VAIX Co-founder Andreas Hartmann speaks to Gambling Insider about how AI can help shape the future of gaming.
How did your career path lead you to the gaming industry?
I have been running product, technology and operations in internet, online gaming, mobile and fintech companies for over 19 years now. Before co-founding VAIX, I was CPO at PPRO Group (fintech) and Upstream (mobile commerce).
By trade, I specialise in online product. Five years in Silicon Valley as Product Manager before that job description existed (quite old, I know). Via a friend, I met the founders of PartyPoker and moved from San Francisco to Gibraltar; interesting choice location-wise but good from career and learning experience side. From pre-IPO PartyPoker, over UIGEA, to the merger with Bwin, it was a wild ride. From launching the industry’s first core account & CRM system, real-time event detection to merging poker networks #3 and #4, it’s been gaming left, right and centre.
Since Silicon Valley, a lot of the work and research I did in product was always in search, query understanding and information retrieval, important parts of what is now considered Artificial Intelligence, with two of the four patents I have in that area. That’s why AI was so exciting for me and I quit my job as CPO to start VAIX.
What gaps in the market did you see when co-founding VAIX?
The core observation was AI is going to be like mobile, just much more profound – and everybody knows what mobile did and is doing to our ecosystems. The second was gaming had not adopted AI, and especially Deep Learning, in 2016/17. AI has been around for a long time: Amazon has been doing collaborative filtering since the ‘90s. But the technology advances enabled viable commercial Deep Learning really gave the field a boost in 2015.
The third was the belief gaming will only gradually adopt AI. Running product and roadmaps in gaming for a decade, I saw mobile being adopted late, social, big data and the list goes on. Gaming is an industry with much better margins then e-commerce or fintech so has much less pressure to innovate.
Having been on the buy-side, and having developed a deep understanding of the challenges in delivering product internally and with third parties, we saw an opportunity to develop services that remove challenges, launch early and create value for visionary operators.
Why do you think Deep Learning is going to have a more profound impact than mobile?
Mobile is an innovation which made thousands of new use cases possible, via a new user interface. AI is and will profoundly change what sits behind the interface. We already have lots of examples: smart spam filters, your car parking autonomously, real-time translation of languages, or Alexa or Google Assistant at home. Everywhere you look, AI is already affecting us. It will go into so many more areas, including gaming.
What can be done to make AI and Deep Learning more prevalent, especially within gaming?
Very broadly speaking, I think we have two areas of pull. One is top-line: for example, deeply personalised recommendations to make the user’s experience more entertaining and hence get people playing more. We have run a number of projects which all show significant uptake of gaming activity and sustainable long-term revenue with a near zero effect on churn.
The second one, partially born out of regulatory pressure, is compliance and responsible gaming. There’s immense interest in what the industry can do to predict problematic gaming behaviour. For that, you need Deep Learning models. They provide better accuracy, identify game patterns faster and then keep learning thereafter.
Are there any other ways to use AI within gaming?
In relation to the previous points, pick the right approach for introducing AI. Ten years ago, many firms said “let’s do mobile” but didn’t know exactly what to address, and ended up just putting their desktop UI onto the first iPhones, not considering implications such as screen size and bandwidth. Finding a real use case or business problem where the technology works is essential. So don’t just go and try to “make AI”, but try to solve a specific problem. For example: to increase player engagement and improve use of your extensive portfolio of casino games, then build recommendations onto your homepage. Or sports: the normal sportsbook has tens of thousands of markets – just put the ones that really make sense on page one.
A good example is Bet365. It is one of the most successful sports betting operators with a staggering depth in content; this makes finding the event and market a user wants a daunting task, which includes many taps from the mobile homepage. Visiting the desktop site makes for less clicks, but requires skills comparative to those of a Wall Street broker to comprehend and navigate the UI. The solution: just take the five most relevant events and give them to the user. AI helps with this specific challenge – instead of just introducing AI for the sake of it.
What are the major challenges you and other AI firms face in 2019?
Like with any new trend, the biggest challenge for AI providers is getting onto the roadmap of an operator. At any point, there are always dozens of priority projects pushing to get development resources, from revenue drivers to regulatory must-haves. One can only help with this by making the value of the technology very clear and then help minimise the requirement for operator’s resources during delivery.
Also like with any new trend, AI is a buzzword internally and data or other teams, already more than busy with priority projects, want to drive all associated projects. Besides the opportunity cost of such teams not doing the key projects already on the backlog, Deep Learning, while significantly better in performance than “shallow” AI, is a lot more complex and work-intense. Operators trying to do DIY AI solutions internally spend considerably more management time and investment versus doing it externally with a company like us. I remember the days when PartyGaming ran its own mail server. Now every company is using external services like Mailchimp.
The third challenge is obtaining data. While VAIX is able to build a quality production-ready model within one to two weeks, getting the data can take months, depending on the many stakeholders, permissions and processes involved.
Voice-based products have presented a number of early challenges for gaming companies. How can these be overcome?
Voice-based products are one application of AI. Anything Alexa, Google Assistant or Siri does is based on Deep Learning. Early challenges exist in every product. One additional challenge in voice-based is getting to understand the domain.
An example: we are currently working on a product for bingo chat hosts using Deep Learning to analyse and monitor player chat. Arguably one of the biggest retention factors in bingo is its chat and relationship between players and the chat host in a room. Bingo players have their own lingo: from “wtg,” to “2tg,” to “blnt.”Even the fairly smart Siri or Alexa would have serious issues if put into a bingo room. A good Natural Language Processing model for bingo requires knowledge and understanding of the group, as well as a lot of training. But, with a quality algorithm, we can produce an admin tool that adds a lot of value just by knowing a player, what they are talking about and their mood.
So, on the way to the perfect voice-based agent, there are many challenges. But also many steps on the way which add value. While many only talk about the final goal, in gaming, there’s gold all along the pathway as you gradually make things better while working towards that final goal.
Does this involve a process of manual training?
AI in general, and NLP in particular, always requires training. In the case of bingo, ideally a chat host, or someone with a love for the game, is best placed to train a bingo chat model.
The model needs to understand the context. This requires human training. A very good example is the word “Scrabble.” If you see this word, you’d think of nothing particularly good or bad. But it can have a negative meaning in gaming context. People say “I’d rather play Scrabble because I always lose here”, or “we just had a really nice weekend with the family, playing Scrabble.” The two contexts are completely different: one is a negative and the other a positive sentiment.
Does VAIX have any big plans in 2019 for its voice-powered interfaces?
Voice and chat are very good examples of what Deep Learning can provide. We have a few projects and features in the pipeline which really empower the chat host to manage a bingo room. Then there are products for chat host managers to analyse and optimise chat. Operators mostly use promotions and bonuses as a retention tool. In bingo, the industry has largely been ignoring the most important retention mechanism of all: the social chat.
Happy users who are in a positive social environment stay longer in the chat, then buy another card or two, and continue to improve the social chat atmosphere and further increase play.
Where does VAIX aim to be this time next year?
We want to be the platform of choice for Deep Learning personalisation. We firmly believe that, just as you use an email provider, a CRM platform or CMS, you shouldn’t build the personalisation mechanism yourself. VAIX wants to be the industry’s AI provider of choice because there are economies of scale from the data partnership. We want to be best in market for predicting player value, player activity and player churn.
Finally, beyond just chat, we are aiming to make more accurate predictions of users’ problematic gaming behavior – and faster to allow operators to act before the damage is done.
Being at the top of people’s minds for Deep Learning in gaming is the goal for us. Not everywhere, but the specific high-value use cases we are adding value in, and delivering significantly better than others in the market today.