Insights Built with IBM Watson

Share this post:

Following the success of the Leatherhead and Women’s World Cup projects, the Emerging Technology team has extended the functionality into the world of American Football and the NFL with Insights Built with IBM Watson.

Insights Built with IBM Watson utilises data feeds provided by SportsRadar, which, amongst many others, includes a breakdown of in-game events. Over the course of the 17-week season, the 32 teams play 16 games each, taking one week as a bye, with the season culminating in the Super Bowl, usually taking place in February, following a knockout tournament competed by the top teams from the season. As such, a huge amount of data about all of these games is recorded.

Conversational Interfaces

Having previously trained Watson to understand the language of Football/Soccer, the UK Emerging Technology team have built on the architecture of the Leatherhead and Women’s World Cup projects and extended the functionality and Watson’s understanding to the NFL.

Not only does this provide intuitive access to the raw event data from the SportsRadar feed, but also includes complex and cognitive insights which are useful for a more in-depth analysis of the game. In addition to the raw event data, Watson considers a variety of other factors, including the weather conditions for each game, provided through IBM’s Weather Company APIs, to give a more holistic view of all influences on the game. This combined approach enables us to uncover a great number of insights from the data that would be incredibly difficult to infer without the assistance of IBM Watson.

Simple demonstration of Watson’s ability to understand stats and game state. Data taken as of Week 3 in 2019.

Having the ability to generate insights is great, but it would be no good without a way of being able to explore the data to uncover them, and in particular making the user experience detailed, interesting and intuitive. This is where the power of IBM Watson Assistant shines through. The Insights Built with IBM Watson solution contains a trained instance of IBM Watson Assistant that can be queried in a number of different ways to reveal these insights.

IBM Watson is trained on domain-specific language. This enables user’s, in this case the FoxSports analysts and producers, to explore the data, looking at niche in-game scenarios and unlocking the patterns that would otherwise take hours of filtering and searching.

Let’s walk through a few examples of the types of information we can provide.

Situational Awareness

Watson understand the game-state throughout the data, what down is it, how many yards they are from the goal line, how many minutes left in the quarter/half/game, the score, how many yards from achieving a firstdown are they, is it raining? This just scratches the surface of Watson’s understanding, but it enables the analysts to explore the data set using just the questions/queries they’d normally use in natural language.

If we’re looking to analyse a particular team’s decision making at a key point during a game, or for a particular match up, we can ask something like:


IBM Watson understands from the question/query that the best way to portray this data is in the form of a comparative visualisation. The visuals make it very clear that the Steelers much prefer to pass, vs the Cowboys who much prefer to rush. Not only this, but we can also get the context of these numbers in relation to the rest of the league. The Steelers have the 5th highest passing rate in this game state, in contrast, the Cowboys have the 5th highest rate of rushing.

Deep-Diving Analysis

If we want to then deep-dive into the passes thrown by the Steelers for this scenario in more detail, we can ask something like:


Roethlisberger’s targets on 3rd down with less than 4 yards from a firstdown.

Routes ran by Switzer when he was the target of a pass on 3rd down with less than 4 yards from down.

Here, Watson presents a range of visuals that are appropriate to the question at-hand. It shows the targets a particular QB has thrown the ball to, through to the routes that particular receivers have run when targeted by the Quarterback.

In a similar fashion, we could also analyse the rushes that a particular team does. In this example, we consider the “red zone”, when a team is less than 20 yards from goal.


Watson’s response when requested to analyse rushing in the redzone. It’s possible to explore each player’s data individually, see an overview of “all players” and inspect the details of each individual rush in the playlist on the right.


To punt or not to punt?

When a team reaches 4th down, more often than not they will decide to either punt the ball (kick the ball deep into opposition territory) or attempt a field goal (kick the ball from the ground and over the posts, which would result in three points). The decision making process here can be a factor of many things, but fundamentally it will come down to the kicker’s ability, and the weather, particularly the latter when playing in an outdoor stadium.

Watson, using data from IBM’s Weather Company, has the ability to understand and filter on this contextual data.

When punting, you want to go as deep into opponent territory as possible, but not beyond the goalline, otherwise that results in a touchback. The returning team will then get the ball on their own 20 yards line. If you can punt it out of play within the 20, or punt and then tackle the receiving team within the 20, you’ve done well as a kicking team.


This kind of question will help in-game, both offensively and defensively. Given the weather right now, can we go for the punt? Is it worth the risk? Watson gives us this data immediately and could be influential in close games and steer the outcome of games.

If we query the system about the punts by a particular team/player (with or without filters on yfd, yfg etc) the system delivers us detailed information about these punts. Not only can we see the punt distances and start and end positions, but we can see the hang time of each punt and the height it reached.

Punts made by the Dallas Cowboys


This image shows the information returned about the breakdown of the 178 field goal attempts in the 2018 regular season. Highlighted is the attempts by Jake Elliott, with 2 good field goals, 1 wide left and 2 wide right with a maximum kick distance of 56 yds. Not visible on this screenshot, but also returned, is the detail of all attempts by players of every team. Interestingly, we can see from that insight that only 83% of field goal attempts occurred on 4th down, with the rest split between 1st, 2nd and 3rd downs.

Field goal success visualisation


Broadcasts and next steps…..

The insights uncovered from the use of the Insights Built with IBM Watson solution has been broadcast each week on-air by Fox Sports.

Example snapshot of the “Insights built with IBM Watson” segment shown during the Week 1 2019 Fox Sports NFL Broadcast

Having a comprehensive, aggregated collection of data in an easily searchable and navigable way enables the broadcasters to discover a new level of insights. In turn, this gives Fox Sports the means to pass this detail on to the viewers, highlighting key game points, player interactions etc that would otherwise be missed, engaging them further with the NFL. This has been an extremely rewarding  project to be a part of, and we can’t wait to be able to share more with you in the near future!


Note: All screenshots and data included are from the 2018 (previous) season, unless otherwise specified.

Master Inventor & Emerging Technology Specialist

Liz Maple

Senior Inventor & Emerging Technology Specialist

More Innovation stories

Rainbow Octopus – Animate a 3d character with ARKit & IBM Watson

Rainbow Octopus is an open-source project created to show how a developer could use IBM’s Watson services in ARKit and Unity to control and manipulate a 3d animated character. Specifically, via the Watson Unity SDK, we capture and send speech to the IBM Speech To Text, and interpret the results using Tone Analyser and Watson […]

Continue reading

What Makes a Great Wimbledon Champion

IBM has been supporting Wimbledon’s pursuit of greatness for 27 years. In recent years, much of that contribution has come from the Hursley based Emerging Technology team. This year’s challenge? To find out what makes the greatest Wimbledon champion of all time. A team of statisticians, tennis analysts, coaches, journalists and developers assembled to start […]

Continue reading

Kappa Architecture on Bluemix

During the last three years I’ve been responsible for helping the US Masters, Roland Garros and primarily Wimbledon understand how people are engaging with the events on social media. Their requirements have led to a messy combination of stream processing, batch processing and static indexing. I’ve started to look at how a Kappa architecture might […]

Continue reading