Our client, a debt collection agency specializing in communication and transport debt, faced a challenge in performing valuations of non-performing loans (NPLs) before auctions. In our customer’s market segment, relatively little data was provided by the portfolio selling entities. As a result, bids provided at auctions were made based on intuition.


We started the process by interviewing bidders and doing general analysis of the data provided to our customer before past auctions. It became apparent that despite a large volume of data being provided, its value was limited. As a result, our team concentrated on more general data inputs such as timing when obligations arose, historical performance of NPLs sold by the parter, age of debt, etc.

A model emerged, offering the highest accuracy, based on ARIMA algorithms. Our model predicted the percentage of the the face value that is to be recovered 12 months after purchase. The system was implemented using BRAINHINT’s AI Farm system and results were provided on demand. We built bidders a simple GUI to provide input data and download a report in the form of an Excel file.


Implementing our system allowed our customer to bid on higher-value portfolios while skipping unpromising auctions.

Accuracy of the model was over 17%, which as admitted by our customer, turned out to be significantly higher than an educated guess provided by an expert.

  • +17%

    accuracy of the model

  • improved efficiency of investment funds

Contact us

We will get back to you within 24 hours.