The ancient Chinese board game, Go has little to do with professional American football but they have one thing in common: predictive analytics.
While football features 60 minutes of brute force and theatrics across a field, Go is a subdued dance of black and white stones around a square board. But if a computer model were to play each game, it would ask the same question: How do I maximize win probability?
To Frank Frigo, co-founder of EdjSports and its parent company EdjAnalytics, the football-Go connection is obvious: Both are games of strategy and skill—and both can be modeled by a computer.
Frigo, a former backgammon world champion, along with co-founder and astrophysicist Chuck Bower, studied computer programs that defeated human chess and Go masters1 to build predictive data models for the sports world.
“The one thing that you’ll find in all of those games is that when you model them, and the machines surpass the humans, they’re focused solely on win probability.”
Frigo says there’s an important difference between a predictive analytics strategy that maximizes win probability instead of points scored.
At the end of the 2017 season, one EdjSports football client won a national title employing Frigo’s brand of data-driven play style.2
It was a notable achievement for data in a sport that’s beginning to embrace deep predictive analytics3 to make better decisions, from weekly opponent scouting reports to draft-day personnel decisions.
But if coaches have warmed to cutting their instincts with big-data predictive analytics, it may take longer for fans’ perceptions to adjust to the seemingly risky decisions that the data calls for.
Customizing Predictive Analytics to Meet the Team’s Needs
The predictive analytics model that EdjSports uses for its professional football clients draws from 20 years of game data to simulate hundreds of thousands of possible scenarios, games, and seasons—all customized for every team that uses its technology. That customized analysis informs decisions throughout the year, from game-time lineup and play calling to offseason trade and draft choices.
The data, acquired from professional teams for a vendor fee, includes …
- play-by-play information about players on the field
- clock count
- ball position
- win-loss records
- past-decision outcomes
This historical data establishes the scenarios in which a coach might want to test assumptions that form the basis of the predictive analytics. To do so, the coach pits a virtual team against its virtual opponent, both of which are built from in-season data—player stats, win-loss records, play outcomes, and so on—that updates weekly.
Applying Predictive Analytics to Real-World Sports Scenarios
If a coach’s upcoming opponent has a defense that’s famously impervious to rushing, the coaching staff might use the predictive analytics model to simulate common scenarios where instinct would tell them to pass. The model might confirm the staff’s instincts, or suggest other options that better maximize win probability because it cares only about a single metric: Game Winning Chance (GWC).
“Other tools help teams maximize points, but EdjSports sees the best option as the one that maximizes GWC, even if it requires a counterintuitive style of play”, says EdjSports President Tony DeFeo.
Pro football teams aren’t allowed to use computers or predictive analytics during games, so EdjSports is part of the pre-game prep for the coaching staffs and their analytics teams, which vary in size and sophistication. Coaches run the upcoming matchup through thousands of simulations and use a printed-out compilation of the results to inform their play-by-play decision making.
In the post-game report, coaches can analyze everything that happened during the game, and use their findings to shape the next game’s strategy. How much did one particularly explosive play by the opponent, a key turnover or even an injury cost a team? Frigo says all of those factors can be measured through the predictive analytics GWC model.
Predictive Analytics Reveals the Benefit of New Plays
The GWC, a supremely important predictive analytics metric, is the key to understanding some of the perplexing but ultimately rewarding behavior that football fans witnessed in the 2017 season.
“The EdjSports platform,” says Frigo, “often points to more aggressive actions that could be counterintuitive. And I think that creates a more exciting brand of football.”
Aaron Schatz, president of Football Outsiders, a site that popularized football analytics back in 2003, agrees.
“It just so happens,” he says, “that analytics in football say that it works better to be aggressive, and being aggressive is more exciting football.”
Going for it on Fourth Down
One notable example is common wisdom surrounding fourth-down decisions. League analysts have long known4 that, from a win probability perspective, coaches should more frequently take risks on the fourth down. But, as Frigo says, “in the face of the media and scrutiny of fans, it can be hard for a coach to do that.”
Predictive analytics is changing that. The reigning champs used the language of win probability5 to justify decisions throughout last season, and football as a whole seems to be finally catching the statistical fire that has swept through other leagues.
Schatz says that football’s late arrival has less to do with a lack of available tools, and more to do with risk-averse leadership.
“The [analytics] block has always been a fear, I think, on behalf of coaches and general managers, to do certain things that analytics suggested because of a fear that if it didn’t work, they [would] get fired,” Schatz says. “What most distinguishes last year’s championship team was an owner who gave the coaches a green light to use analytics.”
The Shift from Gut to Data Requires Broadcaster Buy-In
The shift won’t be complete, however, until the media gets on board and educates viewers about the on-field changes they’re witnessing.
DeFeo feels that some commentators have been “intellectually lazy” in stamping unconventional decisions as “gutsy” when really they’re rational and calculated. If, instead, they learned the GWC-centric style of play that leverages predictive analytics, they could better place viewers in the coaches’ shoes.
But DeFeo does say that some broadcasters have already expressed interest in using EdjSports to enhance the storytelling aspects of their commentaries. It hasn’t happened yet, but when it does, he imagines that it might have an effect on the football-viewing experience similar to the effect that revealing players’ cards had on televised poker:
“All of a sudden you have information that’s not available to the players themselves, or, in our case, the coaches.” At least not in the moment.
By using a tool that cares only about win probability and involving fans in its adoption process, DeFeo thinks football could finally transition from laggard to leader in the realm of sports analytics. “This tool would immediately put football in the driver’s seat,” he says. “Maybe just behind baseball.”
This content is produced by WIRED Brand Lab in collaboration with Western Digital Corporation.