Ahead of their presentation at this year’s GiGse, co-founders of the artificial intelligence-based algorithms and software specialists nQube, Dr. Jason Fiege and Dr. Stasi Baran, took the time to discuss their burgeoning business, why they’ve used their expertise to target the gambling sector, and the power of creating a positive experience.
Can you outline what has happened to the business in the 12-months since you won the Launchpad competition at GiGse and explain where you are currently?
The GiGse Launchpad competition was pivotal for us as a company. Even just preparing for the competition itself allowed us to think about our company in a different way. As scientists, we were coming from a space that is acutely focused on research and development. We were thankful for the invaluable guidance of Launchpad organizer Melissa Blau from iGaming Capital. She provided great advice and insight to all of the companies who were pitching on the Launchpad, helping us to refine the presentation and message that we were conveying to the audience. The preparation for the Launchpad had us thinking about our company, where we would be in one to five years, and how to think about our slot floor optimization product from a commercial point of view.
Prior to the Launchpad, we were working entirely with mathematical models of slot floors based on artificial data. This approach is normal – with any kind of data modelling problem, one usually starts with theories and artificial data to refine ideas and test methods. But that can only take you so far, and eventually, real data is needed to make progress. Exposure from GiGse allowed us to obtain our first real data sets, which we used to challenge our models. We quickly discovered how subtle this data really is, and it led to a long process of refinement – eight generations of our model in total, and many, many iterations of our computational techniques – until we were convinced that we were getting everything right.
Our thinking has evolved substantially over this time. We began heavily focused on the slot machines themselves, but the model has evolved to become very player-centric – we use historical data to follow carded players from machine to machine around the floor, group them into segments, and calculate the probability that each player segment will play each machine. Un-carded players present additional challenges, but we’ve learned how to deal with their play as well.
At this point, we have created a detailed mathematical theory of how players interact with a casino slot floor, we have demonstrated that it fits real data extremely well, and we have refined our methods to optimize the mix of machines and their physical locations on the casino floor. The underlying theory is subtle and mathematically beautiful, with a lot of features that look surprisingly like physics.
Additionally, the Launchpad exposure really helped us meet the right people. It allowed us to connect with casino properties and collaborators, like iGaming Capital and Casino Science out of Vancouver/Seattle, and we have begun our trial program, in which we test our software on small, medium and large land-based casinos to determine how much of an uplift we can produce for them. In fact, we met the first participant of the trial program on the plane as we travelled home from GiGse!
How important is it for the industry to have events such as the GiGse Launchpad?
The casino industry is shifting towards technology driven solutions, and events like the GiGse Launchpad are a perfect way for the industry to interact with innovators, and to learn about new technologies that they might miss out on otherwise. Likewise, events like the GiGse Launchpad give a voice to small, innovative companies, and a way for these companies to get much needed feedback from the gaming industry.
What topics/issues will you be covering at GiGse 2018 in Miami?
We will review how our business and technology changed for us since the LaunchPad. We will also discuss some of the interesting and unexpected discoveries that working with AI has allowed us to make. One of our favorite stories to tell, is that just days before GiGse last year, we noticed that our AI system was shutting machines off on the floor. We had no idea why this was happening – we didn’t tell it to do that, and we thought that the program was broken. We even briefly considered cancelling our trip. After studying it carefully though, we discovered that the AI was actually doing something very clever – it had decided on its own that it was beneficial to turn some machines off because they were distracting people from busier areas where they would spend more money. It was absolutely correct – in this particular case, the most profitable floor configuration actually contained fewer slot machines.
Through our work over the last year, we’ve gained valuable insights into how players interact with the casino floor. We will share some of these insights. How can we best use data to determine the optimal mix of machines for the slot floor? What spatial attributes of a casino (e.g. table games, restaurants, bars, entrances, etc.) influence performance of nearby machines? How do players cluster on the floor? What is the benefit of having open sightlines on the floor?
The importance of player segmentation falls out naturally from our models, and segmentation has become paramount to what we do. It is really the key to understanding how the slot floor works. This has led us to think about advanced AI-based techniques that allow us to more intelligently segment players for marketing campaigns, and determine the optimal level of rewards that they should be offered. So, we definitely plan to talk quite a bit about segmentation. We will use our AI-guided evolutionary computing platform to demonstrate an effective method to find the best number of segments and optimal segment boundaries, while also optimizing the performance of a marketing campaign.
We believe that there are also potential applications of our technology to iGaming. We will discuss some ideas, and we look forward to feedback from the audience.
Do you see yourselves as Scientists or Entrepreneurs – what’s your self perception?
We’ve spent a lot of time in the lab and we’re now making the transition to being entrepreneurs. Entrepreneurship gives us the opportunity to see our research impact the real world, which is an opportunity that isn’t always afforded in a purely academic setting. This is exciting to us!
As physicists, we’re used to thinking about complex data, mathematics, physical principles, and computational algorithms. We don’t think of physics as a school subject – it’s a way of seeing the world, defined by an unshakable belief that even the most complex systems – including the chaos of a slot floor – can be understood with mathematics. Our models look like statistical physics inside, and even our AI system is built on complex mathematics and statistics.
