For a long time, Koblenz-based bicycle manufacturer Canyon found it a challenge to draw up reliable forecasts for bicycle sales. In the highly competitive bicycle market, this difficulty translates into a considerable financial risk. As part of a four-year project, though, Professor Dr. Arnd Huchzermeier and his doctoral student Christoph Diermann of the Chair for Production Management were able to be of assistance.
A forecast set too high or too low can have serious consequences for the company. In one case, Canyon will find itself stuck with bicycles manufactured at high cost; in the other, company turnover drops – taking customer satisfaction along with it. The main challenges facing the bicycle manufacturer are the long lead time for the manufacture of frames, quick replacements of new models, some of which are sold for just a year, as well as the strong yet hard-to-predict growth of the company itself.
The cooperation between the researchers and the bicycle company began back in 2012. First, a group of experts generated forecasts based on standardized data. The researchers then used these collected forecasts to compute a median value. At the end of the financial year, the magnitude of estimation error was determined by comparing actual sales figures against the computed median value of the forecasts. This process was repeated over the following three financial years to determine a systematic estimation error.
Huchzermeier and Diermann had to apply what is known as a ‘segment-specific correction’ to the systematic estimation error, correcting the value upwards. In the process, they divided the Canyon bike range into categories, such as road bikes and mountain bikes. To avoid overfitting, the breakdown was kept from becoming too fragmented. This way, the economists were able to locate the optimum segmentation level; with segment-specific forecasts, they managed to achieve a deviation of just 2 percent. Now, Canyon is in a position to make forecasts with nearly 100-percent reliability, cutting its financial risk to a minimum as a result.