Google Cloud: NCAA March Madness

To showcase the power of Google Cloud and its ability to quickly turn data into insights, we used AI and machine learning to make live in-game predictive spots during the March Madness tournament.

Introducing the world’s first real-time predictive broadcast spots using AI and machine learning… aired during the halftime of NCAA March Madness.

Months of hard work and inventing a whole new breed of broadcast TV commercials using AI + machine learning + a dynamic video technology.

The results?
🦁 2018 Cannes Lion, Gold (Creative Data)
🦁 2018 Cannes Lion, Silver (Digital Craft)
👉 43 million viewers
👉 +91% product lift

  • Role Head of Creative, Cloneless
  • For Google Cloud | Eleven 🔥 2018 CANNES LION WINNER 🔥
  • Date March 2018
  • Type Broadcast Television

AI and machine learning create the world’s first real-time predictive broadcast spots.

To showcase the power of Google Cloud and its ability to quickly turn data into insights, we used AI and machine learning to make live in-game predictive spots during the March Madness tournament.

Our team of data scientists analyzed decades of NCAA data plus the first half of a game to make a prediction about what would happen in the second half. That prediction became the basis for a hyper-contextual, real-time commercial.

To pull it off, we created a tool kit of dynamic creative assets that could create a spot about whatever Google Cloud predicted. The tool kit was loaded into custom-built, Google Cloud-powered software that could assemble, edit, render and traffic a television spot directly to the network in less than 10 minutes.

In the end, it would be a live end-to-end product demo, in front of 43 million viewers.

Execution/Planning
The process began by partnering with the NCAA and moving decades of their historical game data to Google Cloud. We spent months architecting the data pipeline and creating a set of possible narratives based on the 2018 tournament and matchups. From there we developed systems that could choose the best narrative given a game’s trajectory and build it into the creative.

Leveraging decades of college basketball data, we created models to analyze the data. These models fed our proprietary software that was built on the Cloneless platform and also used Google Cloud technology. This technology was combined with a team of data engineers who monitored live feeds of the games.

We built a workflow for predictive themes like free throws, turnovers, and bench strength. For each theme, we used regression and classification modeling techniques. We trained these models on each team individually, as well as their combined totals, creating 21 prediction options per game.

The live data was integrated into the final spots using dynamic creative elements that could be tailored to the prediction. They included a live scoreboard to set real-time context, supers that could change depending on the predictions, and animations that could include the colors and logos of the teams that were playing. The output was six hyper-relevant, real-time ads. Each ad featured a prediction and was aired during the high-profile March Madness Final Four and Championship games.

In the end, we successfully demonstrated the power of Google Cloud to turn data into insights in real time. And we did it live in front of 43 million viewers, entertaining March Madness fans and our heavily indexed target audience among them—giving Google Cloud instant relevance with businesses everywhere.