0
The biggest challenge today is that underwriters do not have time to review a high volume of loan applications and make decisions at the pace the market demands. Besides, biases are introduced into the loan application process when assessing whether someone is a creditworthy borrower - biases that are sometimes conscious, sometimes structural, and sometimes simply the result of an underwriter applying heuristics that were never designed to be equitable at scale.
Here, I am in conversation with Vasudevan Swaminathan, president of Zuci Systems. We discuss how their platform, Halo, helps in credit underwriting using machine learning.
The machine-learning credit underwriting platform takes into account both primary and alternative data. Primary data is data lenders provide, while alternative data comes from various sources, including fraud data from players such as TransUnion and Experian.
Traditional credit underwriting was built for a different lending environment. When the volume of applications was manageable, and the borrower profile was relatively homogeneous. A manual review process backed by a credit score worked adequately. The underwriter had time to review the file, the credit score provided a standardized signal, and the lending institution could make decisions within an acceptable timeframe.
That environment no longer exists for most lenders.
The volume of applications has grown significantly as digital lending has expanded access to credit. The borrower profile has diversified as fintech lending has reached segments that traditional credit scores do not capture well. And the expectation for decision speed has compressed from days to hours to minutes as digital-first borrowers compare lenders based on how quickly they can get an answer.
ML payment risk assessment addresses these pressures directly. A machine learning model can process hundreds of variables simultaneously, draw on both primary and alternative data sources, and produce a credit decision in a fraction of the time a manual review would take. The model does not get tired. It does not carry the cognitive biases that an individual underwriter brings to a file. And it improves over time as it processes more data and receives feedback on the outcomes of the loans it approves or declines.
The lending ecosystem is evolving, and multiple players are targeting low-value lending.
It is moving away from traditional lending models based on credit scores alone. This environment is critical for individuals and businesses to grow. Every financial institution, microfinance company, and fintech app requires objective, logical information to support underwriting.
AI credit scoring for payments has become a distinct discipline within the credit underwriting space due to borrowers' payment behavior. It is one of the most reliable signals of creditworthiness available.
A borrower who consistently pays their bills, manages their digital wallet, and maintains standard transaction activity is demonstrating financial behavior that a traditional system cannot capture.
Halo is lender-agnostic, and it can work in any environment - independently or in collaboration with an existing engine. It can process cash advances, personal loans, auto loans, and complex business loans. The success of Halo depends on the quantity and quality of data available.
Data is critical, especially clean data. Data is one thing. Cleansing it and having the right data is another. Halo first identifies that the data is good, and from there, it starts building the model. It takes time to get the desired results.
50% accuracy is what Halo starts with, and it helps you progress your accuracy levels from there.
The credit underwriting AI model that Halo builds is not a static system. It is a learning one. As the model processes more applications and receives feedback on outcomes — which loans performed well, which defaulted, which were approved or declined incorrectly - it refines its weighting of different variables and improves the accuracy of its predictions.
Machine-learning loan payment data is one of the inputs that make this learning process particularly valuable. Payment behavior is a leading indicator of financial health in a way that many traditional credit variables are not. A borrower's income level indicates their capacity. Their payment behavior reflects their willingness and discipline. Halo combines both to build a more complete picture of creditworthiness than either signal provides alone.
Fraud data from TransUnion and Experian adds a risk issue that primary lending data does not capture. A borrower with no negative credit history who appears in fraud databases adds a risk profile that a credit score alone would not identify.
Halo can help build more lending applications over time and help avoid biases that may be introduced into the system, whether through data or by underwriters.
The bias-reduction benefit of AI-based credit decisioning deserves particular attention. Bias in credit underwriting is not always the result of deliberate discrimination. It can be structural - built into the data itself if historical lending decisions were biased or the result of underwriters applying mental shortcuts that correlate with protected characteristics without explicitly using those characteristics as criteria. A well-designed machine learning model, trained on outcome data rather than demographic data, can make more consistent and less biased decisions than a manual process.
For microfinance companies and fintech lenders targeting underserved segments that traditional institutions have historically excluded, the bias reduction enabled by ML payment risk assessment is not just a compliance consideration. It is a market access consideration. Lenders who can demonstrate that their underwriting is objective and consistent can reach borrower segments previously locked out of formal credit due to the limitations of traditional scoring models.
The 50% starting point Halo begins with is often better than it sounds in a low-information lending environment. A lender making decisions on thin-file borrowers without any ML support often operates at a similar level of accuracy but lacks visibility into where errors occur and a mechanism to improve over time. Halo makes the accuracy visible from day one and improves it systematically.
Watch the video in which Vasudevan Swaminathan discusses the Halo platform and its application to credit underwriting in detail.
