Rooter is a sports community platform in India that provides personalized sports and mobile gaming content. Through this platform, they provide a voice to more than 300 million fans of sports and gaming. They have an innovative user generated live audio and video content platform that engages users and their personalized sports feed and score cards are available in eight different languages.
The name of the game for a business like Rooter is user engagement. One of the ways they planned to increase that metric was by incorporating predictions on individual players, teams, and other game statistics. This would allow their users to assemble and follow their fantasy cricket teams, specifically in the Indian Premier League (IPL).
In order to remain the biggest sports community platform in India, Rooter knew that they would need a clever way to continue to outperform their competitors. They decided to up the ante and make sure their cricket prediction and sports analysis was something that would keep their users returning to the platform more frequently as the IPL season progressed.
Knowing that this was a tall order, they started looking for an AI software to help them achieve the status of being the platform with the best predictions for their users. As they had a shortage of experience and infrastructure around AI and machine learning, there was a significant barrier to Rooter for the development of accurate prediction models prior to the launch of the IPL season.
After researching several options, they began a conversation with Stradigi AI and began to make quick use of the Kepler platform.
After a quick onboarding period lasting just a couple of weeks, the team at Rooter had access to an AI platform that allowed them to create, test, and deploy predictive models. They could quickly and easily integrate and productionize these within their existing system without the need for extensive AI or data science experience.
Guided by the Customer Success and Solutions Architects at Stradigi AI, Rooter began by taking publicly available data from the previous seven years of IPL cricket tournaments. From there, they performed factor analysis, analyzing multiple player and team statistics to generate Batsmen and Bowler ratings for each player.
Rooter made use of two different workflows available in Kepler: Clustering and Global Feature Impact. With these in play, they were able to segment batsmen and bowlers to determine key success factors. The information from this output helped them to optimize the AI models.
In all, Rooter built six prediction models focused on key metrics relevant to assessing team and individual performance in a cricket match. The output included Team, Player, In-Game, Match predictions, all of which gave them models that were able to much more accurately predict the outcome of a team and individual’s performance during a specific match.
The prediction models they produced were transferred to the Rooter website and app prior to each game and updated mid-game to provide an in-game prediction feature. All of this was done using Kepler’s API to connect their prediction models to their custom CMS.
Rooter has been incredibly pleased with the performance of the Kepler platform and the ability to accelerate the delivery of AI into production. Especially given their lack of AI and machine learning experience at the beginning of the project.
The Kepler platform reduced the time that it took Rooter to retrain and redeploy prediction models by 93% over previous AI projects. They deployed six performance prediction models within their first six weeks using Kepler.
Over the course of the season, Rooter averaged better prediction accuracy across multiple prediction types by a factor of nearly 10% as compared to their other competitors who also make use of prediction models for their users.
Most importantly, Rooter improved their user engagement statistics—both pregame and in-game—by nearly 20% in the 2020 IPL season when compared to the previous season.
Not only was Rooter successful in incorporating their first AI-powered feature into their platform, now that they have seen these great results, they are also adding more AI projects into their product-development and innovation roadmap.
DATA SCIENTISTs
WEEKS TO FIRST PRODUCTION MODEL
MODELS DEPLOYED IN LESS THAN TWO MONTHS
