How AI is changing football forever

Welcome to this week’s Deep-fried Dive with Fry Guy! In these long-form articles, Fry Guy conducts an in-depth analysis of a cutting-edge AI development. Today, our dive is about Gridiron AI. We hope you enjoy!

Football season is officially here, and for many that means drafting fantasy teams, wagering on games and players, or engaging in heated discussions with other avid sports fans! What if there was a way to boost your chances of becoming smarter than your friends, winning your fantasy football league, and beating the house? Gridiron AI is here to offer an AI-powered solution that might help you do just that.

WHAT IS GRIDIRON AI?

Gridiron AI is a user-friendly platform which utilizes AI and machine learning to project weekly player stats and predict game outcomes for the National Football League (NFL). This includes predicting outcomes such as money-line bets (Which team is going to win the game?), spread bets (Will a team win/lose by more or less than a sportsbook-set number of points?), player stats (How well will a specific player do in a given game?), and game scores (What will be the final score of a given game?).

Gridiron AI does not add any human bias or analysis to their outcomes. Whatever the algorithm spits out, that is what they post. This allows for an unbiased opinion, purely based on data and nothing more. This process eliminates human opinion and the emotional factor which can significantly influence fantasy football and sports betting decisions.

HOW DID GRIDIRON AI BEGIN?

Gridiron AI began with a LinkedIn post back in 2015. One of the co-founders was tired of hearing about peoples’ un-backed opinions about sporting events, and he posted about wanting to work on a project which could create unbiased, purely data-driven projections. A data scientist (and avid Philadelphia Eagles fan) named Andy Troiano saw this post and had a desire to contribute. He said, “We are just a few guys who love sports and analytics, and we are really just doing what we love to do.”

One of the strengths of the Gridiron AI team is that they are small. If they want to implement changes or improvements, they can do so. If they want to take their time and really focus on developing the foundations of the technology, they are under no corporate deadlines. This allows for a fun and passionate work environment.

WHAT IS PROVIDED ON THE PLATFORM?

If someone is into sports betting, they can pay $5/week during the football season to get access to all of Gridiron AI’s picks for the week, based on the betting odds and lines set by national sportsbooks.

If someone is into fantasy football or wants to partake in activities which track individual players, the platform freely provides their rankings of the top 12 player projections for each position, each week. Then, for $5/week, users can access all of the algorithm’s projections, for every player.

Fantasy football owners also have the ability to save their team information on a Gridiron AI account. This way, they don’t have to search their players each week. Rather, the platform will allow the user to easily access data for their players and see who is projected to perform the best for each week. It is also worth noting that the default scoring method used on the platform is similar to that of ESPN or CBS scoring, but it can also be customized for points per reception (PPR) leagues or other outlying rules which might influence scores and rankings of players.

HOW ARE THE PROJECTIONS BEING MADE?

Gridiron AI uses a few different models that ultimately feed their final projections which are displayed on the site, primarily based on probability distributions. Troiano says, “The way that we try to leverage AI is to build and model probability distributions … so we basically have AI give us 500-1,000 possible outcomes for a given game or given player based on all of the available data, and then we use that to roll it up into what is seen on the site.”

For individual players particularly, the algorithm weighs potential upsides and downsides based on the available data and generates projections based on a range of outcomes (upside projection, downside projection) which can allow for a more accurate assessment of a player’s potential each week. It allows the user to see who has great potential for a breakout game and who are the safer plays for that week.

The AI algorithm uses over 500 features/stats (i.e., rushing yards, receiving yards, share of team receiving yards, receptions, receptions/game over 10 yards, etc.) from multiple years of past performance layered with information about the teams they are on, such as the offensive line, coach, and team success probabilities. When all this data is compiled together, the deep learning algorithm is able to make quite accurate predictions.

Gridiron AI’s projections are extremely accurate. On average, they are within 4-5 fantasy points each week, much more precise than NFL.com, FantasyPros, and Fantasy Football Today. Their accuracy for spread picks has been around 52-53% and moneyline picks have been around 60-65% (varying slightly from year-to-year). Gridiron AI’s success rates can also be found each week on Gridiron AI’s Twitter/X page.

WHAT ARE THE LIMITATIONS OF GRIDIRON AI?

Gridiron AI has a lot of great things going for it, but a data-driven model like this also has some shortcomings that are worth taking a look at.

One of the limitations of using an AI algorithm to make projections is the injury and suspension factor. It is difficult for the AI to grasp the concept of potential injuries or suspensions and how they figure into projections. Although the Gridiron AI team is working on a way to get better at this, it will continue to remain a problem for all prediction models, because injuries are quite unpredictable.

Another difficulty of these projections, and projections more generally, is the sheer craziness of football. There will always be unexpected upsets, blowouts, and players who over-perform or under-perform. For this reason, this type of analytic-based projection model might not always offer the best projections in each individual case, but over a wide sample size, it will have great success.

Lastly, it’s worth noting that some people don’t enjoy analytic thinking when it comes to sports. As Troiano stated, “The thing about fantasy football is that it’s really hard to remove your personal opinion and feelings about players.” His point is well-taken, as many people have bad experiences with players on their fantasy football teams in the past or tend to like players from teams they are fans of and dislike players of teams they dislike, even if these players or teams could contribute to their best outcome based on the data. It is the individual preferences of the fans, whether data-driven or emotionally driven, which makes sports projections and discussions so fun.

WHAT DOES THE FUTURE HOLD FOR GRIDIRON AI?

Gridiron AI has an amazing opportunity to grow from it’s current 9,000+ users as it continues to gain traction and popularity with sports fans. In the offseason each year, the team is also working incrementally to improve the platform and add additional services. Some of these future features might include:

• Expanding to other sports

• Offering projection explanation summaries using a large language model (LLM)

• Providing the ability for users to talk with an LLM about their fantasy team specifically

• Offering player prop probabilities (Ex. There’s a 30% chance this player will have over 95 yards)

• Preseason total win/loss predictions for teams and season predictions/futures markets

Gridiron AI has endless possibilities for the future, and the team continues to be focused on improving functionality and projection algorithms to refine their foundational model. This will allow them to build incrementally on top of these projections to continually present a service that can be used by sports fans worldwide. In a world where AI is making its way into almost everything, including football, Troiano aptly remarked, “It’s a crazy time to be alive … It’s a great time to be alive.”