Genius Sports is a UK-based sports data company who is leveraging computer vision and deep learning to lead the digital transformation of sports intelligence, working on an international scale with top sports organizations such as MLB, PGA Tour, and the Premier League.
This May, the NCAA announced its 10-year partnership with Genius, the goal of which is to develop and implement new data capture and analysis services, coaching insights, and real-time statistics that can be used to improve gameplay and increase fan engagement.
We spoke with Gal Ozery, Senior Product Owner at Genius Sports, who oversees product and technology for Genius’s AI team.
“Building a product on bad data means building a bad product.“
Genius was considering outsourcing all of its labeling and tooling to a third party. As a product owner, Gal had major reservations about the quality of the training data that would be produced by an outsourced labeling team using their own tools, into which Gal and the team at Genius would have no insight.
“Having no visibility into our labeling process gave me serious concerns about accuracy. Building a product on bad data means building a bad product. Not controlling your data — the most important part of your product — is a totally unacceptable risk.”
A further concern for Genius was around ability to control their product roadmap. Building and training useful models at the world-class scale that is Genius’s standard is an extremely iterative process. “Without a collaborative tool, you have no ability to quickly iterate on the skeleton of your product. It’s a 3–4 day process to implement any changes, which you pay for in lost progress and additional charges.”
“I have a team, instead of just a tool.”
Gal quickly began the search for another way: “I attempted to use other tools. I was left feeling like I was doing something wrong. The interface was not intuitive. I looked for 2–3 days for certain features, only to discover that they did not exist in the product.”
Other tools were too complicated and difficult to set up: “I checked out other tools. I’m busy; don’t have time to figure out something complicated, and I can’t pull an engineer from another team to do something like that.”
Gal discovered Labelbox and began experimenting. “No one’s interface was as simple as Labelbox. I used the in-app chat function to ask questions about my project. The first time I tried it, Brian answered instantly.”
Once Gal’s team got started using Labelbox, “the Labelbox team reached out proactively, detecting an issue and emailing to let me know they were working on it.”
Working with Labelbox means that “I have a team, instead of just a tool.”
“Having Labelbox means there’s one less thing to worry about, so I can focus on the areas of the product that are in tune with my role.”
“Our machine learning lead loves Labelbox. He’s the end user for this data; we’re labeling the data for his team. He’s who matters.”
Genius up-leveled their relationship with their labeling team: “We would not have been able to engage our remote team without Labelbox. This was the biggest thing we could give them to ensure their, and our, success.”
“The only thing that comes before Labelbox is recording the data.”
Gal’s team enjoyed the experience as well: “The labeling manager said that the team thought using Labelbox was fun. They were proud of the fact that that they were able to finish 1,000 complex images in only 2 days.”
Gal put it simply: “For quickly integrating a labeling tool into your process, other competitors do not match up.”
“I’m looking forward to the opportunity to work with Labelbox on new enterprise features. I want our engagement with Labelbox to continue to grow.”
“Labelbox has become essential to our process. It is the beginning of every single deep learning exercise we do. The only thing that comes before Labelbox is recording the data.”