22 November, 2022

Eye of the machine

OSAI CEO Roman Garin speaks to Gambling Insider about starting his data technology company, how he hopes it changes the gambling market and achieving 100% reliability.

Could you give us a brief overview of your product and how it works?

In 2017, I had an idea to replace the human scout with a computer. I have over 14 years of experience in the sports betting industry, so I know how the scouting works; all of these human factors that could somehow relay the data that betting companies buy. When I started, we built the technology over three years and by 2020 our site had become a company for physical sports. Then Covid-19 happened and we decided to start with table tennis.

It isn’t a big sport and we were a small company, but because we needed to build the team and technology further it made it tricky to start with a big sport. When we started, we said the main thing for us was that it works in real-time; because we’re talking about the betting industry, the real-time and the quality of the data, accuracy of the data; it is all really important to get right because these are major issues. Of course, they compared this with the human scout and when we’re finished, we should be better than the human scout.

Was the AI Data Scout your first vertical?

Yes, that was our first vertical. The scout takes each frame, analysing the game – not just the raw data, player or ball position – and predicts the score before the referee. When we do that, understand it takes a lot of data, much more than the human scout has. Using table tennis as an example, the human scout will see where the ball is, who is serving and who scores the point; for our system, we use where it bounces on the table, speed, player hand position etc. All that helps us transform everything into data that is useful for the betting industry.

The first one is the sport analytics to help the player understand how they could be better than other players, to see what other players do in their games; it could help them become better. Otherwise, we have also created our Mixed Reality system that takes physical sports and mixes it with a virtual world, because younger audiences can be lost and we want to attract them more. The young audience buy into esports more, so we mix reality for them. With the virtual background created in real-time, rendering in 18 milliseconds with a high-framerate camera, and a computer to cut the delay between the camera and the upload to the server.

If this works well for real-time betting in table tennis and tennis, how does this adapt for other sports?

We’re a software and technology company; for us the data is abstract, so the software could be transferred to any other sport easily. When we started, we only thought about table tennis, but as this tech has gotten better, we have started thinking about how we could implement this in other sports.

But because we are still a small company, we can’t just go into the big sports today; we are working as a business and we need to be stable where we are, then to invest in where we want to go. That is our approach, sport by sport. 

How reliable is the tech for people that want to adopt it, especially for those who want to place bets on the statistics?

In total, 97.5% is the minimum percentage of accuracy we have. This 97.5% doesn’t mean every match we do has a 97.5% reliability rate; we actually have a 100% rate if you look at our system when it is in use. We get our results by comparing them with the referee, we compare our prediction and when there is a difference, we make a correction to provide the right data – so when the people get the data from us, they have 100% accurate data. When we started, we realised that referees make mistakes, but because we don’t provide the referee with information, whatever they decide is right – our system has to go along with. But all the deep data we provide still opens the new markets for betting because humans can’t provide this kind of data.

When you look at the data for lots of other sports, it is all based on humans, who of course can’t do everything, but we could. We could help betting sites be different with them because they could open new betting markets based on our data. In table tennis, you could split the table into 16 squares, you could make bets on where the bounces will be, for example. It’s also about speed and accuracy. Every day we work hard to improve our system and all the products we have in production have a high accuracy rating. In Rocket League, it's about 99%.