Your browser doesn’t support HTML5 video
Imayan Digital
Despite digital advances, the self-employed struggle to get their first formal loan.
Banks may intend to start with small loans, but are constrained by the lack of a standardised mechanism to assess them without:
Fintech startups in emerging markets like India are using AI-based credit scoring for micro-loans to serve the new-to-credit customers.
They leverage alternative data and AI to reach segments previously unserved by formal credit systems.
Despite advancements, delinquency risk lingers, potentially amplified by aggressive growth tactics.
With traditional input, significant features of AI models mirror that of traditional rules engines. Even SMS data offers marginal improvements, if any.
Confined by the input, AI can at best enhance speed and avoid human biases, but its transformative potential may remain untapped
I believe that key to underwriting the unserved lies in diverse, behaviour-based data.
Searching for a parallel model to learn from, I had an unlikely, yet intriguing, inspiration from the local loan sharks!
Beyond their debt recovery skill, I see, informal lenders rely on 3 key strengths to decide:
First 2 strengths are a byproduct of their proximity, however, judging the borrower by observing seems to hold the key.
The real challenge and opportunity is in extracting this valuable insight and structuring it to replicate at a large scale
At the heart of this solution is a computer vision model that analyzes facial micro-expressions during vernacular psychometric questionnaires.
The target is repayment behaviour. Therefore, it is essential to collect the debt repayment data for our sample over a duration of 6 to 8 months.
This is presuming that most loan products in India have monthly repayment cycles.
Roughly 10 months from the start of data collection, we enter the phase of analysis and model building
Correlate the repayment data with the earlier recorded facial micro-expression data in the context of the psychometric responses and their demographic data to identify predictors of loan repayment and default to arrive at the digital credit underwriting model
Post model building, we test our new system. A suitable DOE is recommended here:
The repayment performance shall be continuously monitored over the 6 to 8 month period.
At the end of the pilot period, the performance of the test group (approved through our digital underwriting model) is compared against the control group to evaluate the performance of our innovative approach to decide next steps
Using the insights gained from this pilot, refine and improve the facial micro-expression model. This could involve adjustments to the underlying algorithms, the psychometric questionnaire, or the criteria used to interpret the facial micro-expression data.
Provision collection and use of hyper-local information similar to the loan shark lesson, to refine the decision
Facial micro-expression analysis already has proven successes in fields such as fraud detection, personalized advertising, and emotion tracking.
Its potential was on full display at the Singapore Fintech Festival (SFF) wayback in 2019, where they demonstrated real-time, high-accuracy emotion detection on a group of individuals simultaneously.
Any credit assessment technology must respect local and international laws, including India's upcoming "Digital Personal Data Privacy Bill," ensuring data privacy and security
Ethical considerations such as preventing bias in AI, respecting user privacy, and providing an avenue for credit decision disputes are crucial
Imayan Digital LLP is a DPIIT recognized Fintech startup striving to simplifying value creation through Digital Transformation.
We welcome feedback, ideas, and collaborations to refine this innovative idea further. Let's revolutionize the lending landscape together!
The Author, Muthukkumaran K, is the Founder and Managing Partner at Imayan Digital LLP. An Engineer from GCT and an MBA from IIM Lucknow, Muthu comes with 13+ years of corporate experience. Muthu has successfully led various ML based Fintech projects for a leading Bank and an NBFC
This bite-sized article presents one potential solution in Digital Underwriting for assessing credit risk using technology and is intended for informational purposes only. It does not guarantee success and the proposed system needs to be tested thoroughly to ensure ethical and practical considerations are met. It is crucial to comply with data privacy laws and obtain informed consent from all participants. The ideas and information contained in this article do not constitute financial advice