The challenge

One of the main challenges when MiniFinans provides fast consumer loans is to assess and rate creditworthiness of potential loaners correctly. MiniFinans wanted to improve the existing metrics to ensure interest rates would be paid back in a timely manner. The traditional rules of assessing potential customers are static, such as age, gender and geography, and the changing behavior of customers is not taken into account, which creates an uncertainty for MiniFinans every time they lend money. Since the loan service is provided online it called for a solution that can handle these assessments precisely and quickly.

“The algorithm has saved us time and lowered risk, because now we can assess loan seekers with up to 80% accuracy. This AI project positions us among the first in our business to do so, and it will shape the future of our business.”

The solution

MiniFinans engaged with 2021.AI to use AI and machine learning technology to give more precise prediction of interest payments by a potential borrower. The algorithm uses more than 30 variables and more than one million data points to make assessments of the potential borrowers. The algorithm provides full transparency, which enables MiniFinans to find out exactly why someone is either provided a loan or refused by the algorithm. This also makes the algorithm comply with legislation that says customers have a right to transparency when it comes to assessment of their creditworthiness. 2021.AI delivered the use case, the coding of algorithms, and production implementation of the algorithm. MiniFinans now uses the algorithm to assess and rate all new customers.

The result

2021.AI developed an algorithm and set it up on the Grace AI Platform. We received CSV files with the required data to be stored on the Grace platform and updated once a week. Subsequently, the incoming data was used to make real-time predictions of the payback behavior of potential borrowers and sent to the algorithm via an API.

2021.AI incorporated three algorithms and applied a majority voting method to combine the predictions. Those three member classifiers were used to achieve better classification accuracy. Normally, the entire classification performance of an ensemble classifier tends to provide better accuracy than a single classifier.

3 facts about the MiniFinans project


The algorithm was developed on Grace and is hosted on a Linux server on AWS and implemented in MiniFinans’ infrastructure. The requests are sent to the server via Flask.


The three algorithms incorporated by 2021.AI are Random Forest, Support Vector Machines (SVM), and Logistic Regression.


The data was based on Minifinans’ data storage on customers. The data included age, income, taxes but also used operating system, location, browser and many more.

Get in touch

Interested in taking AI into production and scale to every corner of your organization?