How AI/ML-based credit scoring accelerated a lending application to become an industry leader

Financial industry use case

Scoring in Mongolia at the time

There were mainly two problems with credit scoring that were being used in Mongolia at the time. First, there was no automation used in scoring systems, and credit scoring had to be done manually for each loan request. Due to its heavy reliance on manual work and high operational cost, financial institutions could not offer lower amounts of loan product which is micro-loan. Second, limited types of data sources used by financial institutions, including credit history, proof of steady income, and collateral, meant that people who have no financial footprint or steady income could not afford a loan.

Implementing first AI/ML-powered credit scoring in Mongolia

Our new credit scoring models solved these previously mentioned problems. With end-to-end automation, from data collection to processing and final approval, customers can finish the entire process within 5 minutes, all from their smartphone. We employed multiple models for this scoring system, including random forest and artificial neural network, and created a model that produced the most accurate results. This Scoring model continuously learns and updates itself based on new data sent from the loan management system.

Combining credit and behavioral scoring

Once we implemented credit scoring and acquired a sizable number of customers, we also implemented behavioral scoring to assess and identify high-value customers and upgrade their products with increased credit. Our behavioral scoring for LendMN utilizes customers’ various kinds of in-app activities, and more importantly, their repayment tendencies.


  1. 96% repayment rate compared to traditional financial institutions being around 80%
  2. The application process under 5 minutes
  3. Market leader for micro-lending service
  4. Disbursing 180,000 loans monthly on an average

And Solutions’ Credit Scoring Solution

One of the capabilities of the AND e-wallet that SuperUp benefited from was the e-wallet’s capacity to handle high-frequency transactions. As the SuperUp has many mini-apps and services, the transaction volume was high,
and it needed a wallet that could handle such volume smoothly without failures.

The AND wallet is built with open APIs, which means it has robust integration capabilities with third-party systems such as bank accounts, other e-wallets, bank cards, and over-the-counter channels. As a result, users could top-up their wallets and transfer funds to any channel of their choice. Since the wallet was also integrated with bank payment gateways, users can quickly transfer money between e-wallets and bank accounts without any fee.

Data Collection

Direct questionnaires
User profiling and demographic data collection through direct questionnaires to customers via multiple mediums.

Device information
Device information such as manufacturing date, price, phone calls, contact information, application data, etc., can serve as an additional and accurate data stream.

Third-party data sources
Along with the data from credit bureaus, banks, and other financial institutions, third-party data such as telephone data, utility bill data is also collected to have a better and more accurate look into the customers’ behavioral tendencies.

Data Analysis

Data structuring
Collected data is sanitized and stored with the desired structure for efficient data operations.

Data labeling
Tagging the raw data with one or more contextual attributes for use in model training.

Feature selection and extraction
Identifying, manifesting, and ranking dataset features that make the most sense of data based on accuracy and sensitivity.

Model Training

Our scoring solution uses multiple AI/ML algorithms such as Random Forest, Artificial Neural Network, KNN, etc., based on the business objectives, goals, and nature. It then creates an ensemble model consolidating results from the utilized models.

Self-learning and updating

The active learning model works continuously and is fed with new data sets for supervised/unsupervised learning to make the system more intelligent and accurate.


Our base gamification schema with a loyalty system helps businesses access and identify desirable customers by generating key features that are highly suitable for modeling.