Five years after German retail giant Otto provided investment capital for a start-up analytics software supplier, it achieved a 40% improvement in demand forecasting from the resulting technology.
Otto derives revenue of €11.6bn (£9.3bn) revenues from operations in 20 countries. Around 10 years ago, a tipping point came where growth in e-commerce made it more difficult for the retailer to forecast sales across diverse product and sales channels.
Accuracy needed to be improved to avoid over-stocking items or failing to meet customer demand in time, says Michael Sinn, Otto vice-president, category management support.
The company had used regression analysis systems common to the business intelligence (BI) market, but was looking to achieve greater accuracy in forecasting sales.
The decision to select Blue Yonder’s technology followed a hunt for a new predictive analytics platform, which saw the leading software firms in the market compete for the contract, says Sinn.
Read more on predictive analytics in retail Predictive analytics early adopters focus on individual customer analysis Demand-driven supply chain innovations center on analytics, mobile ID SAS CEO offers analytics over BI and reveals use cases for text analytics
“Blue Yonder’s results were better than any of the other software houses could deliver. We gave them historical data, from which we already knew the results, let them make forecasts and compared the forecasts to the results. There was a huge gap to the next best software supplier.”
Scientist Michael Feindt developed the algorithm behind the Blue Yonder technology while working at Cern, the European Organisation for Nuclear Research.
Feindt went on to found the company based on the technology he had developed, which uses a neural network that changes its structure as it learns the complex relationships between input data and resulting outcomes.
With funding from Otto, the transfer of four Otto IT staff and three years’ joint work in adapting the software to a business environment, Otto and Blue Yonder introduced the software to live business data in 2010. It was able to repay the investment in the first year.
Sinn says the technology could save €10m each year. As well as avoiding under- or over-stocking, the system also helps reduce the cost of transporting secondary orders and means clothing and goods suppliers can confidently produce larger volumes, which help Otto get better prices without hurting supplier margins.
Data management and governance matters Read more about predictive analytics and retail from the research archive Best practices for analytics in retail Predictive analytics for sales and marketing The business value of predictive analytics Drive your business with predictive analytics
The project has revealed some challenges in data management and governance, says Thomas Friese, project manager, prognosis and forecasting, at Otto.
“Our data is used in many different cases for many different issues and nobody knows the complete process of what happens to the data in our systems. When Blue Yonder started to get our data they got a very special view of its quality,” says Friese.
Discrepancies in data arise because systems have been designed – and data collected – for different categories or purposes.
“When selling clothes you have different data about the articles compared with selling TVs. Our systems grew up with our business. When we built systems for selling TVs we did not think about having clean, useful, normalised data across all items. It is like an owning an old house and continually adding extensions,” Friese says.
In this way, Blue Yonder, can act as a watchdog on data quality at Otto, he says. However, the retailer cannot change its data structures just for the purposes of forecasting. Different business units rely on different data structures as part of the operations systems. Any serious problems with the data will be considered as part of a review of the company’s ERP deployment, Friese says.
Blue Yonder itself is able to cope with inconsistencies in the data. Its self-learning software automatically detects changing conditions and adapts to changing data quality, the supplier says.
Data architecture
Otto runs a Teradata data warehouse and extracts data using its own technology in combination with software from Exasol.
The retailer’s IT department has built various client software applications which use data from Blue Yonder’s NeuroBayes Predictive Analytics Suite, supplied via a data interface, to support procurement and supply chain staff in making decisions.
As well as relying on historic sales data from Otto, the Blue Yonder system exploits data regarding item style, fashion, brand, price, when it was released online, which page of a website it resides on and how prominently it features. External factors such as weather conditions and cultural events also play a role, says Otto’s Sinn.
Otto retains a 50% stake in Blue Yonder, although the supplier works with other firms in the retail sector, including customer loyalty firm Dunnhumby, a subsidiary of UK retailer Tesco. It is working to predict the next ten purchases a customer will make on his or her next shopping trip, in order to help Dunnhumby clients – which include Proctor & Gamble, Shell and Coca-Cola – target offers using smart phones.
The next goal for Otto is to use Blue Yonder technology to forecast the impact of a dynamic pricing model on customer demand.
