Esports arena: Digging deep into esports data

The demand for esports betting continues to rise but many online bookmakers have yet to fully understand, let alone leverage, the huge potential the vertical has to offer. This is partly due to the complexities and challenges that come with gathering data and generating odds for esports contests, which can differ greatly to traditional sports.

Traditional sports use manual data collection and sensors, but this is somewhat limiting as you can’t place sensors on players if it risks hampering their performance. This is not an issue with esports contests as they are entirely virtual – in theory, there is a server where all data and actions are logged and stored.

Of course, accessing that server – especially when tournaments are held online or in a location on the other side of the world – is easier said than done. While being given direct server access is the simplest way to obtain data, most suppliers are not granted this permission due to being held offline and/or on private servers. On other competitions, one supplier might buy exclusive access to the server as an official data distributor.

Problems and solutions

This means using computer vision technology that “watches” a video feed and creates performance statistics – it works in a similar way to Tesla’s autonomous car AI engine. Computer vision technology and AI are so powerful and accurate some providers are now using it to generate data and compile odds for traditional sports betting.

Another challenge that esports data and odds providers have had to overcome is the additional layer of complexity that contents have and that change with each game – equipment. While footballers play with the same boots and ball, esports players can customise their character with different items or weapons that enhance their abilities or give them a strategic advantage.  This creates a meta game and a layer of theory crafting for each contest.

For example, an item might be very relevant at the start of a game but not 30 minutes in because the damage doesn’t scale over time. Or an item is good as a counter to your rival’s character, and you end up in a rock, paper, scissors scenario. This creates another facet to the game, data collection and how odds are calculated.

This is being overcome with more complex models and a data intensive approach to prediction. The more complex the rules of the game also usually mean that the game is more data rich. The good news for betting operators is that the way data is processed once collected is similar, if not identical, for traditional sports and esports.

Best practices

But with so much data available, providers must ensure they use it intelligently and effectively. More data means that you need to select the right data points, as not all data points have the same predictive power. The traditional modelling approach is based on classical statistics models which require a lot of fine tuning and a manual exploration of what data points are interesting. Now with the recent AI revolution, models based on deep learning are figuring out what data points and combinations will have the most impact on predictions by themselves. So either you use more complex models or you’re leveraging more knowledge of your data scientists, but in the end the complex model will most likely win.

Of course, the wealth of data available means that data providers and operators can offer players great choice when it comes to esports odds and markets. More data means more betting opportunities and more exotic markets. The size and length of esports contests also offers unrivalled betting activity.

For example, the LCS and LEC finals were full best-of-fives and the recent CS:GO ESL Pro League final between Gambit Esports and Heroic lasted more than six hours. These are huge windows of betting opportunity that also allow providers to gather tons of real-time data and in particular outlier scenarios.

Esports data and odds is still in its early, high growth, high potential stage of development but already we are seeing a consolidation of market prices and maturity. If you look at the bookmakers offering esports odds, the key challenge right now is that markets are often limited and often have short open times – markets are suspended a lot. This is because of data availability, model accuracy, trader knowledge and market pricing.

But with providers now offering better feeds, comprehensive modelling and specialist traders, these hurdles can be cleared allowing operators to tap into the huge opportunity esports betting presents.

Flavien Guillocheau is the CEO of Paris-based esports data supplier PandaScore.