Our goal is seamless integration of generative AI to enhance collections and customer support: Karan Mehta, Co-Founder & CTO, RING

In a recent interview with Express Computer, Karan Mehta, Co-Founder and CTO of RING (formerly known as Kissht), delves into the transformative journey of the digital lending platform. Spanning from 2015 to the present, RING’s digital transformation narrative reflects the seismic shifts catalysed by the COVID-19 pandemic and the advent of generative AI. Mehta provides insightful commentary on the evolution of RING’s operational processes, the transformative impact of India Stack initiatives, and the paradigm shift in customer attitudes towards credit. With a focus on leveraging AI for credit risk assessment, ensuring data protection, and enhancing customer engagement, Mehta offers a nuanced perspective on the intersection of technology and finance in India’s rapidly evolving digital landscape.

Please talk about some of your recent digital initiatives and your digital transformation journey thus far.

Our digital transformation journey has been spanning from 2015 until now. If we were to encapsulate this journey into a book, the last chapter would undeniably constitute the lion’s share—around 60%—as the landscape of digital transformation has undergone a seismic shift, particularly in the last couple of years, catalysed in large part by the COVID-19 pandemic. With the advent of generative AI, this trajectory has taken an even more unforeseen and dynamic turn, surpassing our initial expectations.

Back in 2015, while we identified ourselves as a digital lender, our operational process was far from fully digitalised. Customers were required to visit our website to apply for a loan. Subsequently, their applications would be forwarded to our manual team of underwriters, who meticulously scrutinised documents and customer data before rendering decisions. KYC verification necessitated customers to physically procure and submit documents, often leading to a protracted cycle of document submission and verification, extending up to six days on average.

This procedural bottleneck stemmed partly from bureaucratic hurdles and partly from infrastructural limitations. At that time, essential digital infrastructure such as UPI, Aadhaar, and robust mobile connectivity were still in their nascent stages. Fast forward to the present, the landscape has evolved remarkably. The ubiquity of Aadhaar-linked mobile numbers, now nearing 98-99%, has significantly streamlined KYC processes, rendering them seamless and efficient.

Reflecting on the past, if you had asked me in 2015 to envision an ideal wishlist for enhancing our lending journey, nearly every item on that list has now been realised. The advent of government-led initiatives like the India Stack, encompassing UPI, Aadhaar, DigiLocker, Account Aggregator, and initiatives by NPCI such as eNACH and e-Mandate, has revolutionised the financial ecosystem. These pillars of digital infrastructure have not only been erected but also function with robustness and reliability.

Having a functioning bank account is one thing; establishing trust is quite another. Many countries have introduced their own versions of UPI, but none have achieved the level of ubiquity necessary to make a significant impact. This is where our government has excelled. Additionally, the accessibility of affordable mobile phones and data has been a transformative game changer for us. What’s remarkable is how quickly Indians have adapted to this technology. People of all ages, including my 95-year-old grandfather, have embraced apps like WhatsApp with surprising ease. The effort companies have invested in user interface (UI) and user experience (UX) design has been met with enthusiastic adoption from customers.

What percentage of your business involves extending credit to SMEs or MSMEs?

While we don’t directly cater to the SME space, our focus lies on providing loans to individual consumers. Interestingly, a significant portion, approximately 60%, of our customer base comprises self-employed individuals. These individuals span various occupations, from artisans to vegetable vendors. As an organisation, we perceive these loans as consumer loans, despite the fact that they are often utilised for trade-related purposes.

Our approach involves evaluating the individual’s capacity to repay the loan, conducting KYC checks, and ensuring the suitability of the loan’s purpose. Although these loans may be utilised for business activities, we classify them as personal loans since our primary concern is the individual borrower’s financial credibility and repayment capability.

We enthusiastically support borrowers who seek funds for business growth initiatives and are committed to facilitating their endeavours.

Do you specifically utilise AI to assess credit risk for loans?

Let’s start by differentiating between AI and generative AI, as there’s a significant overlap in how we utilise both tools but for distinct purposes. Generative AI finds its forte in customer support, encompassing voice analytics, transcription, and related areas. Conversely, AI, specifically machine learning models, has been pivotal in our credit analysis journey, a trajectory spanning over six years.

Initially, our credit model operated as a Business Rule Engine (BRE). However, around 2017, we transitioned to a more sophisticated machine learning paradigm. Leveraging AWS services like Amazon SageMaker and Amazon SageMaker Studio, we’ve forged ahead in this domain. To put things in perspective, we’ve facilitated approximately 11 million loans to date, affording us a treasure trove of data for analysis. In essence, lending companies have the most comprehensive customer data compared to other sectors.

Our data repository includes a plethora of insights, ranging from borrower history to transaction records, offering a nuanced understanding of customer behaviour. Our objective is to decode customer habits and translate them into repayment patterns. Such complex analysis is beyond the realm of traditional human underwriting models due to the sheer volume of variables involved. Hence, we’ve harnessed machine learning using AWS to fine-tune our credit models, continuously expanding their feature sets to optimise repayment behaviour prediction.

