“Artificial intelligence,” or AI, is one of those lofty, vaguely threatening terms that, for some, conjures up images of a looming humans-are-not-worthy techno-apocalypse. 

But the reality is much more practical and, as a session at last fall’s Global Gaming Expo (G2E) demonstrated, increasingly accessible to slot managers who presently use a time consuming do-it-yourself approach to get at answers that AI can solve in minutes. 

The G2E session summarized a case study of AI on the slot floor using findings at a small, tribal property in the northwest. The speakers were Dr. Stasi Baran, COO, nQube Data Science; Jason Fiege, CEO, nQube Data Science; and Pat Owens, electronic games director and slot manager for Port Angeles, Wash.-based Elwha River Casino.  

Owens noted that Elwha River Casino, with 138 gaming machines, is one of the smallest casinos in the country. “We are called, ‘The Friendliest Casino on the Peninsula,’ we got called that by our guests so many times that we adopted that as our casino motto,” he said. The casino is on the very tip of the peninsula of Washington state; “about as far northwest as you can go in the U.S. and not get your feet wet.” It’s located on the Elwha River Tribal Reservation which is just outside of Port Angeles, and it caters to “dedicated and faithful guests.” The casino is located in the middle of the reservation at the end of a dead-end road, so it has a limited clientele. 

“We wanted to do this project because it gave (nQube) a chance to test this technology on a small system so they could see how it would work when they built it up bigger,” said Owens. “We have invested heavily in creating the best slot floor experience, but I think any slot manager would say that about their floor. What we needed to find out was what is it about the machines that contribute to people playing? We have a tendency to focus on the game and, as far as the cabinet goes, is it a slant or an upright, and those are the only things we need to pay attention to. nQube segmented it down to how many buttons there are and how many do they prefer; do they like to have a whole bunch of choices or do they just basically want a bet button? And which games draw those players? How will changes to the mix of games impact the floor?”



Owen’s issues at Elwha are familiar to all slot managers, but why use AI to address them? “You have treasure troves of extraordinarily granular data right down to transactions on the slot floor,” said Fiege. “AI takes huge, messy data sets and makes sense of them. We all have an internal model of the world around us and we use it to make decisions; that’s how we survive. Giving computers the ability to do that is really the next frontier in AI. AI can help tremendously in understanding what the data actually means. 

“We’re not trying to replace slot managers and operators; we’re trying to enhance intuition and push things in the right direction.” 

Baran of nQube explained the underlying technology as follows: “Generally speaking, any algorithm that has some intelligence, that does things that humans are normally good at, that is really able to learn, is what we call AI. Deep machine learning is a subset and some machine algorithms can learn from data. When software relies heavily on humans to interpret the results, that’s probably not AI.”

With more and more vendors developing AI solutions for gaming, Baran encouraged operators to look for answers to the following questions: Is the AI component of this technology actually benefitting my business? Is it just taking advantage of marketing buzz, or is there something about this product that’s new and innovative? Does it solve a problem better, faster, cheaper and more reliably than traditional techniques?

The focus of the Elwha case study was slot floor optimization—basically, which games should you buy, which should you retire and where should you put your games on the floor. Machine segmentation is central to any such exercise, said Fiege. “We do segmentations both on the player side—basically behavioral segmentation based on what players have done on the slot floor—and we also segment machines, which is an example of strategy optimization.”

The goal of segmentation is to take a bunch of items and look for similarity between those items. “With games, you’re trying to maximize the similarity within groups and minimize dissimilarity,” said Fiege.  “Slot machine segmentation is a really important part of what we do because we’re trying to put together machines that have similar performance characteristics. What we get from casinos is denominations, manufacturers, cabinet types, machine ID numbers, video/reel, etc. We can key that into our slot segmentation system and start breaking the machines intelligently into groups. Every machine follows a different path down the street. We can do similar things for player demographics as well. The AI system I use is something I built in 2002 and is really focused on these types of problems.” 

The slot floor is considered as an ecosystem. “We have a floor with people moving around playing machines,” said Fiege. “People can’t play the same machine at the same time and if I add a new product to the floor it’s going to take attention away from existing products; if I move something I’m going to impact the floor in other ways that are difficult to account for. We think of the slot floor as a kind of interconnected web where we’ve got players interacting with slot machines and interacting with players in a less direct sense. One of the key things we do is model the slot floor as a whole—I’m not looking at individual machines and saying, ‘This one did well so we’re going to add more of that and this one didn’t do well so I’m going to get rid of it.’ We’re building an entire model of the slot floor, fitting it to the data and allowing AI to actually learn how the dynamics of a slot floor actually work. 

“When you‘re doing this you’re taking into account cannibalization between difference slot categories, product placement on the floor… we’re looking at active transactions from the slot floor and session-level data and using that knowledge to drive incremental revenue,” Fiege added. “What we’re also always trying to do is get really good segmentation data but not increase the number of segments any more than we have to.”

