Technologies used

Jupyter Notebook

Jupyter Notebook

Financial Services

Micro-lending using alternative data for a multi-national bank

Impact by the number

 

35%

less time taken to make credit decisions

<10%

default rate

Financial Services

Micro-lending using alternative data for a multi-national bank

The challenge

A leading bank identified the opportunity to diversify its product suite and offer digital credit services to branchless banking customers. Our customer envisioned the services to provide an instant and convenient way for the customers, who use smartphones, to avail short-term loans of small ticket sizes. However, for their solution to be effective, our client needed to assess the credit risk of the mobile banking customers, who mainly consisted of people with insignificant credit history and/or transaction data.

The solution

Addo developed an AI-powered credit risk assessment platform using diverse internal and external sources of data, such as transaction history, account balance history, loan repayment history, smartphone data, online application form data, social media profile data, etc. These data points were analyzed using the three core modules of the platform to generate a credit worthiness score for each applicant; these included:

  • Account Balance Prediction Module: AI techniques were used to predict the account balance of customers at the date of loan maturity. This shed light on the repayment capacity of customers. Customers with a high predicted account balance were considered better candidates for loan approval.
  • Fraud Detection Module: A set of logical rules were applied to the solution that were aimed at preventing fraud. Along with this, AI techniques to detect fraudulent behavior and anomalies were employed, which took into account past user behaviour, and compared it to the behaviour of the entire population of users.
  • Credit Scoring Module: An expert system was developed that assigned relevant weights to each variable and came up with a credit rating for each customer. Addo used a set of best practices and our experience of creating such systems in the past for this.

We outlined the alternate data acquisition strategy and process flows, designed the high level architecture, and modelled the risk assessment engine using several machine learning techniques.

To build this solution, a pool of highly skilled Machine Learning Consultants and Machine Learning Engineers were engaged. 

AI techniques used: Hybrid Neural Networks, LSTMs, Stacked Auto-Encoders, Random Forest based Regression, Multiple Cox Regressions

The results

  • Reduction in time taken to make credit decisions
  • Reduction in default rate
  • Improved customer experience
  • Data driven and automated credit scoring process
  • Increased accuracy of credit scores
Looking for a
similar project?
Let's talk

Related case studies

Financial Services

Prediction of risky driving behaviour in the elderly

Read more

Financial Services

Customer service chatbot optimization for a leading bank

Read more