Moreover, AI plays a pivotal role in our fraud detection efforts, where accuracy is paramount. The probability of default (PD) prediction, derived from extensive data analysis, serves as a linchpin for our organisation’s profitability. Any deviation or variance in these models directly impacts our bottom line, underscoring their criticality.

Our credit and fraud models serve as the bedrock of our organisation, underpinning our operational framework. With unwavering confidence in the accuracy of these models, we’ve built a robust empire atop this foundation. However, any unforeseen discrepancies in these models would necessitate a significant course correction, highlighting their pivotal role in our organisational landscape.

What measures are in place to leverage machine learning or AI for data protection?

This holds immense significance for us, primarily because it directly impacts our customers. In India, customers generally exhibit a lesser degree of awareness regarding data privacy. Drawing from my experiences, having spent considerable time in the US, I’ve become accustomed to the strong emphasis on data privacy prevalent there. The European approach is even more stringent, further highlighting the spectrum of attitudes toward privacy.

Within our organisation, there’s been robust discourse among the founders, particularly between myself, representing the technology and product facet, and the other business-oriented founders. My stance consistently revolves around ensuring that every decision aligns with what’s best for the customer. While we possess access to a significant amount of data, our objective is to streamline processes for customers rather than solely optimising default rates. Even if default rates were to fluctuate, the underlying principle remains centred on providing fair pricing reflective of the business’s operational costs.

However, there are non-negotiable boundaries that must be upheld, rooted in respecting customer expectations and privacy. In addition to my role as CTO and founder, I actively champion data privacy and information security within the company. This entails meticulous efforts to safeguard data usage and prevent any inadvertent breaches, ensuring data is accessed only for authorised purposes.

How do you compare the GDPR with the DPDP Act, 2023, especially regarding data retention policies and consent mechanisms?

Well, what we’ve observed is that the DPDP Act of 2023 draws significant inspiration from GDPR, particularly in its structure and provisions. Notably, India has shown some leniency in data retention, deliberately leaving room for scenarios where data preservation is crucial, such as under the Prevention of Money Laundering Act (PMLA). PMLA mandates data retention for up to five years, but many institutions, including banks, extend this period to seven years due to regulatory and operational considerations, as well as the desire to enhance customer engagement and loyalty.

For consumer-facing companies like RING (Formerly, Kissht), Zomato, or MakeMyTrip, data retention is pivotal for fostering customer stickiness and retention. While DPDP grants customers the authority to request data deletion, there are certain data sets, like KYC information, that must be retained per other regulatory mandates, even if a customer opts for deletion. This nuanced approach reflects the Act’s balance between individual privacy rights and legal obligations.

However, one area where the DPDP Act of 2023 is in establishing robust checks and balances at the organisational level. While it delineates obligations and regulatory oversight, questions remain about the practical implementation and monitoring of compliance, especially given the vast and decentralised nature of India’s business landscape. Nonetheless, the Act serves to clarify legal boundaries and obligations, providing organisations with a clearer ethical compass. As companies scale and face scrutiny from investors and legal advisors, adherence to data protection regulations will become increasingly imperative, driving a culture of compliance and accountability. While the DPDP Act marks a significant stride forward, some may argue that its implementation could have been timelier given the evolving digital landscape.

Is your storage infrastructure entirely cloud-based?

We’re fully committed to the cloud, and we have been since day one. Fortunately, as a CTO, I’ve never had to grapple with on-premises servers. On-prem servers are a potential nightmare, with incredibly long lead times. Just recently, we had a use case where we needed a server for our call centre, as the telephony system required an on-prem solution. The server vendor informed us that it would take six weeks to deliver the server to our office. Do you know how long it takes me to spin up an EC2 instance on AWS? Just six seconds. With a simple click, I have a machine up and running in six seconds, and it’s even ten times as powerful as what the vendor was offering in six weeks.

The cloud also empowers us to continuously upgrade our technology. In the past, adopting new projects or technologies meant procuring new servers and dealing with the hassle of discarding old ones if the project didn’t succeed. Now, with the cloud, we can easily run POCs, experiment with new frameworks or languages, and quickly switch over or shut down resources based on the outcome. This flexibility eliminates the stifling effect on innovation that was once prevalent.

Partnering with AWS has further streamlined our adoption of new technologies. Services like Amazon Elastic Kubernetes Service (Amazon EKS), which would have been challenging to implement on-premises, are now readily available as managed services in the cloud. This accessibility enables us to leverage these technologies effectively and stay at the forefront of innovation.

How do you propose achieving a balance between the potential threats and benefits of AI, particularly in terms of enhancing security measures?

What lies ahead is essentially an AI cat-and-mouse game, a dynamic we’re already witnessing unfold. The emergence of deepfakes, particularly in documents submitted by customers, has prompted us to deploy detection models to spot morphed or tampered images. Whether viewed positively or negatively, AI is poised to empower nefarious actors to engage in identity fraud, behavioural fraud, and a host of other malicious activities, for which we may not yet be fully prepared.

