FloSports
Retail technology company improved retail promotion and recommendation engine with MLOps
Artificial Intelligence & MLOps
Data Modernization & Analytics
Related Case Studies
![Perform[cb] Logo](https://cdn.builder.io/api/v1/image/assets%2F78dfb6aee9fd4aa4be447bb5ccea13e3%2F3bd5083b2bdd403b940a5ac56631df5f?width=640&quality=90&format=webp)
Perform[cb]
Better Performance
Growth in Offer Claims
Effie Baram
Sr. Director of Platform Engineering
Upside is a North America-based retail technology company that helps people earn cash back on the things they need, and businesses earn proven profit on every transaction. Upside does this by partnering with over 50,000 restaurant, grocery and gas station locations nationwide. Upside powers these experiences through its app and API Partner Platform, which includes Uber driver, Lyft driver, GasBuddy, Instacart and Current Bank apps. All of this value goes directly back to its retailers, the consumers they serve, and towards important sustainability initiatives.
Upside was experiencing an unprecedented demand for its product in 2022 as the platform added more food-related merchant partners. This demand caused difficulties keeping up with issuing real-time and individualized consumer offers. The Upside team knew they needed to implement a Machine Learning Operations (MLOps) solution for platform scalability and more efficient and predictable ML training. Upside needed support developing and deploying infrastructure for the expansion and engaged Caylent for AWS tooling recommendations and to implement the new ML training engine, models and systems.
Caylent performed an initial assessment to understand the customers business goals, technology roadmap and operational requirements and worked backwards to identify core AWS ML services to build upon. Caylent then designed, migrated and tested Upside’s data landscape across multiple Upside environments to train and implement a solution leveraging managed AWS services to reduce the ongoing operational burden for Upside.
The new data landscape solution included core infrastructure, data pipelines, data analysis and engineering tools, data scientist development environments, model registries, feature stores, an ML metadata repository, ML pipelines (history tracking, model version tracking), model serving infrastructure, and model monitoring. By leveraging a purpose-built Amazon DynamoDB-powered ML feature store, Upside can serve inferences to their mobile app customers with sub-100 millisecond round trips.
In addition, Caylent automated supporting infrastructure for the data landscape by implementing monitoring, logging, and alerting for data flows as well as validating data. Caylent also implemented automated CI/CD for data and models that would automatically provision necessary guardrails. Caylent completed the implementation with zero outages and zero downtimes to the previous system.
The new MLOps platform has improved the performance and recommendation engine for Upside’s existing cash back promotions and has resulted in additional functionality and net new offerings around new food verticals. Upside is now able to perform real-time machine learning releases, which have improved both the creation, performance and conversion of consumer cash back promotions.
Continued training of Upside’s implemented ML system has led to a significant increase in the number of cash back promotions being generated in real time to all eligible users across the platform. The performance of Upside’s key recommendation engine has also been improved by 50%.
Following a successful media advertising campaign, the platform has given Upside the ability to accommodate scale, with the robust and resilient architecture now able to support the onboarding of 270% more users and 280% more offer claims this year. Upside has also been able to implement its recommendation engine in two new industry verticals—grocery and restaurants—and announced new venture funding to further growth in new and existing verticals.
Company
Share