Applied Machine Learning to Realtime Approve/Decline Credit for karyana Stores

The Client is a Pakistan-based startup providing online grocery shopping for customers all over the country. The online shopping platform, aiming to be the fastest-growing, offers a wide range of products and solutions to customers, quickly delivered straight to their homes.

As a rapidly growing business, The Client has high ambitions by expanding into new markets using state-of-the-art, cutting-edge technologies. In addition to their B2C online platform, The Client has also envisioned strategic B2B solutions for retail vendors.

The Pakistani market, although rapidly growing, is marred by lack of regulation and tons of irregularities in financial systems resulting in unorthodox lending practices. Despite that, credit lending is a vital aspect of a retail vendor’s stock management and is heavily relied on to run their operations.

Since most retailers are not registered with any financial authorities, creating a framework for assessing a vendor’s score is an extremely complex and technical undertaking. Most current lenders use traditional, non-conventional methods for analyzing the credit scores for unregulated vendors.

With millions of unbanked people who are regarded as ineligible for conventional lending, the need for smarter credit scoring solutions is evident. The Client and CodeNinja had a clear vision – to design and develop a smart AI-based Credit Scoring system to evaluate and determine vendor’s risk assessment.

Key Challenges

Creating an entirely new credit scoring mechanism based primarily on unregulated, non-conventional data points due to lack of information from traditional financial institutions
The client wanted up-to-date credit scoring and potential risk assessments and based on real-time data indicators.
To fabricate a framework to onboard possible vendors that show positive future credit potential.Those who otherwise would be ineligible to obtain credit under a traditional credit scoring system

The Solution

Evaluating vendor profiles based on a variety of data while integrating alternative sources of data, such as:
- total income,
- credit history,
- transaction analysis,
- personal information,
- organizational structure,
- number of employees and much more.
Constructing a mathematical model based on statistical methods and accounting for large amounts of information
Formulating a new way to analyze creditworthiness from alternate sources of data rather than utilizing the traditional scorecard method used for credit scoring.
Using real-time indicators to asses vendor’s creditworthiness and potential risk
Iterating the ML pipeline with multiple algorithms that differently weigh data in the credit scoring pipeline
Developing an elaborate fraud detection model for highlighting bad actors by analyzing multiple data streams to detect inconsistencies