However, the antidote to these threats also lies in AI. By leveraging AI for pattern detection, we can scrutinise whether certain behaviours or device usage aligns with a customer’s typical patterns, thereby aiding in threat mitigation. Moreover, in our security operations, AI is proving invaluable in understanding our infrastructure and autonomously suggesting areas of vulnerability. Instead of inundating us with exhaustive reports containing hundreds of points, these AI-driven analyses distil complex issues into actionable insights, pinpointing the most critical vulnerabilities that demand immediate attention.

This targeted approach allows us to prioritise effectively in a realm where there’s always an extensive laundry list of security tasks. AI enables us to discern high-impact vulnerabilities that previously might have escaped notice without human intervention. As a result, we’re better equipped to address security gaps swiftly and strategically, leveraging AI’s accumulated knowledge from diverse scenarios to identify and safeguard against the most exploitable vulnerabilities. This integration of AI into our security framework represents an exciting frontier, promising enhanced protection in an ever-evolving digital landscape.

Are you currently encountering any technological challenges or focusing on specific areas for improvement in technology?

Our company operates entirely digitally, from loan origination to management and repayment, all processes are conducted digitally. We’ve allocated a dedicated engineering team of 200 members to handle these critical areas. Whether we’re processing 10,000 crores of loans a month or 50,000 crores, the size of this team remains constant. However, our challenges lie elsewhere, particularly in functions like customer support and debt recovery, where human intervention has been the norm.

In customer support, we encounter a myriad of issues spanning various categories, each demanding personalised attention and resolution. Similarly, in debt recovery, understanding the underlying reasons behind a customer’s inability to pay is crucial.

As we anticipate scaling our loan volume to 10,000 crores per month over the next two years, we’re confronted with the prospect of further team expansions. This is where generative AI comes into play at an opportune moment. Generative AI promises to revolutionise our approach by enabling effective verbal communication with customers, understanding their concerns, and offering tailored solutions based on their lending history—all tasks previously reliant on human intervention.

Our objective is to seamlessly integrate generative AI’s capabilities into our workflows to address the challenges of collections and customer support effectively. However, it’s essential to maintain realistic expectations. Generative AI isn’t a panacea; its implementation requires substantial investment, including integration, training, and data customisation. Immediate results shouldn’t be anticipated; rather, we’re prepared for a gradual process of implementation and refinement as we embark on this journey.

How do you plan to address the fact that 60% of your current customers are vendors or artisans, who may not predominantly speak English?

That’s the biggest challenge. That’s been one of the biggest, I think, impediments to adoption of LLMs in India. However, there’s a silver lining to this situation. The teams working on foundational models in India, such as AI for Bharat based out of IIT Madras, have been making significant strides. Additionally, AWS has made notable contributions, with its transcription product built on generative AI supporting a wide range of Indian languages. For instance, our collaboration with AWS’s post-call analytics product has demonstrated effective usage of these models across various languages.

Emerging players are also making valuable contributions to the ecosystem, which is incredibly promising. Leveraging AWS for transcription has been particularly exciting for us, as it covers a substantial portion of our call transcription needs, approximately 95 percent. However, it’s essential to note that even with successful AI implementation, human intervention remains necessary.

A controlled approach ensures that our human teams handle only the most complex issues, while AI addresses the majority of routine tasks, solving approximately 95 percent of our challenges effectively.

In the next six months to one year, what are some of the new digital or technology initiatives planned?

We’re rolling out a suite of intelligent collection analytics tools. As an organisation, we’ve established clear guidelines for how our agents interact with customers during collection calls, aiming to elevate the standard of engagement. Historically, collections has been perceived as a rough-and-tumble business, but we’re ushering in a new era of sophistication for all our customer interactions. Despite our best efforts, there may still be a few rogue individuals among our 6000-strong team. For these cases, we’ve devised a system where any violation of our code of conduct during a call triggers an immediate termination of the call, followed by coaching and guidance for the agent to learn from their mistakes. This product will leverage AWS’s post-call analytics stack to provide real-time feedback to agents during their calls, ensuring compliance and adherence to protocols.

We’re on schedule to launch this initiative in July, having completed the initial proof-of-concepts. This endeavour also serves as a prime example of how we’re harnessing the power of generative AI. Transcription poses a unique challenge, particularly given the diverse linguistic landscape of our call centres. With a mix of Hindi and English, often spoken rapidly and colloquially, traditional transcription methods fall short. That’s why we’ve developed a specialised model that recognises the nuances of this hybrid language, treating it as a distinct dialect rather than trying to fit it into predefined categories. It’s a complex endeavour, but one we’re fully committed to, knowing it will greatly enhance our operational efficiency and customer experience.

digital transformation journeyGenerative AIKisshtRING
Comments (0)
Add Comment