Overview
Nike's digital platforms decide what each member sees, and those decisions run on models that have to stay accurate and affordable to retrain. Our team worked embedded with Nike's data science function, building and maintaining the machine learning systems behind personalisation and engagement.
The challenge
The models needed to raise member engagement and lifetime value while running on pipelines that had grown expensive and slow to rebuild. Redundant data sources and costly retraining cycles were limiting how quickly improvements could reach members.
The solution
We built a recommender system on Databricks, Snowflake and PySpark to raise user lifetime value, alongside a feedforward neural network classifier to drive engagement. We then refactored the ML pipelines behind them, removing redundant data sources and adding automated unit testing to protect reliability and data integrity.
Outcome
The classifier doubled user engagement across Nike platforms, while the pipeline refactor reduced cloud cost and shortened model build times. Findings were presented to cross-functional stakeholders, keeping modelling decisions clear outside the data team.