Last month, Bayes Esports strengthened its portfolio significantly with the launch of Predex, its new predictions data exchange.
Predex, which will be integrated into the BEDEX platform, will use highly accurate probabilities’ for a range of esports events to provide sportsbooks with a range of predictions.
Dr.-Ing Notger Heinz (pictured, right), Principal Data Scientist at Bayes Esports, sheds some light on this new predictions data exchange including some of the challenges associated with developing accurate probability models.
SBC: For those that might not know, can you begin by telling us about Predex? What is it and how does it work?
NH: Predex is an abbreviation that stands for Predictions Data Exchange. With Predex, we are offering probabilities that are calculated by our sophisticated probability models to our customers. Using machine learning methods, we have trained our algorithms on thousands of matches with millions of data points. As a result, our probability models can predict the outcome of a match or side market much more statistically accurately than any human ever could.
There are also no down times to the probabilities provided via Predex. There are no breaks in our feed and markets are never closed by Predex itself. Using these probabilities can give sportsbooks a significant advantage.
Firstly, Predex enables them to calculate more accurate betting odds or improve their own probability models, leading to them having an advantage over their bettors and to higher margins.
Secondly, sportsbooks can also take advantage of the near 100% uptime of Predex to offer more agile betting options that require odds and probabilities to be calculated on short notice.
SBC: How does this fit into your BEDEX platform?
NH: Predex is integrated with BEDEX. That means it’s compatible with all other BEDEX offerings and uses the same pipeline and API. For many of our customers that are already using our Live Trader Dashboard, Predex will already be available, requiring no further integration at all.
SBC: Can you tell us how Predex sets itself apart from other prediction models? What are some of the key benefits that it brings to your sportsbook partners?
NH: Prediction models are only ever as good as the data you train it with. At Bayes Esports, we have exclusive access to the largest portfolio of official live data in the industry. Ultimately, this means that when compared to other prediction models, ours are trained using not just more data, that data is also more accurate and granular.
When we compare the Predex probabilities to others available on the market, we noticed that whenever the two models disagreed with one another, the Predex probabilities ended up being correct 66% of the time. To put it differently, whenever another probability model predicted a different outcome than Predex, our probabilities were correct in two-thirds of the cases. In these crucial cases where games and markets are close and nothing is clear cut, the probabilities we offer via Predex will enable sportsbooks to be on the winners side more often than if they relied on other models.
Furthermore, we have a 99% prediction uptime. This allows sportsbooks to offer more live betting options, such as individual round winners or next objective taken. These options require sportsbooks to be able to react and adjust their odds depending on the game state on very short notice. Slower moving probability models or manual models force betting operators to close their markets much more often or disallow them to even consider offering them to bettors.
SBC: How is Predex removing any biases associated with prediction models?
NH: Many probabilities traded and used in (esports) betting are still created by the traders themselves. These traders can be experts in their field and have tremendous knowledge of the sport or esport they are working with on a daily basis, but at the end of the day they are prone to their own biases and misconceptions. Even if you consider yourself as extremely knowledgeable, yet neutral in the sport you are calculating probabilities in, you still run the risk of overestimating a team’s recent performances or undervaluing a player’s role in a team. It’s impossible for us to be entirely neutral and objective.
Predex only offers probabilities created by algorithms trained with data and data only. They are free of any subjective observations or biases and provide the most neutral perspective possible in esports betting.
While you can argue that the expert knowledge a trader can provide regarding non data based factors do bring crucial value to the calculation of probabilities in multi-facetted sports and games, we observed that, strictly statistically speaking, that knowledge does not provide any noticeable value to the sportsbook in the long run.
There might be some fringe cases in which a trader’s expertise can change calculated probabilities for the better, but over a longer period of time, going with the probabilities calculated by our algorithms will be the safer and better option.
SBC: What are the key challenges associated with developing accurate probability models for traditional sports events? And how does this differ when it comes to the esports space?
NH: The key challenges all start and end with data. In traditional sport, the main challenge is getting access to a large enough dataset that you can use to train your probability model in the first place. The problem there comes not only from getting data from a large enough sample of matches, but that data also needs to have enough data points to cover all kinds of different variables that could influence the outcome of a match.
Since data in traditional sports in large parts still has to be collected manually, there are many variables and potential data points that are not tracked, at least not to the point where they would become an option for machine learning tools. That means in traditional sports, probability models can be somewhat inaccurate because they cannot factor in every single detail that could be relevant.
In esports, the amount and granularity of data is a non issue, at least on paper. Due to its digital nature, data points of every single action and detail do exist and they can be used to train probability models as best as possible.
The challenge comes from how often game titles in esports are updated and changed. Major patches that change the way the games are played on a fundamental level are not uncommon at all in esports. Every single time this happens, previous data becomes less valuable, to the point where it might even be unusable. Having your probability models hold up and continue to be accurate in times where the games themselves are changing requires a great understanding of the data and your model.
As such, maintaining high-level probability models in esports can be very resource intensive for sportsbooks. With Predex, sportsbooks can leave that task to us. All they need to worry about is how they want to use the probabilities, not how they are created.
SBC: Going forward, how will Predex enable more sportsbooks to explore what is possible in esports betting?
NH: Predex gives sportsbooks access to probabilities for hundreds of markets across the three most popular titles in esports, giving them more options of what they want to offer to their bettors. In particular, the near perfect uptime of Predex probabilities enables sportsbooks to offer engaging side and micro markets that can move and change quite rapidly in esports.
Being able to consider these options and know that the probabilities they base them on are as statistically accurate as possible, will give sportsbooks the tools and the confidence to explore the potential of esports betting. There is a lot that is theoretically possible in esports betting due to its digital nature and the granularity of the data available, Predex will help our customers to turn those possibilities into reality.