During the recent INMA Los Angeles Tech Innovation Study Tour, we visited the Fox Studio Lot (home of The Simpsons!) and had the privilege of hearing from Melody Hildebrandt, chief technology officer at Fox Corp. Their AI use cases fall into three categories:

1. Create: Tell better stories. 

2. ⁠Curate: Meet people where they are, improve content packaging and discovery. 

3. ⁠Carry out: How can we improve execution of repetitive tasks?

Two key insights: Hildebrandt pointed out that for the last category, Fox found it most effective to use a bottom-up process, harnessing the energy and ideas of entrepreneurial employees, while the other two had been more top-down. 

The other distinction was that Fox decided to buy technology for the last category, while it preferred to build “highly custom” applications for the other two.

“Data is the new gold. We don’t want to undervalue it. The first two are highly strategic, so we are building,” she said.

Some of their internal tools include:

• Predictive graphics for live sports. These were created in-house. The production team focuses on optimising storytelling as they are fed real-time insights throughout the game as it airs. For example, if a particular pitcher is facing a particular batter during a baseball game, the tool provides them with an analysis of each player’s statistics and how this particular play could go. “It allows us to tell differentiated stories by putting data in the hands of non-technical users,” Hildebrandt said.

• Semantic search for Tubi’s 250,000 titles. Tubi, owned by Fox, is the most popular free CTV streaming service in the United States, and content discovery is a problem, as any of us who have tried to decide what to watch on a streaming service know. “How can you infer intent from users? LLMs are incredible at this,” Hildebrandt said.

• Clipping video to repackage it into different forms. Content is marked with different categories, using an in-house model. This tool uses computer vision on chyrons to understand the entities and topics that are relevant. “Creating video is really compute intensive,” she said. “Here, we are running AI across existing video.”

• Finding the right video clips. This tool has a semantic understanding of text articles and pulls appropriate clips based on how relevant they are. “The newsroom loves it,” Hildebrandt said. “It is a painful process otherwise, and they know it will lead to more engagement.”

• Streamlining product work. This tool ingests stakeholder interviews and pulls out use cases, pain points and opportunities, and creates a summary of discussion and action items and requirements for developers. 

• Data analysis. This tool streamlines tedious manual file comparisons between ratings services and log events, identifying errors in reporting, a daily task for the ratings team.

If you’d like to subscribe to my bi-weekly newsletter, INMA members can do so here.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .