Collaborative Intelligence: Erschließen Sie den Wert von kombinierten Modellen

Collaborative Intelligence: Erschließen Sie den Wert von kombinierten Modellen

Wie können wir die Genauigkeit der Nachfrageprognosen für schnelldrehende Konsumgüter bei komplexer Bestandsplanung verbessern?

aXialyze hat eine Prognosetechnik entwickelt, mit der sich die Zahl der Fehlbestände bei schnelldrehenden Konsumgütern erheblich reduzieren lässt. Diese Technik ermöglicht es Ihnen, das richtige Produkt zur richtigen Zeit auf die wirtschaftlichste Weise zur Verfügung zu haben.

Challenge

There is an inherent complexity in the forecasting subject, especially for food retail companies.

With a continually growing number of products to forecast and the multiplicity of forecasting models, it is challenging to determine which model will produce the best forecast accuracy for a given product in a given period. There is also a need for constant trade-offs between target service level and inventory costs.

Combining forecasts models using ensemble learning methods can improve forecasting accuracy, thus boosting profitability. Furthermore, it makes sense to steer the learning models towards optimistic or pessimistic forecasting depending on the product’s characteristics.

Methods

aXialyze saw a new way to combine forecasting methods in retail supply chain management. Tremendous value can be created by combining existing methods that are practically computable in a retail environment with thousands of forecasts to be generated on a frequent basis.

Moreover, instead of just using simple combination models, aXialyze makes use of Ensemble Learning techniques. Hence, individual models are combined with multiple machine-learning models to improve the stability and predictive power of the forecasting model.

Finally, one can complete the value of this method for decision making by differentiating treatment of products based on their ABC rotation classification. You can however also go beyond that and apply the preffered bias at article level using an item-by-item approach.

To harness the power of this approach, we introduce an asymmetric loss function to bias the model towards a specific treatment for each product category. Hence, you can individualise the learning method for each article.

This was the last step in creating tools that can be individualised to the product categories based on ABC and hence let you decide on the most efficient stock level.

  • Products in very high demand (A) can never be stocked out
    hence the forecast is biased towards overprediction, eliminating variability in demand.
  • Products in low demand (C) often just take up space and tie up capital
    hence the forecast is biased towards underprediction, enhancing profitability without losing customer loyalty.

Results

Employing combination techniques in forecasting had an essential impact on the two key profitability drivers in the business, product availability and stock levels. Boosted accuracy allowed for more efficient inventory management, which in turn translates into an improvement in service level while reducing safety stock.

This showed that efficiently putting consumer expectations and needs at the centre of inventory planning can even have economic benefits, improving your business performance.

1920 1024 Main
Share
Previous Post
Innovative Forecasting-Methoden für ein agiles Unternehmen: Strategien und Lösungen
Close