First Look: Blue Yonder

Blue Yonder is based in Germany (Karlsruhe) and was founded in 2008 by an ex-CERN researcher, Prof. Michael Feindt and focusing on solving big data problems such as demand estimation. Several solutions targeted at retail were released over subsequent years. In 2013 they launched their Data Science Academy. The Blue Yonder platform for predictive applications launched last year. This is targeted at making it possible to automate very large numbers of decisions and embedding analytics and machine learning into operational systems. Blue Yonder has a team of over 80 data scientists and makes over 100B predictions per month on their platform.

The Blue Yonder platform is for developing Predictive Applications. These applications are very focused on automation and Blue Yonder aims to solve problems with ongoing and continuous data integration, the predictions themselves and operating at scale.

The platform performs three specific activities:

• Collecting data
• Predicting outcomes
• Driving decisions

The platform has flexible storage, REST APIs, pluggable machine learning pipelines and both HTML 5 and API-based runtime components running on a multi-tenant, secure cloud infrastructure. The machine learning core uses both proprietary and open source machine learning algorithms. These can be accessed directly, useful when a new problem is being addressed, or packaged up to focus on the specific problems of trend estimation, classification and event prediction. For some problems they have pre-packaged models. Everything done on the platform is done in python and takes advantage of the wide range of python tools and algorithms available.

For integration, the platform combines data services for access to public data, a flexible storage platform, job control for machine learning and data integration jobs and APIs to work with external systems both on-premise and batch. A web application can be built on the platform to configure and manage the applications developed. The data model, prediction model and application development are all points of customization.

The decision-making element of the platform considers costs, predictions, probability distributions etc. and drives the result back into systems such as ERP and CRM systems. Blue Yonder often performs A/B testing comparing the results of these automated analytics with human decision-makers. They find, as everyone does, that the algorithms generally outperform humans. In particular cognitive biases are problematic with, for instance, loss aversion resulting in poor decision-making.

In addition to the platform, Blue Yonder has several pre-packaged products and focus areas:

• Forward Demand for daily and weekly demand forecasting in retail (pre-packaged product)
• Dynamic Pricing to find ideal price through rigorous analytic experimentation (about to become their second packaged product)
• Replenishment
• Risk Management
• Return Management
• Customer Analysis
• IoT Predictive Maintenance –with a special focus on being able to sell machine availability

More information on Blue Yonder is available here and they are a vendor in our Decision Management Systems Platform Technology Report.