OTTO improves demand forecasting accuracy up to 30% with Google Cloud's TiDE model

Overview

OTTO, a leading German ecommerce retailer, used Google Cloud's Time-series Dense Encoder (TiDE) model trained on Vertex AI and deployed on GKE to analyze multivariate time-series data such as seasonal demand patterns, pricing changes, and promotions. TiDE improved demand forecasting accuracy by up to 30%, helping OTTO reduce inventory costs, minimize stock-outs and overstocks, and support dynamic pricing, with particularly strong responsiveness for volatile products like iPhones versus more stable items like men's jeans.

The challenge

OTTO needed to improve the accuracy of its demand forecasting to improve inventory management and customer satisfaction, and found working with complex multivariate time-series data challenging using its baseline approach.

The solution

OTTO's forecasting team, Team Lumen, implemented Google Cloud's Vertex AI, BigQuery, and Google Kubernetes Engine (GKE), integrating the Time-series Dense Encoder (TiDE) model trained on Vertex AI and deployed on GKE to analyze multivariate time-series data such as seasonal demand patterns, pricing changes and promotions.

Reported business value

The adoption of Google's solutions resulted in an up to 30% improvement in forecasting accuracy, reduced inventory costs, and increased customer satisfaction through timely product availability.