AI in finance: interview with Project Lead on innovative factoring solutions

We sat down with our researcher Ausra to discuss how AAI Labs helps to revolutionise approaches to credit scoring by using bespoke ML models.

We appreciate you speaking with us today. Could you please start with introducing the project?

The project initially aimed to solve the credit scoring problem in the factoring domain for a factoring client operating in Europe. Three countries of the client's operating locations were chosen for the project based on market familiarity and data availability: Lithuania, Latvia and the Netherlands. Because of the client's business model, we wanted to predict not only whether the debt would be returned on time, but also whether it would be returned with a slight delay of a couple of months. This problem was what we initially started with, but during the project the scope eventually expanded, and we solved related problems for the client: predicting the probability of default of a company, the appropriate credit limit based on the data of debtors and their clients. We improved several processes for our client: provided evaluations of new clients that were used during the company's onboarding process as well as separate evaluations of debtors already existing in the client's portfolio which are useful during the annual review process.

Can you tell us more about the problem that required a new solution?

The problem existed long before we started working on the project. In factoring, companies are usually concerned with conventional loans. For SMEs, the current models lack accuracy and might not cover all the factors that clients want. So, we built these models to solve their day-to-day problems: when a new company comes, the lender needs to evaluate the potential success of the case. That is why we created a machine learning model, which is now used in the client’s daily operations.

Since the project is over, can you elaborate on your personal view of how AI affects factoring companies?

The results of this project benefit companies in several ways. First of all, due to using machine learning models instead of manual processing of new companies, the process of onboarding is sped up. This means that companies can get funding faster. Also, when working on the models for the client, we paid attention to creating the models that, so to say, don't leave money on the table. Due to the previous procedures some companies might have been rejected in the past by manually created rules of evaluation, although they might have returned the loans successfully. Our machine learning models aim to reduce the amount of falsely rejected cases which brings the client additional income which would have been missed otherwise. Thus project has a long-term financial benefit to the companies in need of loans thanks to the more accurate evaluation by our developed model.

Do you see the potential for this product to be used outside of this region? Can it be adapted elsewhere?

Although the results of this project can't be simply transferred as they are, they can be adapted to other regions. The main problem is having different regulations and reporting standards in different countries. But we now know what works and what doesn't, and which models and data processing transformations are effective. So, it would be easier to do this than creating everything from scratch.

What was the greatest achievement during the project?

For me, it was exciting to see the outputs of one of the scoring models go straight into the client's internal system. The realisation that our models were essential in important decision-making was great. Close and constant communication with the client was very important, and it made the project results better quality.

Were there any risks associated, how did you mitigate them?

Every model has certain risks, usually associated with the accuracy of predictions and their changes over time. We did a lot of testing of our models and recommended time frames for the client when the models should be retrained. This was to account for changes in the market and global situations.

Do you still participate in the maintenance of this model, or has the company taken over?

We have fully onboarded the client team, and the company has now taken over. They took over the code, the training process, and the whole knowledge of the model. One of the goals of the project was to make sure that the client is familiar with the maintenance process and has full ownership of the developed solution.