MEET DARRYL WOODFORD

In the Sports betting landscape, there is a man who is combining years of experience in academia, game design and social media data analytics to build comprehensive models across dozens of sports.

Meet Darryl Woodford, a seasoned data architect. Here he tells you how he combines his unique set of skills in the hunt for an edge in sports betting.

Darryl is the co-founder and Chief Technical Officer with Cipher Sports Technology Group and has been building predictive models for nearly 20 years. His journey began in the UK, with studies in engineering and game design, including a masters thesis with the University of Copenhagan on the agency of avatars in virtual environments. From there he was a research fellow at Queensland University of Tech where his focus was finding ways to measure and evaluate audience engagement using social media.

But sport was always calling… “I played a lot of Poker in the UK during the poker boom,” Woodford said. “But that got harder so I moved into sports betting where I was a pre-game and in-play punter for a number of years. But that got harder so I started building models instead,” he added.

Through the use of ‘Machine Learning’ Woodford and Cipher Sports created Fantasy Insider. “We founded Fantasy Insider, projecting how individual players would perform through things like Draft Stars etc. From there we moved off that into sports betting,” he said.

“We basically took what was a rating for how each player would perform and extrapolated that into a team model, running simulations. If you swap out someone like Lance Franklin and put in a different player, how much impact would that have on Sydney for example,” Woodford explained.

Darryl Woodford is always learning, and the depth of modelling continues to develop. “When we started off we did the player’s predictions and tried to extrapolate those.
We’ve got more sophisticated with sometimes 150-200 features in those models,” he said.

“We’ve also expanded into more simulation models. For cricket and tennis we do ball-by-ball and point-by-point, which makes sense given the complexities. We currently do 20-25 thousand events a year, and are branching into fighting sports at the moment.”

Woodford speaks about the growing challenge of finding edges in sports markets.

“Betting markets are becoming more efficient. You are seeing advancements in AI and in data analytics. So edges are harder to find, but you can still find advantages,” he said.

“The US have legalised sports betting which creates new efficiencies. People start betting on sports they don’t know about, or on their home state out of partisanship and that can skew the odds,” he explained.

“You can look at models on sports that don’t get huge attention, like the NBL for example. If playing bigger markets like NFL or NBA I’d look at smaller markets like points, rebounds, assists, first quarter betting for example where edges might pop up.”

As Woodford points out, not all sports are the same. The key is identifying what the most relevant data is to a result in that particular sport.

“It varies by sport. Team stats in some sports can tell you most of what you need to know. In some sports, individual data combined to run simulations can be better. Once you get the prediction out, you either trust your model or you don’t. If you start second guessing when to bet, you will get into difficulties. Do you trust your process? That is the underlying question,” he said.

“In the feature engineering process that’s when the human judgement comes in. What are the key features you want to focus on? And have you got the right ones? If the answer is yes you should trust your model”

Australia is proving a more difficult source for data gathering, with companies like champion data monopolising some of the key areas. Woodford explains how to source the relevant information.

“You can buy historical odds data, but you can also scrape a lot of it. If it’s a hobby project you can look at some websites that give you good free data. NBA.Com or basketball reference for example,” Woodford said.

“For prop betting they can be quite effective. AFL Tables isn’t too bad either for information on a basic level.”

Every aspect of the game comes with risks, and Woodford explains the necessity to be able to trust your data source.

“Where are they getting their information and are they reliable?” he asks.

“In the early 2000s, The models weren’t that sophisticated so you could find edges. Someone was using a model for Wimbledon and CBS had the rights but were showing matches on a 2-3 hour delay and they lined up the website scores with what they were showing on tv, not what was live. As the match went on, the odds started making no sense. It shows you have to know where your data is coming from and what they are doing,” he said.

“A few other points to be cautious of are is the edge too big? If it is you may have missed something in your model. Also has the market overacted to a player going out? Which creates a bigger edge,” Woodford explains.