBITDA impact.