Client story: financial services
Optimizing the prediction of loan repayments with 80% accuracy
Enabling real-time predictions of potential borrowers’ payback behavior with high classification accuracy
A financial services company enables quick and transparent access to capital through smaller consumer loans for individuals. They believe that people should have the freedom to decide how, when, and where they wish to spend their money and aim to become the preferred loan provider in Denmark.
With 2021.AI, the financial services company has innovated and improved customer assessments, resulting in more precise predictions for potential borrowers’ loan repayments.
Download the case
Download the PDF-version of “Optimizing the prediction of loan repayments with 80% accuracy” client story
The challenge
Small consumer loans are almost always unsecured, meaning that the lender will not demand collateral to accept and approve a loan. There are, however, some minimum requirements that the borrower should be able to meet. The main challenge for the financial services company is therefore to ensure that they can quickly and correctly assess and rate potential borrowers’ creditworthiness.
As part of this effort, they wanted to improve their existing metrics for evaluating loan applicants. The historical metrics for assessing and rating potential borrowers, such as age, gender, and geography, were static. They did not consider borrowers’ changing behavior, which created a degree of uncertainty every time they lent money.
Since their loan services are available online and most of the small loans can be applied for and accepted in a matter of minutes, the company needed an automated solution to provide real-time creditworthiness assessments.
The solution
The financial services company engaged 2021.AI to leverage AI and machine learning technology to give a more precise prediction of loan repayments by a potential borrower. They shared customer data points, including age, income, taxes, operating systems, location, browser, and more. Based on this data storage, 2021.AI set out to improve the classification accuracy using an ensemble classifier.
Typically, the classification performance of an ensemble classifier provides better accuracy than a single classifier. Therefore, to achieve better classification accuracy, three algorithms were deployed: Random Forest, Support Vector Machines (SVM), and Logistic Regression. A majority voting method was applied to combine the predictions. The algorithms were developed, deployed, and continuously monitored by leveraging 2021.AI’s platform, Grace.
The results
The AI algorithms deployed use more than thirty variables and over one million data points to assess potential borrowers. The algorithm provides complete transparency, enabling the company to determine why the algorithm either provides someone with a loan or refuses them. This transparency also ensures the algorithm complies with legislation that says customers have a right to information regarding the assessment of their creditworthiness. 2021.AI delivered the use case, the coding of algorithms, and the implementation of AI algorithms. The financial services company now use the algorithm to evaluate and rate all new loan applicants.
Project highlights
- Building an AI solution to enable real-time predictions of potential borrowers’ payback behavior.
- The deployed algorithms achieved 80% classification accuracy.
- Significantly helped lower the risk of accepting new loans
- Saved 1m-2m DKK per year
About the project
The project set the foundation for other models in the company including lead generation model and a fraud model.