© The Indian Express Pvt Ltd
Tags:
Journalism of Courage
The introduction of the Unified Lending Interface (ULI) will be a “big boon” for artificial intelligence (AI) players in the financial services space, according to Joydip Gupta, head of Asia-Pacific at Scienaptic AI. The New York-headquartered AI-driven loan underwriting platform company – which is looking to expand it Indian client-base to private banks from non-banks and small finance banks such as UGRO Capital, Uni Cards, and Jana Small Finance Bank – doesn’t see the ULI as a competitor, with Gupta saying it will help remove a huge obstacle for firms such as Scienaptic AI: the lack of standardised data. The ULI, which is currently at a pilot stage, is an initiative by Indian authorities to allow easy access to authenticated data from various sources, with lenders expected to connect to this platform through a ‘plug and play’ model. In an interview with The Indian Express’ Siddharth Upasani, Gupta also discussed the recommendations of the Reserve Bank of India’s (RBI) recently-released report on developing a framework for responsible and ethical enablement of AI in the financial sector, efficiency gains from AI, and the type of jobs that would be at risk. Edited excerpts:
What role is Scienaptic AI playing in the financial sector?
It is mostly underwriting and adjacent areas. Fraud is an adjacent area and one which is becoming very prominent – regulators are trying to put early warning signals and almost mandate that. A big use-case now is what happens after a loan gets disbursed: portfolio monitoring, early warning signals of stress, cross or up-sell potential to the customer – or pre-approving someone even before they come for a loan.
At the very base level, we have a platform that can be provided to the lender either as a SaaS (software as a service) offering or an on-prem (on-premises) deployment, where the lender can – without doing any coding, but just by knowing Excel and PowerPoint – be trained and they can create the journeys like on a piece of paper. Ultimately, whatever they’re doing manually can be automated fully.
To make this platform operational, we connect it to various data sources. We have partnerships with 30-40 different APIs (application programming interface). They are connected based on the client’s requirements and the whole journey gets executed in less than a second. Some clients might want to improve any internal rules and ask us for suggestions. That’s where we bring in our scorecards and models. Those are our IP (intellectual property), which we train and keep validating and monitoring annually using real data. We can just drop that into the client’s platform.
Would the ULI end up competing with you, considering it has the government and the RBI’s backing?
Actually, that won’t be the case. We don’t produce the raw data – credit bureaus, banks, and others do that. We take the raw data, enrich it, and build intelligence on top of it. This completes the picture for a lender. What does the lender want ultimately: should I give a loan to this person? If so, how much should I give? And if there are slightly higher risks, can I change my interest rate? That is where we come in.
What the ULI is essentially saying is that there will be a unified interface where the data providers will provide the data and people like us, who are helping banks consume the data, will use that data. Today, if you ask an SME (Small and Medium Enterprise) for their financials, they will give you something in Excel, a printout, or at best something from Tally; there’s no common system. Yes, there is the RoC (Registrar of Companies) and it will give some data. But account aggregators still don’t give bank statements for the larger current accounts; they give for individual accounts and proprietorship current accounts only. In fact, a company like us struggles because we go with our models and clients say they don’t have any data and ask us to scan PDFs. We do it, but that’s not our forte.
So, ULI (will be) a big boon to players like us because, hopefully, it will incentivise the givers of data to give the data and it will lead to the standardisation of schemas and let us focus on what we do best.
The RBI report on developing a framework for responsible and ethical enablement of AI in the financial sector wants to promote AI in the manner of Unified Payments Interface (UPI) for payments: it wants a fund for data and compute infrastructure and high-quality financial sector data infrastructure to help build models. Will the strategy that worked for UPI and payments work for AI in financial services too?
The committee’s proposal makes sense, but AI presents different challenges than payments. UPI succeeded because payments needed a standardised infrastructure. AI needs something more sophisticated: shared infrastructure to democratise access to high-quality data and compute, plus an AI innovation sandbox for testing.
