The solution improved the accuracy of credit risk predictions, sped up the onboarding process, and reduced the rejection of viable businesses. The project helped Factris prioritize high-value clients and manage financial risk more effectively. The models have since been fully integrated into Factris' systems, and the solution can be adapted for other industries, including retail, logistics, and healthcare, for data-driven risk management.

The challenge

In the factoring industry, accurate credit scoring is critical for evaluating the risk associated with financing SMEs. Traditional credit scoring models tend to be limited, especially when applied to smaller enterprises, as they often lack the data coverage and sophisticated factors required for precise evaluations. For Factris, this led to slower onboarding processes and the potential rejection of companies that could have successfully repaid their loans.

Additionally, the client needed a solution that could not only predict whether debt would be repaid on time but also consider slight delays and other variables. The system had to be adaptable, capable of handling evaluations for new clients as well as existing debtors, and scalable for multiple regions with varying regulatory standards.

The solution

We developed a machine learning-based credit scoring system. The system predicts the likelihood of debt repayment and credit risk, tailored specifically for the SME factoring domain. The models use multiple data inputs, including company financials, court records, and historical payment behavior, to create accurate predictions and help Factris make better loan decisions. Notable features included API integration for real-time scoring and outlier handling for the most financially significant companies.

Implementation process

Data preparation and feature engineering involved cleaning and refining data, identifying key predictors, and generating new features to enhance model accuracy. For model development, two primary models were created: one for predicting repayment and another for evaluating outliers, such as companies contributing the most revenue. Integration and testing included embedding the model directly into Factris' systems using APIs and Docker containers for real-time processing. Regional adaptation required developing separate models for Lithuania, Latvia, and the Netherlands to account for regional differences in data and regulatory environments, with these models retrained based on new insights from each market. Upon completion, Factris fully took over the maintenance of the models, with AAI Labs ensuring a smooth onboarding process and comprehensive knowledge transfer.

Result

The machine learning models sped up the credit evaluation process, enabling faster onboarding of new clients. The project delivered a 96% success rate for accepted cases that were correctly evaluated and only a 4% error rate for cases that went bad due to delays​. This allowed small and medium-sized enterprises (SMEs) to access funding more quickly. 

The improved accuracy of the models also meant that fewer viable businesses were wrongly rejected, leading to an increase in approved loans and overall revenue for Factris. 20% of companies were responsible for 80% of the revenue, which shows how the model helped prioritize high-value clients and manage financial risk​. 

According to Aušra, Project Lead at AAI Labs, “One of the greatest achievements was seeing the outputs of the scoring models go straight into the client's system, playing a key role in decision-making processes.”

Future opportunities & applications

While the solution was designed specifically for this client, the underlying principles can be adapted to other regions and industries. The key challenge in transferring this solution lies in regional variations in regulations and reporting standards. However, with a solid foundation and deep understanding of which models and data transformations work best, future adaptations can be achieved more quickly than starting from scratch.

Additionally, industries such as retail, manufacturing, and healthcare can use these models for various applications like risk assessment, customer credit analysis, and improving operational efficiency. For example, retailers can enhance demand forecasting, while manufacturers can optimize inventory management. In logistics and supply chain management, predictive models can forecast vendor reliability and mitigate financial risks in complex supply chains. These models can also be applied to insurance, energy, or real estate, where accurate risk predictions and financial forecasting are crucial for data-driven decision-making and operational success.