Electrolux Project

Electrolux, a global leader in household appliances, originated in 1918 with its headquarters located in Stockholm, Sweden. Initially specializing in electric vacuum cleaners, the company has expanded its portfolio to encompass various major consumer appliances. Alongside its primary brand name, Electrolux, the company's prominent brands include Frigidaire and AEG.

Headquarters

Stockholm, Sweden

Founded

1918

Industry

Household appliances

Company size

45,000+

Challenge

The substantial sales volume of home appliances inevitably corresponds to service calls—instances where customers encounter issues with their purchased appliances. During the warranty period, typically lasting one year, Electrolux dispatches technicians to resolve these issues directly in customers' homes. The primary objective of the project was to forecast the volume of service calls for these appliances based on historical data.

Results

A regression model using Markov Random Fields has been developped, using one year of training data in order to predict the number of service calls for the next month. In the graph the Service Call Ratio is used which is a derived metric from the number of service calls itself, where a renormalisation is taken into account by the number of sold items. The actual value of the SCR is shown by the black dots whereas the prediction is shown by the blue dots, a blue band is added to reflect on the uncertainty coming from the sales volume estimate. The red and green bars indicate the relative error per bin. Here the result is shown on all the sold product in the food preservation category. With a mean sales volume of 176735 items per month. The Mean Absolute Percentage Error on the SCR is 4.2%

Summary

A sunburst interactive graph shows the overal result on all the product hierarchies; product line, group and subgroups. Low mapes correspond with product hierarchies for which the prediction is very good indicated by the pink color, whereas the less good predictions are shown in dark green. The overal prediction quality is very good and always below 14% MAPE.