We like to build powerful computational tools to address tough technical problems, both in science and in industry, and it might surprise you that they are often the same tools.
From our point of view, we are doing real science, and we are also entrepreneurs. The entrepreneurship part comes less naturally to us, but we’re learning, and find it exhilarating.
Why gambling – what made you target this sector?
There were three things that were particularly attractive about the casino industry. 1. The incredibly vast and clean data sets collected by this industry, which are any scientist’s dream; 2. the significant financial opportunities available in this industry; and 3. the slot floor optimization problem that we identified is a perfect fit for the AI-guided optimization technology that nQube had already invented.
The opportunity to have fun at work was also very attractive. We get to do science in an industry that is centered around people having fun. What could be better than that?
What other verticals are you active in and is there a disconnect with gambling in any way?
nQube is also involved in AI-guided optimization of trading strategies, primarily for futures markets. We offer a B2B algorithmic trading service to financial companies engaged in trading. This service utilizes the same underlying AI technology as our work optimizing slot floors, although the mathematical models are quite different.
You use the phrase ‘Happy people spend a lot of money’ could that be a mantra or motto for the gambling industry?
Absolutely – It’s all about creating a positive experience. One way to think about our technology is that we are optimizing slot floors to match player preferences. A well-matched floor improves the gaming experience for players, and happy players spend money. Everyone wins.
It’s a big transition from the Science Lab to the Casino Floor - have you had to negotiate a cultural shift and how have you found it?
There are many cultural differences between academia and the other industries we work in, and the casino industry. The casino industry has been extremely positive and welcoming. When we started in this industry, we were completely open about the fact that we had powerful tools to offer, but lacked domain knowledge, and we’re still learning every day. Along the way, we met a number of key people who became very excited about our technology, and they helped us to make this transition. Notably, we met several of these key people as a direct result of the GiGse Launchpad.
Gambling has a difficult reputation with some observers, is this something you considered and how did you/have you responded?
Gambling can be very polarizing to people. We have occasionally experienced negativity when explaining what we do to people outside of the industry. We respond with the positive side – gambling fuels state economies, it funds tribal communities, and in Canada (where we’re from) it helps to fill government coffers.
We actually debuted our technology at the Gambling and Risk-Taking Conference, run by UNLV in 2016, where we were exposed to a lot of research on problem gambling. We are sensitive to this issue, and once we are more established, we would like to develop techniques to identify and mitigate early-stage problem gambling behaviors, using the same data that we use for slot floor optimization.
When a casino operator engages nQube, what do you supply, how do you supply it and how do you work with them to analyze the results?
Our solution requires large amounts of computational power, so we typically perform the optimization and analysis securely offsite. One of the very interesting aspects of our AI-guided optimizer is that it’s multi-objective. This is a rare capability, which allows the system to see the whole range of possible solutions to their problem, all at once. From a practical point of view, we can show operators the trade-off between new purchases (or number of machine moves) vs. the uplift that they can expect on their bottom-line. It gives operators an effective tool to see the potential impact of their investment of time and/or money.
We also provide software that allows the operator to browse their new floor configuration. They can see what the AI recommended for their property, and they have the ability to make manual changes, such as moving slot machines from one location to another or completely removing them from the floor. If the operator moves a machine, the projected uplift is recalculated for the entire floor, based on that move. Interestingly we’ve never been able to improve the uplift by moving machines around manually. The AI always does a better job.
Our software is great at handling constraints. Only want to change 5% of your floor? We can build that in. Need to have certain cabinet types together, or in a particular region on the floor? No problem.
It might be difficult to convince a casino to remove or move slot machines, how do you anticipating responding to them and how do you convince them to believe in your product?
We saw the AI shut down some machines in a very particular, and somewhat unusual, test case. We don’t expect this recommendation to be the norm. Because we can explore multiple solutions, we’ll be able to show operators what their uplift looks like with or without the slot machine changes.
Based on your field trials, what percentage of uplift from drop can a typical casino expect by utilizing your product?
We’ve been modelling data sets provided to us by several casino operators, but our solution hasn’t been out in the field long enough to measure actual uplift results. However, our model fits the data to within 1.5% mean error without machine segmentation, and well under 1% error when machines are segmented into types, so we are confident that the model is working extremely well. The same model shows that it is relatively easy to achieve an uplift in win in excess of 10%, by swapping out or moving a small fraction of the machines. Even higher uplifts can be obtained by swapping out or moving a greater number of machines.
Putting North American gaming under the microscope where do you see the big challenges and the big opportunities occurring?
We have started in this industry just as it is experiencing a major shift towards innovation – driven by both forward thinking and necessity. Because this shift is current, we encounter operators who are skeptical (the challenges) when it comes to new technology almost as often as those who are excited about it (the opportunities). The industry in general wants and needs innovation, but it can be challenging to break into gaming. The difference in culture between the casino industry and innovators outside of the casino industry, can make it difficult for small companies to gain traction.
In general, the big opportunities are in insights from the “big data” collected by casinos. Data science provides the opportunity to maximize operational efficiencies, maximize casino floor performance, and tune casinos to suit the entertainment preferences of a changing demographic. The answers are already there in the data. It's just a matter of creating the tools needed to find them.