November 3, 2020
Gazelle – PHD Sales forecasting
Sales forecasting lies in the heart of every company’s operations. Typically, historical data can be used to compute accurate estimates for future sales. Gazelle, however, is in a special position. Their industry changes rapidly, new bikes are introduced each year. Next to that, they have plenty of SKUs and productional challenges. This results in their historic sales data not always being reliable, making it a challenge to generate a forecast.
To tackle this challenge, we created a model that can forecast demand based upon similar products. The model automatically compares all products, models, or SKUs and clusters them to a desired aggregation level. By doing so, we automatically correct for outliers in the historic data. The resulting model generates forecasts with a much higher accuracy compared to currently known methodology.
The algorithm was developed as part of the PhD research of Jesper, the methodology will be published into a forecasting journal in the future.
Jesper SlikSenior AI Specialist