- Pilar Lara
- Use case
The issue of raw material demand in perishables
To manufacture a product, it is necessary to use a perishable raw material with a very limited shelf life. This raw material is expensive to store and dispose of.
As it is a key raw material for production, the usual practice is to over-purchase to avoid any plant shutdowns. This leads to additional costs due to the purchase price, the disposal of unused material, and its storage.
Until now, planning has been carried out by an expert using a well-known commercial demand planning software. They mention that they are buying between 20% and 30% more than what the software and expert calculate, as it is a product with high demand variability.
There is a two-year historical data of detailed consumption of the raw material that we use for analysis.
An analysis is conducted to determine the factors that influence the consumption of the raw material, and all internal variables are identified. External variables such as weather conditions are also studied. There is potential for a more accurate prediction.
Proof of concept
As a proof of concept, a very basic predictor is presented using only internal variables, and a test is conducted with old data, resulting in an accuracy rate above 80%.
After analyzing the results, a positive evaluation is made as it is observed that the demand predictions are much more accurate than those of the demand prediction software.
Minimum viable product
For the minimum viable product, weather prediction variables from AEMET are added. Various models are tested, and a more sophisticated one is chosen compared to the one used in the proof of concept.
With the introduction of external variables and the new prediction model, a prediction accuracy of 90% is achieved. Additionally, the maximum prediction deviation does not exceed 15%.
An interface is provided that daily supplies demand prediction data, as well as graphs of demand status and prediction errors. For a few months, the demand expert uses the system alongside the prediction systems used until now. In the same interface, feedback is provided on the predictions for further analysis.
At the end of the trial period, the feedback is very positive, and only minor adjustments based on the feedback need to be made.
Final version and outcome
Once the model is retrained and the necessary changes are made, the app is integrated into the company’s intranet, providing access to the entire planning and logistics department.
The team’s confidence in the model is very high. Excess product margins rarely exceed 10%, and the amount of raw material that needs to be disposed of has been significantly reduced.
This obviously results in significant cost savings, and now the processes are more sustainable by maximizing the utilization of acquired raw materials. Since its implementation, there hasn’t been a single day when there has been a shortage of raw materials due to demand prediction issues.
As always, to maintain industrial secrecy, the data shown here is altered and/or anonymized. Therefore… everything is based on real facts except for the parts that we have completely made up 😉
If you find this use case interesting and believe it could be applied in your case, do not hesitate to schedule a meeting with us.