Falke KGaA
FALKE is an internationally active family-owned company specializing in high-quality apparel. To ensure the availability of its products worldwide, the company must plan for projected sales accordingly. To this end, FALKE had already implemented a sales planning system based on Board in collaboration with celver, the results of which are then incorporated into production planning.
The Company: Uniqueness in Series Production
Since 1895, the family business FALKE has stood for fashionable clothing made from high-quality materials, processed with perfect craftsmanship and attention to detail. The company produces sweaters, bodysuits, tights and stockings, underwear and sportswear.
The facts at a glance:
✓ Simplification of the planning process
✓ Less manual work
✓ 21% improvement in forecast quality
✓ Trust in the data
The challenge: Too much manual work
To ensure global product availability, projected sales must be planned accordingly. To this end, FALKE had already implemented a sales planning system based on Board in collaboration with celver, the results of which are then incorporated into production planning. The system provides key insights for subsequent processes: When will which types of socks be needed, and in what quantities? And which colors and sizes are in demand? All these insights also impact subsequent processes, such as which yarns need to be purchased for production. However, previous forecasts were implemented in a very complex manner and required significant maintenance effort. Since the forecast quality of the statistical method used was insufficient, sales planners had to repeatedly intervene manually to obtain plausible results.
“We currently have approximately 1,500 different items in our product range,” explains Werner Redicker, Head of Sales Planning at FALKE. “With a range of this size, manual re-evaluations simply aren’t feasible in the long run.” The company was therefore looking for a forecasting method that would provide more reliable data and thus reduce the workload for the relevant department.
The solution: Precise forecasts for each item
The data scientists at celver therefore worked with the FALKE team to test 15 different state-of-the-art approaches, 12 of which are now in use. These range from statistical methods such as ARIMA and Holt-Winters-Brown, to neural networks like Deep AR and N-BEATS, and machine learning with ARIMA Boost: For each item, the system automatically determines the most suitable forecasting methods each month based on their predictive quality and combines them accordingly. Subject matter experts were involved from the very beginning to provide direct feedback on the plausibility of the results. “It quickly became clear that the quality of the forecasts could be significantly improved this way,” recalls Werner Redicker, referring to the initial test results.
This new approach is currently being applied to FALKE’s basic items, as they have a more stable sales history compared to fashion items in current trend colors or seasonal patterns (e.g., St. Nicholas motifs). However, even among the basic models, there were some outliers to consider. For example, customer behavior changed significantly during the COVID-19 pandemic: instead of sheer tights, significantly more hiking socks were sold. These temporary special effects had to be factored into the forecast to avoid distorting the results.
But even without a pandemic, there are always outliers that must be factored into the forecast. These include extraordinary events, such as product relaunches. For example, as part of a sustainability campaign, certain products were taken off the market and then relaunched with accompanying promotional campaigns. To ensure stable forecast figures, the resulting peaks in sales figures had to be smoothed out. To do this, the items were linked, and the unusually high sales figures were distributed across the preceding months with lower sales.
“We were truly surprised at how quickly we were able to improve our forecasting accuracy in collaboration with celver,” says Dr. Paul Schneider, Team Leader of Data Analytics & BI at FALKE. “The core project was completed in just two months—and entirely remotely. However, our collaboration with our business departments worked wonderfully even online and delivered excellent results.”
At the beginning of the month, FALKE’s sales planners now receive automatically generated figures for the annual forecast by item. The current forecasts are imported directly into the existing Board applications via an SQL server—the associated calculations take place in the background without the user noticing.
“A major advantage in terms of user acceptance is that the planning process itself hasn’t changed,” Werner Redicker sums up. “Our planners now simply receive much more accurate figures in the system they’re used to, allowing them to continue working with confidence.” Compared to before, forecast quality has improved by 21%. As a result, those responsible for planning now have to worry significantly less about basic items and can focus entirely on more complex issues.
In recent years, planning has been heavily influenced by the effects of the pandemic and disrupted supply chains. “We’re slowly getting back on track,” Werner Redicker concludes. “Now we’re regaining a sense of actual demand and can further expand the potential of the new forecasting methods.”
Your contact person
