Anyone who witnesses the highs and lows of sporting drama should be instantly sceptical when they hear of systems that claim to be able to predict outcomes. Even in a world with AI, Big Data and machine learning, there is nothing that can get in sync with the emotions of sport. Do you think, for example, that any supercomputer in the world could have predicted that Simone Biles would have returned from the Tokyo Olympics with just one bronze medal? Sport is just all too human an activity for computers to predict perfectly.
And yet, there is a growing consensus that the future of sports betting will be inextricably linked to computers. Perhaps the most famous example of this is the IBM Supercomputer, WATSON, which has been used for Daily Fantasy Sports (DFS) predictions in NFL. IBM has been partnering with ESPN for a few years on the project, which also includes an app. But to be frank, it’s had mixed results. DFS is huge in the United States, with tens of millions of players. If WATSON was able to deliver a winning formula, everyone would be using the app.
Computers can use structured data
That gets to the point we made earlier. WATSON was able to beat a host of Jeopardy champions all the way back in 2011, an event that was seen as a huge breakthrough for supercomputers. But you can teach a computer general knowledge (structured data); it’s a different story with sports predictions where the use of unstructured data is applied.
For instance, we give the example of Simone Biles earlier. Biles, as most are aware, is one of the greatest gymnasts of the modern era. She was odds-on (as low as 1/10 with some bookmakers) in every discipline she competed in at the Tokyo Olympics. However, she pulled out of several events once the Olympics began, citing stress and mental health issues. Eventually, she did take part in one event, the balance beam, and she duly won bronze.
The example of Biles is important because the structured data would tell you that Biles should win gold on the balance beam. Why? Because she is the reigning world champion. But the unstructured data – in this case, Biles’ frame of mind during the Olympics – would tell you something different. The point is that computers can understand structured data much better than we can – crunching millions of calculations to make a prediction. It’s important for activities like investing in the stock market. But humans understand unstructured data better – at least we do at the moment.
Distractions could cost Kane the Golden Boot
Another example could be Harry Kane. The structured data should tell us that the England captain is going to have a brilliant 2021/22 Premier League season. During the summer, he was the betting market leader to be top scorer. He was also the leading scorer last season. But due to a long summer of transfer speculation regarding a move to Manchester City, it’s clear that Kane is under a unique type of pressure to perform at the same level. The structured data says back him to be top scorer again, but the unstructured data (Kane’s relationship with Tottenham fans) should raise a red flag. Perhaps he will get things together, but there is no algorithm that can explain what Kane is going through this season.
The key to successful sports betting is to marry the structured and unstructured data together. Top bookmakers try to provide that service for customers. For example, William Hill sports betting online provides an insane range of stats and data models on its website for the majority of sporting events. And the unstructured data is dealt with through blogs and podcasts. The latter, William Hill’s The Punt podcast, is well worth a listen for breaking down the stories behind the odds and markets.
But even if we are provided with those tools, there is no guarantee that we use them correctly. Precedence does matter in sport, but it also doesn’t matter. We tend to look for patterns in data sets, and they can be useful. Yet, we also use statistics that don’t really point to anything useful for the sports betting prediction, or that provide a kind of “false positive”. In the simplest terms: Trends can be misleading.
Trends can be deceiving in sports
Another example showing the difficulty with trends: Manchester United have played Leeds United three times since the latter was promoted to the Premier League in the summer of 2020. Across those three matches, 14 goals have been scored, meaning games between United and Leeds average an insanely high 4.65 goals per game. But the scores were 6-2, 0-0, and 5-1. And that means the 0-0 give us little indication of the other two games and vice versa. Which result is the anomaly? The 0-0, or the two high-scoring games? We might be tempted to say the 0-0 is the anomaly, given there are more high-scoring games, but the data set is simply not large enough.
You do see websites now offering sports betting algorithms that can crunch huge volumes of data and arrive at a prediction. But you must accept the caveat that statistics can only do so much. Some of these sites make wild claims about the efficacy of their models, but you should always have some scepticism about them. They indicate what is supposed to happen based on prior information, but that can go out the window in an instant.
So, what does the modern savvy bettor do with the tools at their disposal today? It’s our view that you should not pay for sports betting tips, although some punters do swear by them. But using data – even the stats that are freely available on bookies’ sites – is always a good place to start. Marry that data with common sense and intuition, and you will do just fine. You can also use those sites that offer data-based picks. But bear in mind that the bookmakers use their own algorithms, and they are usually one step ahead of the betting public.