Demand forecasting for a Belgian dairy company

A leading Belgian manufacturer and supplier of high-end dairy products contacted our team for help with its production planning. They plan their production on a weekly basis, but this is no easy task as the company makes such a wide variety of products. There are also many uncertainties involved with supplying raw materials and purchasing end products. Demand forecasting is a technique that attempts to estimate future orders or requests. It can then pre-empt problems by identifying potential shortages in resources or capacity.

The problem

Fluctuations in demand are a major challenge for the supply managers at this dairy company. Furthermore, as increasingly personalised products and services enter the market, interactions with customers and customers’ expectations are also increasing. Raw milk is delivered and the company extracts skimmed milk and fats. They then recombine these two basic ingredients in different proportions according to the type of product they intend to produce. Four important aspects of the planning are:

  • Freshness and storage restrictions – The company produces natural dairy products which means it can only store these products for a certain length of time and there are also limitations on the quantities it can store while maintaining freshness.
  • Pack size and production planning – The dairy company can produce as many small bottles per hour as it can large bottles, though it uses fewer raw ingredients for the smaller pack sizes.
  • Preventing cross-contamination – After certain products have been produced, they need to shut down the production line and cleanse it before producing a different product. The company has to take into account the time needed to change products.
  • Risk of undocumented knowledge – – A single employee had taken care of all the forecasting work for over 15 years. None of his expertise had been written down or transferred to anyone else. This company risked losing all of this precious knowledge and expertise if he was absent for long periods of time or left the company.

Our solution

We tackled this problem in 3 stages:

1. Analysis

  • We looked at the existing demand forecasting to see where to integrate AI.
  • We identified the data sources that we could extract, including past sales orders, and shipping and price lists. We then looked at how to feed this data into the model.
  • We pinpointed any anomalies in the data.

2. Development

  • We developed a solution by selecting the best forecasting algorithm that could predict the required weekly volumes for each product line until we had reached the business success criteria (accuracy >90%, weekly basis, split by vendor, etc.
  • We fine-tuned the final model, debugged it, and got it ready for production.

3. Implementation

  • We looked at where the model would be deployed, either in the cloud or on the company’s existing infrastructure. We also looked at feeding back results into the existing forecasting process.
  • We created pipelines for the data to flow to the model.
  • We resolved the issue of ‘cold starts’ where a new product appears that does not have historical data. We used data from comparable products to forecast demand for these new products. We designed the AI model in such a way that it could ask questions to humans to support this process.

How we added value

  Wasted end products reduced by 30%

We optimised production schemes so the volumes required each week could be accurately forecasted

 Sales increased by 1 % as in-demand products remained in stock

 The company freed up staff (0.5 FTE per year) thanks to our automation of its manual forecasting process



Receive news about AI.
This field is for validation purposes and should be left unchanged.