Behind the code: Fantasy Football insights in practice

Author Chris Jason is the Senior Director, ESPN Fantasy Products

ESPN and IBM have teamed up to bring a new level of insight to fantasy football team owners that correlates millions of news articles with traditional football statistics. Watson is built on an enterprise grade machine learning pipeline to read, understand, and comprehend millions of documents and multimedia sources about fantasy football. The ESPN Fantasy Football with Watson system has been a significant undertaking with many components.

This article is the eighth in an eight-part series that takes you behind each component to show you how we used Watson to build a fair, world-class AI solution.

AI fantasy football in practice

Throughout the 2018 National Football League (NFL) season, millions of fans played ESPN Fantasy Football in a head-to-head format. Each week, fantasy football team owners had the option of using Watson to set team lineups. This was presented to fans in a few different forums — most notably player screens in the ESPN Fantasy App, but also in segments aired on TV and in digital content. Empirical-based decisions that are supported by Watson help to minimize the temptation of starting a player just because he is on your favorite team or is one of your favorite players. In September alone, over 5.5 billion insights were produced for the 9.8 million users that accessed the ESPN Fantasy App, and Watson complemented with evidence the 2.4 billion minutes fans spent in the app that month. This unprecedented level of depth and insight from unstructured data complimented by ESPN’s traditional player statistics and analysis provided a comprehensive and detailed story about each player. Watson used some of that content as evidence to explain deep insights.

Toward the end of the 2018 ESPN Fantasy Football season, over a thousand players participated in a survey to measure the utility and impact of the Watson AI system. From the active survey respondents, over 80% of the users who used the feature said the Watson AI insights helped them to enjoy fantasy football better. The more a fan followed the NFL, the more likely they used Watson.

Some notable feedback included a need for a consistency metric, clear definition of boom/bust, simplified point potential, boom cumulative occurrences, and additional transparency around the models. Other users wanted the ability to compare more players that are on or not on a specific roster. These are all items we will be looking at as we continuously improve the experience.

Throughout the 2018 ESPN Fantasy Football season, a series of former NFL players, ESPN personalities, IBM employees, and a movie star competed to win the coveted IBM Watson Fantasy Football League trophy. Teaching them about Watson was not easy, but accomplished. After the draft, each team manager used Watson to select their starting lineup.

Starting lineup

Over the course of the season, the AI insights became invaluable. Field Yates used the insights to compare running back score distributions with success. Bonnie Bernstein removed her favorite player bias by using Watson to pick a quarterback. Stephania Bell made a compromise between her head and heart by examining media buzz about players. Charles Woodson used the boom and bust metrics to decide whom to pick up from the waiver wire. John Urschel selected his wide receiver starters based on projection curves. At the end of the season, Jerry Ferrara said that the top three things he learned throughout the season were:

  1. Don’t ignore the boom potential because you are afraid to bust
  2. Let Watson work for you on the waiver wire
  3. Running backs are the key for a successful season

Congratulations to Stephania Bell for taking home the trophy, but remember, Watson is already training for the next season.

Image of Stephania Bell

In Stephania Bell’s championship synopsis, she said that her top three Watson features were:

  1. Ability to quantify risk versus reward using boom and bust projections
  2. Player comparison tool to visualize players
  3. Fun of being her own team manager augmented by machine intelligence

Personally, I have found it gratifying to run into fantasy football players that are using my AI system. During a New Year’s Eve event for 2019, I met a soon-to-be friend from Missouri who was telling me about a new tool he uses for selecting his team’s lineup. The tool was Watson! He relied on the boom/bust metrics and enjoyed the distribution curves for comparing players. He said that he could easily find differences in players by looking at their score distribution shapes.

In another example, an IBM product manager, Raphael Sacks, who had never played fantasy football before, used Fantasy Insights with Watson to go 9-4 and win his league’s regular season before losing in the league semi-final and settling for third place.

Every week, he used Watson’s Boom/Bust predictions to choose starters and to pick up free agents. In particular, Raphael found the Boom/Bust feature handy for choosing Flex starters, as there were often several wide receivers and running backs with very similar ESPN point projections. To be thorough, he often added the top few free agents to the boom/bust visualization and picked up a free agent kicker late in the season who proceeded to have several very productive weeks. When the Boom/Bust feature was very close, he used the Buzz analysis. This feature was especially useful in deciding whom to start at the quarterback position because he had two top 15 quarterbacks on the roster. While using Buzz, he chose the higher-performing quarterback in 13 out of 16 weeks.

This blog has shown how ESPN Fantasy Football with Watson was used by millions of users throughout the 2018 Fantasy Football season. Watson converts difficult decision into evidence-based choices. Get ready for the next ESPN Fantasy Football with Watson! #WinWithWatson

The ESPN Fantasy Football logo is a trademark of ESPN, Inc. Used with permission of ESPN, Inc.