This project shows how education, research and business can complement and strengthen each other. The students gained hands-on experience with real data and an actual business challenge, while FrieslandCampina benefited from their fresh ideas and scientific knowledge.

Credit limit modelling

Companies regularly buy products on credit and pay for them retrospectively. A limit on this type of credit is needed to prevent the accumulation of excessive debt. Determining an appropriate level of debt, however, is a difficult and time-consuming process, says dr. Inez Zwetsloot, the director of the AI4Business Lab. How can you properly and quickly model the right credit limit?

To answer that question, the three students developed a model that uses multilayer perceptron (MLP) neural networks to more accurately estimate customers’ creditworthiness. The MLP network is an advanced type of neural network that recognises patterns in data that traditional statistical models fail to detect. The students compared the performance of these MLP models with classical methods such as linear regression and decision trees. The latter are often easier to interpret, but are less precise than MLP networks. The students' analyses showed that MLP networks are also flexible enough to be applied to customer data.

From study assignment to a usable model

The students each developed a new MLP network application based on customer data from FrieslandCampina. They then each wrote their undergraduate thesis on the tool they had personally designed. “In the next step, they compared the different models to determine which one best fits the requirement, works best and is most easily explained”, Zwetsloot says. “Based on that comparison, the students jointly wrote a recommendation for the company, outlining the various pros and cons.” The entire project was completed in just twelve weeks. During the process, the students were in close contact with Zwetsloot and thesis supervisor Bart Lameijer, as well as Daniëlle Jansen Heijtmajer (Global Director Finance, ERM & Shared Services) at FrieslandCampina.

Upon completion of the project, the multinational decided to actually implement the tool. Daniëlle Jansen Heijtmajer wrote the following about the students in October 2024: 'They provided us with a dynamic credit model that we will use to ‘automatically’ rate new customers/refresh the credit assessments on existing customers – for us a substantial continuous effort, that is greatly helped by this approach.'

Learning by doing

At the AI4Business Lab, undergraduates, graduates and PhD students work on real AI challenges faced by businesses. In the lab, under expert supervision, they unleash their academic knowledge on practical application cases. The project described here is just one of the challenges submitted by businesses that the lab works on each year, Zwetsloot says. Businesses in the lab's network feed the programme by submitting problems that students may be able to solve. So FrieslandCampina submitted this issue itself.

Obviously, the AI4Business Lab expects and hopes that more student projects will have an impact in the practical business environment. This example clearly demonstrates that AI research is not just theoretically valuable; it can deliver tangible and innovative improvements in business processes.

Want to know more?

The Impact Centre involved is: AI4Business

• Dynamic Customer Credit Review, a case study with FrieslandCampina (2024).