Logo_w_o_bg.png
Image by Isaac Smith

Forecasting Sales

Our preseason forecasting solution is state-of-the-art and built using the latest tools in time series econometrics as developed by our partners at the Vrije Universiteit. The estimated reduction in costs for the client due to our improved predictions was approximately 6 percent in 2019 and 2020. Overall the solution allows for better optimized production and smoothed operations.  

Want to read more about how we did this and what we can do for you? Scroll down for more details or contact us!

The case

The production cycle in fashion could take up to one year! How do you decide how many items of which product to produce? A key player in the textile and footwear industry is confronted with this decision repeatedly for every retail season.  

 

Some articles are carryovers: we produce them year after year, whereas others could be new and experimental. What is the best way to decide how many items to order and manufacture? We need to make a forecast. A good forecast takes into account historical demand together with other factors: marketing budgets, fashion trends, seasonality. We developed the demand forecast and a module that suggests how many items to purchase. Sometimes it is better to order more items than we expecting to sell!

Solution

In cooperation with the client, we developed an econometric model that takes a large amount of historical data on transactions, marketing, stock and more as input and returns the demand per article per season. Not all articles have information on historical transactions. Therefore, we develop another model for new articles separately. The new articles forecasts are based on similar articles that were sold in the past. 

 

A good buying decision uses a demand forecast, but it requires a bit more. Sometimes business should order more items than what the forecast suggests, not to risk running out of stock during the season. We developed a system that suggests the buying quantity based on margins and forecasting uncertainty.

 

The solution, developed in Python, stores the data in the cloud via the Amazon EMR big data platform and runs preprocessing using Spark. It allows for a wide arrange of user settings which enabled our client to integrate it into interactive dashboards. These can then be used by business to develop the buying strategy that optimizes for a variety of criteria such as expected revenue while imposing restraints on budgets per product groups, stock and marketing budget. 

Impact

The forecasted numbers have been used both in the US and Europe with positive feedback from the client. The overall estimated potential of the solution is to deliver a 6 percent reduction in costs. On top of that, the solution speeds up decision making through automation and helps buyers to make final decisions by providing a good benchmark for the majority of articles.