An easy-to-use AI-based credit scoring system to provide microloans with 4 times lower NPL rates due to leveraging Big Data
Challenge
Our customer Mobicom Corporation is the first and the largest Mongolian-Japanese venture. The telecommunication company covers over one-third of the mobile service market. Recently, the company’s subsidiary Mobifinance (MonPay)launched a new way of issuing instant loans.
It allows end customers to apply for instant loans, get approval, and receive finance in no time.
The loan decision is based on a large amount of legally obtained data, available to the telecommunication provider: such as payment history or usage statistics.
The goal was to improve risk management and grow profits by decreasing NPLs, expanding the customer base, and automating decision-making. The outstanding loan number was growing and the business was growing. The in-house scoring system didn’t provide the results we were aiming for. Besides, it required a lot of time and effort from the team
The customer was searching for a powerful credit scoring product, which is able:
to work with Big Data
process imperfect data, like missing fields
evaluate the impact of certain data types on the output
create multiple models for experimenting and choosing the market strategy
take the maximum of what AI and ML can offer to build high-quality predictive models.
Approach
The GiniMachine team gathered customer’s requirements and explained the system implementation step-by-step. Fast onboarding and easy use of the product were of primary importance for the Mobicom team.
These objectives can be reached due to a set of heuristic methods along with a custom implementation of the decision tree ensemble method.
Such a combination helps to apply modern ML techniques to almost any raw data set and automate technical redundant tasks like data preprocessing.
This way company’s risk officers and data analysts get more time left for creative tasks, experimenting, and analyzing the resulting models. With the ready-to-use dataset, GiniMachine creates and validates a scoring model in 2-10 minutes, depending on the amount of data.
Result
Mobicom telecommunications company managed to reach its objectives and now it can grant loans in several clicks with the optimum ratio between risks and possible profits.
After 17 months of building prediction models and scoring new applications, the NPL rate dropped by 4 times: from 18.9% to just 4.4%.
The customer’s choice was to deploy the credit scoring system on-premise. It took a bit of additional effort related to the allocation of server capacity and balanced server performance.
The GiniMachine team detected the issue at a glance and helped to fix it at the testing stage, so the solution started building scoring models and showing a Gini Index of more than 0.6 right after, which is a good result.
The Mobicom team notes the high performance and reliable prediction power of models. It required no special training to start working on the solution. GiniMachine uses unstructured data, and big data, processes missing data, and finds non-obvious dependencies.
As a result, the customer benefits from fewer non-performing loans and higher acceptance rates. The tool can offer insights into lender’s data and can be used for exploratory analysis.
Mobifinance (MonPay) also started using GiniMachine for customer churn prediction and collection scoring, so soon first results can be measured and presented.
GiniMachine is a must-have for any non-financial businesses that are making their first steps in financial services. It works with behavior-based data and helps to work with AI/ML in a friendly interface and take data-centric decisions in seconds
Oyunchimeg Shagjjamba, CEO at Mobifinance NBFI LLC