The slot mix is a major focus for Owens at Elwha. The casino opened in March 2009 and he came on board that December. “When I first started there we had 106 games and two vendors,” he said “Since that time, we’ve expanded to 138—which is as many as we can possibly fit into that space and still make it comfortable for our guests. I made changes to the floor; I didn’t have the variety that I wanted with two vendors. Now we’re up to six vendors, which is about maximum for a casino our size. I wouldn’t go any higher because we can get those six vendors to get us the games that we want. I try to change games every month; I’m always looking for the best games as all slot managers are.



According to Owens, Elwha River Casino installed the nCube solution in part to find answers to the following questions:

  • What slot machine attributes affect performance?
  • What factors affect machine performance relative to the customer base?
  • How would changes to the mix of machines impact the performance of other slot machines on the floor due to market saturation and cannibalization effects?
  • What products should Elwha invest in and which products should be retired?
  • What uplift in overall win will be achieved with the investment?
  • Owens said the AI exercise produced the following results for Elwha:
  • AI is able to explore tradeoffs in an intelligent way and flag harmful low performers for removal; and
  • AI was also able to make recommendations for new purchases, by properly taking into consideration supply and demand trade-offs. 

Elwha was provided with two alternate sets of purchase/retirement recommendations:

  • 8.8 percent of the floor, AI found a potential 3.8 percent uplift in theoretical win; and
  • 18.3 percent of the floor, the expected uplift was 5.2 percent. (Newer versions of the code achieve higher uplifts.)

Owens said nQube information allowed Elwha staff to ask more detailed questions of suppliers, including:

  • What can they offer that my players will play—what applies elsewhere might not apply at Elwha;
  • AI provided recommendations from within slot machine segments (based on slot machine characteristics), and Elwha asked what vendors could offer that match those segments; and
  • Intuition vs. logic—comparing AI recommendations with changes that have been made to the floor.

nQube also provided Elwha with time savings by aiding in the following tasks:

  • Analytics and purchasing/retirement recommendations;
  • Report generation; and
  • Marketing tools based on slot machines and player segmentation. 

In his presentation, Owens stressed the time savings aspect of AI in particular. “I spend about two hours a day working with the previous day’s numbers; what that tells me and what I can do with the floor,” he said. “What games you bring that have a cannibalization effect. We all know that when you bring in new games and you have a limited clientele, you don’t bring in new customers for those games—you’re just spreading the customers you have all over. So it’s important to keep an eye on where those guests are going, what we should invest in and which ones we should be getting rid of. AI gives us a good basis to predict just exactly what will happen if I bring in these games. The purchasing and replacement recommendations are very good. Since it doesn’t have all the vendor catalogs of incoming games it doesn’t recommend specific games, it recommends by segment. For instance, people at your facility prefer simple bonus rounds, fewer buttons, slants to uprights… that sort of thing.” 

Games can be popular not because of a particular theme or genre but because of the physical aspects of the game itself, the comfort level of people sitting in a chair and playing the game. “AI is able to explore these tradeoffs in an intelligent way and I would say in a much faster way than a human being can,” said Owens. “So rather than spending a couple of hours on the gaming data from the day before, I can run a report and find within minutes what I would be digging to find otherwise.”

The obvious choice is to just remove the lowest performing games on the floor, said Owens. “But if you look at those statistics more in-depth, then you understand that just removing those games is not the simplest solution. There are all these other factors involved. Sometimes, poorly performing slot machines actually are a benefit because they tend to bring in people who see it as their favorite game. They’ll come in, play it then they’ll look around at the new stuff. So it’s good to keep a couple of those games around once in a while just to bring people in. We have several games that no other places in Washington have because they’re obsolete and yet they’re favorites of some people who come in and spend some time on, then they get up and explore the floor.”

Such idiosyncrasies can be found on every slot floor, which is among the reasons why humans have nothing to fear from AI.  

“AI will never replace a slot manager, or anybody,” said Owens. “That’s not what it’s meant to do. It’s just a tool. One of the things that AI does not have is human intuition. For instance, one of the games that was recommended to be removed was the only game that our very top player will play. If we get rid of that game, that player will not come into the casino anymore and we’ve lost our best customer. So we obviously can’t take that game off the floor. 

“What this tool gives me is something to look at; it will give me a game that is as close to that in cabinet style and other attributes that I can use to gradually try and wean this player off of that one machine so we can get this obsolete game off the floor.”

AI is also able to make recommendations on which games, types and segments of machines to bring in to increase revenue. “It provides information within gaming machine segments that give me ammunition to go to my vendors and say, ‘I’m not looking for the latest and greatest game, I’m looking for a machine that does something specific,’” said Owens. “Low volatility, high volatility...  we tend to look at that more than we do bonus rounds, for example. We have found that people are getting really bored with bonus rounds. Everybody gets up and goes to the restroom when they get the 100 free spins because they know there’s nothing they can do until that’s over with, so they’ve got the time. They like the rewards of that, they don’t like having to sit through it. Then there’s more and more happening combining the kinds of bonus rounds so the player can pick the one they want, but sometimes players don’t want that many choices. It’s those sort of things that we need to focus on that you can’t tell just by wandering the floor.”

In the future, Owens sees AI as a tool that every slot manager should have in their arsenal. “I think of how it has benefitted me with 138 games and I can imagine what it would do for someone with 500, 1,000 or 2,000 games on the floor,” he said. “This is going to save them a lot of time.”