AI has become a convenient tagline for attracting attention. But in the world of sports betting, it is an essential component, driving not only fan engagement, but betting odds, user interfaces and providing insane amounts of data for teams, scouts and punters to know exactly what their preferred players could do at any given time. Sportradar’s Behshad Behzadi breaks down how and why AI should be implemented, and what it can bring to the table.
Imagine knowing what every player will do before they even do it. Imagine being able to train a team, calculate odds, even know when a player is likely to pass the ball – the possibilities are endless. And the sports technology group Sportradar has that capacity in their hands. Chief Product, Technology, and AI Officer Behshad Behzadi tells AGB exactly how important this data (when processed properly) is, while examining the advantages and limits of AI.
“I had been at Google for 18 years before joining Sportradar,” notes Behzadi.
And his contributions to the company were not few: Co-founder of Google Assistant, Co-founder of Google Lens, Co-founder of Google Smart Display. The list goes on and on.
But the executive saw in Sportradar something exciting. A chance to step away from the managerial toil that inevitably comes with success in a massive company, and to focus on his true love of data – and all that can be done with it.
It’s no coincidence that Behzadi holds the title of not only Chief Product, Technology Officer, but also that of AI Officer. The inclusion of the AI component espouses Sportradar’s approach to its services, leveraging AI to process information that fuels not only sports betting, but coaching, talent scouting, optimization of user interfaces, fan engagement, advertising, back-end support and streaming, among others.
AI is essential, but customers need reassurance
When asked whether Sportradar could operate at its current scale without the use of AI, Behzadi unequivocally answered “no”. But that doesn’t mean that it is rampantly deployed across all areas of the business.
Computer Vision, deployed in early 2023, allowed the company to collect “100 times more data than a human”, meaning that without it “you would not be at the same level at all when it comes to predictive analytics and creation of the odds and prices”. This has further evolved and integrated into a myriad of services it provides. “The gold mine is always data. And Sportradar has, I think, more than anyone else in the world, sports-related data”.
Moving to integrity, “we need to prevent fraud and cheating, because otherwise this industry will not survive. So that is also machine learning and AI”.
But these are the more “classical, traditional AI examples”, notes the expert. Generative AI is becoming a key component of future operations, not just for Sportradar.
One such example is Content Studio, an AI-powered platform using real-time data to generate sports articles, videos and audio content in seconds.
“It takes one game and then creates hundreds of different stories, formats, target to different audiences, in different languages. If you don’t have this scale, and scale matters […] you couldn’t have that,” notes Behzadi.
But looking more at the betting side, the generation of odds can be of concern to punters wishing to know exactly how such odds were calculated.
Enter the world of Bet Concierge – a chatbot designed to provide answers to match-related questions and smart betting suggestions directly linked to the betmakers’ bookslips.
“If you can’t explain why this odd is this number […] your people will leave because they think you are cheating them”.
“AI governance that we have at Sportradar is trying to help us on where to apply and where to not apply,” notes Behzadi, highlighting a key part of the concern around AI use.
For example, allowing AI programs themselves to define bonuses and bonus frequency, discounts and reductions is a slippery slope – and could potentially hurt their sportsbook customers if rampantly applied. “We would not know their business logic or policy for that,” highlights the executive.
AI models and practicality
“We do both,” notes Behzadi, in regards to purchasing AI models or developing them in-house.
Sportradar has some fully home-grown AI models, but “we use models from Amazon, we use open models from Google and others,” noted the executive. “We kind of expand it with our data […] and then try to improve it better for our use cases”.
“We also have built something quite relevant,” notes Behzadi, “a foundational model for basketball, where we […] took all the data that we have from NBA videos – all the tracking data that we have which is practical, for every single player.
“We have 29 body points and 60 times per second […] Now we have a model which can simulate every single player, very similar to the players. For example, this player in such situation is passing, is looking this way but giving their pass that way”.
But the model isn’t static, it’s self-learning. “We didn’t give any indication to the model, but the model learned”. By providing the “foundational model”, the system can self-teach.
This means that you can “build 10 times, 10,000 times the simulation of the game. And you can compute lots of different statistics. What is the expected possession value in this very specific case, including the direction of the head and hands, the angle of the body, and what is good in that situation.”
“You can use it absolutely for betting,” notes the executive, but it goes far beyond – especially relevant for coaching and “how to improve a player”.
This is particularly relevant in the Synergy Coaching & Scouting platform– focused on basketball’s coaching staff and professional scouts – which can take that data and – instead of needing to attend far-away games, or study endless game reels, evaluate the data points and use it to coach or procure the best players out there. Truly a game changer, as all teams in the NBA, WNBA and NCAA utilize the platform.
But these models cost money and punters only want a certain amount of options to bet on.
“People, at the end of the day, still bet on a few number of more popular things. When it’s pre-game or during the game, there are certain bets which (are main stead)”. Offering thousands of betting options simply because the technology can do so doesn’t mean it will increase punter uptake.
UI and AI/reading the mood
“VIP players are the people that every bookmaker cares about, and you don’t want to touch it and potentially risk that your VIP decides to go somewhere else. But the cost of not changing might mean losing a new generation of users.”
Behshad Bezhadi
To get around this, the User Interface (UI) can be tailor-made, particularly now that you can use AI to help build a solid user interface that is bespoke to each punter.
“You have to build an experience,” indicates the executive.
This covers multiple areas: “you should look into what’s the journey of the fan using the product,” notes the expert, “you make it engaging but make it easy, make it pleasant”.
One highly interesting component is understanding how and when to advertise.
“The idea is to try and improve advertisement for your brand by knowing how the emotions of people are changing in different moments […] When a local team is winning and the city is happy, that day is a good day to spend a bit more money on advertising […] Targeting the right time, right person, right message – and that message can be different for you or me despite it being about the same game – you’ll be optimized for the fan experience, for the success of that advertising campaign”.
AI native
“We have tried to become an AI native, become an AI-first company. And the biggest change is really about that […] We are one of those smaller number of companies to whom AI was important enough that we called someone a Chief AI Officer. And that is how we want to internally operate […] The fact that for the past 20 years something has been sold somehow else is not a good reason to continue to do it that way,” solidifies Behzadi.
This approach has guided Sportradar’s growth and rampant success in its partnerships with top sports brands worldwide – utilizing a bespoke approach to AI that acts as a facilitator rather than a catchphrase, a valuable tool when selectively applied within distinct parameters, with readily visible results.
