From RFID to real-time AI: How a decade of AWS and NFL Next Gen Stats has rewritten the playbook – News

In the world of professional sports, “data-driven” is often a term tossed around to describe basic box scores. But for the National Football League, the last 10 years have represented a fundamental shift in how the game is measured, analyzed and even played.

This week, as the league reflects on a decade of its Next Gen Stats or NGS platform, the story isn’t just about football — it’s an excellent example of how cloud-native infrastructure and machine learning can transform an industry in real time.

What began in 2015 as a tentative experiment with radio-frequency identification or RFID tags has flourished into an artificial intelligence-led set of experiments and decisions fueled by Amazon Web Services Inc. Today, the partnership between the NFL and AWS serves as a model for the “intelligent enterprise,” processing millions of data points per game to deliver insights that were once considered impossible to quantify.

The origin: From tracking to intelligence

A decade ago, the NFL’s “Next Gen” journey started with hardware. The league embedded RFID chips into every player’s shoulder pads and within the football itself. Twenty ultra-wideband receivers were mounted in every stadium to capture the X/Y coordinates of all 22 players 10 times per second, and the ball 25 times per second.

“Football, for 100-plus years, has been a box score game,” Mike Band, NFL’s senior manager of research and analytics, noted in a recent retrospective. “You had yards, touchdowns and tackles. But those numbers only captured a sliver of what unfolded on the field.”

The early years focused on low-hanging fruit, metrics such as top speed and player separation. However, the raw data was just the substrate. The real breakthrough came in 2017 when the NFL formalized its partnership with AWS, moving the project from a tracking experiment to critical league infrastructure. By 2018, the league opened its tracking data to all 32 teams, putting every franchise on a common analytical footing.

Scaling the stack: The SageMaker era

The complexity of the questions grew: How difficult was that catch? What is the probability of a sack? The NFL needed more than just data storage; it needed advanced machine learning capabilities. The league turned to Amazon SageMaker to build, train and deploy models that could handle the high-velocity data streaming from the field.

In addition to SageMaker, the NFL has adopted many AWS tools, including Amazon Quick, which is an agentic AI-enabled workspace that acts as a set of “teammates” for business users. The NFL is using Quick to deliver real-time, interactive visualizations and answers to different stakeholders, including fans, broadcasters and analysts.

The first major milestone of this partnership was “Completion Probability,” launched in 2018. Built using an XGBoost machine learning model, it factored in 10 variables, including receiver separation and quarterback pressure, to assign a percentage to the likelihood of a catch.

Today, that single model has evolved into a library of more than 75 machine learning models running simultaneously. What’s equally impressive is the sheer scale of the data being generated and analyzed:

• Data ingestion: Every snap triggers the creation of a massive amount of physical data.
• Latency requirements: Models must return results in under 100 milliseconds to be relevant for live broadcasts.
• Volume: The system now produces between 500 and 1,000 unique stats per play.

It’s important to note that though NGS was initially created for broadcasters and fans, the data backbone now underpins everything from officiating and schedule creation to the “Digital Athlete” — an AWS-powered injury prediction tool that helps teams identify when players are at increased risk of injury.

During a media panel in San Francisco this week, Julie Souza (pictured, right), global head of sports for AWS, and Mackenzie Herzog (left), vice president of player health and safety for the NFL, discussed the impact AI has had on injuries. They explained it was the combination of the Digital Athlete and tens of thousands of simulated games that led to the dynamic kickoff rule, banning of the hip drop tackle, and a redesign of helmets, all of which led to the lowest injury rate the NFL has seen in decades.

Decoding the ‘game within the game’

One of the most recent and complex innovations to come out of the AWS-NFL lab is “Coverage Responsibility.” For decades, defensive performance was a statistical “black box.” If a quarterback didn’t throw at a cornerback, that corner’s effectiveness was invisible in the box score.

Using spatio-temporal transformer architectures, the same type of technology behind modern large language models, NGS can now identify defensive assignments in real time. The system can tell if a safety was disguising a coverage pre-snap or if a cornerback was “left on an island” in man coverage. This transforms the eye test of scouts into hard, verifiable data.

The league has also democratized this innovation through the “Big Data Bowl,” an annual competition where data scientists from outside the NFL are invited to solve league problems using tracking data. Many of the metrics seen on Amazon Prime’s Thursday Night Football today, such as “Pressure Probability,” originated as submissions from this open-source community.

The next frontier: Optical tracking and skeletal data

As the league looks toward the next decade, the NFL is already moving beyond the X/Y coordinate. The next evolution of Next Gen Stats involves “optical tracking,” using 4K camera arrays to capture the full 3D pose of a player.

Instead of seeing a player as a single dot on a screen, the system will soon track 30-plus points on a player’s body, including joints such as elbows, knees and hips. This skeletal data will unlock a new dimension of biomechanical analysis, allowing teams to analyze a quarterback’s throwing motion or a lineman’s leverage with millimeter precision.

Lessons for IT leaders from Next Gen Stats

For IT leaders and enterprise architects, the NFL’s decade with AWS offers three key takeaways:

1. Context is king: Raw data is a liability until it is contextualized by machine learning models.
2. Infrastructure dictates innovation: You cannot run real-time AI on legacy, siloed systems. The NFL’s AWS cloud stack is what makes subsecond inferencing possible.
3. The ecosystem approach: By combining internal expertise with external talent, such as with the Big Data Bowl, and vendor partnerships such as AWS Scientists, the NFL accelerated its R&D cycle by years.

A decade ago, Next Gen Stats was a novelty. Today, it has become one of the most critical components of the NFL. As the league moves into an AI-first future, the “Next Gen” label seems almost modest. Business leaders should follow the continuous innovation model the NFL went through to deliver immediate value on the data being generated and then build on the success.

Zeus Kerravala is a principal analyst at ZK Research, a division of Kerravala Consulting. He wrote this article for News. 

Photo: Zeus Kerravala