The committee’s recommendation for a dedicated fund to finance indigenous AI model development addresses a real problem. Unlike payments, where competition happens at the application level, AI benefits from shared foundational infrastructure while companies compete on implementation – think of it as building better roads so everyone can drive faster, not forcing everyone to use the same car.
India’s financial data is fragmented across institutions, making it hard to build representative models. A shared backbone could help smaller players access quality data and compute power they couldn’t afford alone, while preventing the bias that comes from training on narrow datasets. This levels the playing field without stifling competition. Companies will still differentiate on how they use these tools to serve their customers better.
Your clients are primarily non-bank financial companies (NBFCs) and fintechs. Is there a tendency for banks to depend more on in-house models?
Yes, the clients are mostly NBFCs and fintechs. We do have a few banks like Jana Small Finance Bank and Unity Small Finance Bank. We are also talking to a couple of other small finance banks. But the lion’s share, yes, is not banks.
Banks tend to have large in-house teams and a sea of data scientists building models. But they do look at products from startups in a PoC (proof of concept) or PoC++ way. They don’t take these products to replace their current main processes. And because of our size and bandwidth, it’s tough to go to these banks and play with their PoCs for a year or two and wait for them to make their decision. So, it’s just that we have taken a conscious stance to start with fintechs who are the early movers on AI and digitisation and then to the NBFCs. Now, we are getting in front of the banks. Private banks are next in our journey.
A big reason to use AI is efficiency gains. Are these gains meaningful in the underwriting process and banking in general?
Let’s say you have a branch-based business and you are giving microfinance or MSME (Micro, Small, and Medium Enterprise) loans to non-metro, non-salaried, non-Tier I prime customers. Typically, you get some data from them and an underwriter will punch some numbers in Excel and calculate certain ratios and based on them recommend a loan amount.
Now imagine if this is standardised across all branches while maintaining legitimate branch-to-branch differences: if the east zone has a higher risk, there should be a different tolerance for it. A tool like ours allows you to do that – it preserves the nuances of the different segments of a business even while standardising the whole process. Of course, 10-20 per cent of decisions might require a human call. But 80 per cent, let’s say, can be standardised. If a company wants to double in size in the next two years, they would have had to double the number of underwriters. But if they’re using a tool like ours, they will only add 10 per cent more because 80 per cent of the work is automated. That is the efficiency we are bringing in today.
Tomorrow, sentiments of a customer could be captured during a conversation with a branch official through the audio of the conversation and qualitative comments written in a tablet. If there was a Large Language Model (LLM) which can capture all these different forms of data – audio, video, unstructured text – and give it a structured meaning, which then feeds into a model like ours, then that one conversation replaces two-three visits to the customer. I think that’s not too far away given the rate at which LLMs are coming.
It’s difficult to have a conversation about AI without talking about job losses. What sort of roles is Scienaptic replacing and what would the typical worker in these roles look like?
Currently, because of platforms like ours, the need for underwriters is less for processing X number of files. So far, because the Indian economy is growing, it has not led to reduced underwriter jobs; companies are able to grow faster without having to hire that many people. But if this continues and broadens, I think one of the jobs that will obviously be at risk is underwriting. This is not a very high-end job. So, they will have to upskill or reskill themselves.
But it’s not just underwriters. Take a Loan against Property, for instance. If you look at the handoffs in the whole process, it is not just underwriters, but a bunch of middlemen – somebody who takes files, goes to the field, etc. If the process is digitised and automated, the operational intermediaries will go away. So, COOs (Chief Operating Officers) of banks are really looking forward to AI to make their operations very efficient. But at the same time, that’s where jobs will be lost. And it will mostly be in tier-III cities. In tier-I cities, a lot of this has already been digitised.
But there is a whole new aspect which may lead to job creation. AI, in the future, could help banks identify opportunities to lend that they may be ignoring today. If a drone, for instance, can inform a lender about a farmer’s field – what he grows, his productivity – then the bank can forecast his cash flows using estimates for demand for what is being grown, change in price of that item, rainfall, etc. And if the bank has information on the farmer’s family members – children, their health, age – then the income and expenses can be forecast. Suddenly, banks can make a much more educated guess and could be willing to lend that farmer some money.
