If you’ve been following our NFL picks, then you’ve probably seen some articles about our “AI picks.” Since the season began, we’ve offered weekly predictions for every game in the NFL. We built an AI system that uses machine-learning algorithms to predict the outcome of each NFL game.
The AI predicts the probability of a team beating its opponent as well as the margin of victory for said team. By predicting the margin of victory for a team, we can make more educated bets based on the point spread. Our AI currently uses no player data. We rely on a custom mix of boxscore statistics that goes back 20 years. We will be incorporating player data into our system(s) soon, when more of our internal processes become automated.
A Hard Look at the Numbers
For the 2019 NFL season, we kept track of every single prediction we made in order to see how we would fair against Vegas. Given it’s our first try, the results are rather promising.
Moneyline Bets
When betting straight up on every NFL game, the AI system picked the correct winner around 63% of the time. Many NFL experts have a win-percentage of 60% to 70%. Our results are good compared to the experts, but they’re not great. We would likely be classified as a bottom-tier expert if we were to let our AI continue as is. And we don’t want to produce bottom-tier results, we want to be among the best. We are confident that, with enough fine-tuning, our NFL AI system can become just as good as all the top-tier experts. Incorporating more meaningful and relevant real-time data will likely be the push that brings us to the level of a top-tier NFL expert.
Bets Against the Spread
In order to profit when betting against the spread (ATS), a bettor would have to win more than 52.4% of his bets. Going into week 8, our AI seems dead set on breaking even. We are hovering around the 52.4% ATS, which was what we expected given the difficulty in creating a system that can beat Vegas bookmakers (they’re good at what they do for a reason).
So far, our best week was week 7, when our AI went 9-4-1 ATS.
The one game we pushed on was Cowboys vs. Eagles where the opening line was Cowboys -3. Our predicted margin of victory was also Cowboys -3.
Our worst week was in week 5, when our AI went 5-9-1 ATS.
Look back to those games and it’s easy to understand why the AI performed poorly. There were multiple upsets and almost one-third of the games were decided by three points or less.
One Thing to Note:
Our bets against the spread are if we were to bet on every single NFL game given our predicted margin of victory. If our predicted margin of victory fell below the line, we bet on the underdog. If it was above the line, we bet on the favorite. We provided no further conditions on how to bet other than where the predicted margin of victory fell in relation to the point spread. Down the road, we will incorporate human intuition to pair up with our AI. The most accurate and meaningful results are produced when both a human and a computer can work together harmoniously.
Future of TheSportsGeek AI
Right now, our game result predictions are only incorporated into the NFL. We will be expanding our NFL AI to include predictions regarding betting totals, home/away team scores, and futures wagers. Likewise, we will be expanding the use of our systems to other major sports like the NBA, MLB, NHL, soccer, etc. in the future.
