Jordan Peltier, Chief Data Officer at PandaScore, says that humans and machines must work together if businesses are to maximise the potential of AI when it comes to esports betting odds.

Artificial intelligence has been one of the most discussed industry topics over the past 12 months and it’s easy to see why. AI can be transformative for operators, suppliers and bettors, and while there are concerns around adversarial AI, when used in the right way, it’s a game-changer.

At PandaScore, AI plays a crucial role in the odds we deliver esports betting operators. This is because it is incredibly effective at handling complex, high-dimensional and unstructured data and finding patterns within it – these patterns are needed to predict the outcome of events.

This, of course, is the foundation needed for determining the odds for each market that we offer. But the power of AI goes beyond this.

It can adapt dynamically to new data which is just not possible with old statistical models like parametric distributions as they struggle to capture the non-linear relationship between variables so require additional manual engineering which takes time and resource.

But does this mean the AI machine is taking over from human trading teams? Absolutely not.

Leveraging the capabilities of AI and combining it with specialist human traders is the optimal approach because you have models that can capture the complexity of esports supported by human traders with a deep understanding of each discipline.

Odds should always be based on true probabilities and while there have been significant breakthroughs in AI – I’m talking about transformer-based models such as Large Language Models – it is still prone to making errors, especially when it comes to forecasting.

Let me explain by way of percentages. AI alone can get you 90% of the way to determining the true probability of an outcome but you need the human specialist trader to get you the final 10% of the way to perfection. And perfection is a must when it comes to odds.

Ultimately, this is what allows operators to offer high-value odds to their customers while ensuring a strong and stable margin from their esportsbook. What’s more, it only takes one small error for sharp, savvy bettors to exploit a mistake and that’s not what operators expect from their data partners.

It must be remembered that AI models are only as good as the data they feed from and still require training. Human traders help in this regard, providing a layer of security for core actions like settlement and odds checking when the model goes off market.

In some instances, the trader will have access to data the model doesn’t see of wasn’t trained on – things like a last-minute change to the player roster.

This is why our human traders are responsible for picking the tournament and market coverage that will most appeal to our operator partners, and our traders are on hand 24/7/365 to support operators in their efforts to offer a top esports betting experience to their players.

The training of AI models is crucial and again sees our trading and data science teams work together during the entire lifecycle of each AI model we build from designing the model to testing and interacting with models in production to ensure they are performing.

This is a virtuous cycle with our data scientists training the most optimal models that prove to be the easiest for our traders to manipulate. Traders also give a ton of deep and detailed feedback which our data scientists can use to improve the models.

The idea here is that the trading team has little input once the model is rolled out and often only make one input per match with the AI model then calculating all of the odds and markets. This ultimate is what powers our BetBuilder and PropBet products.

Having specialist traders for each esports discipline is also important. Each video game is like its own sport and while there are some similarities between some titles, the differences have an impact on the markets and odds offered and that’s exactly why specialist traders are a must.

Take shooter-round based game like CS2 and Valorant and MOBA games like Dota2 and LoL. With the former, there’s a lot of repetition in the gameplay which takes place round after round. This means betting markets are focus on rounds and kills for teams and players.

With MOBA games, there are objectives within the game that have a direct influence on who wins the match – things like kills, towers, nashors, inhibitors and dragons. This is complex – and doesn’t exist in traditional sports betting – and is why pricing teams must master in-game dynamics.

Again, this is something that AI can’t do on its own right now.

AI is a key pillar of our business, and we are incredibly proud of how we have used it in collaboration with our data scientists and traders to create models that deliver top odds for our operator partners and their players.

But working with AI is all about staying ahead of the game and our work is never done. This is why we continue to test, iterate, innovate and do all we can top perfect our models. This would not be possible without our incredible team which stands as a testament as to how machines and humans can work together in harmony.