Likely to raise capital by end-FY27, says Sudipta Roy, MD & CEO, L&T Finance

Added: 2026-05-22

[00:00:03]Welcome to Money Control. We're joined by Sudepto Roy, MD and CEO of L&T Finance. Sudepub joined the company in 2024. He took over as a CEO that year.

[00:00:13]Um, and since then he's brought about a lot of changes. Two things are very visible at LNT Finance as we speak. One, it's a highly utilized NBFC with a near 100% retail ratio. The second thing and this might sound a little bit of sci science fiction a word that I very often keep using even when I speak to Sudipto when um uh um at times um words like Nostradamus etc. But these have brought in a lot of changes to the way the company is now underwriting loans and the company is now monitoring loans.

[00:00:49]Welcome Sudipto to this interview. I'll start right away with this. Um the most interesting takeaway if I were to look at your Q4 numbers and also for the full year of 2026 um it's the fact that the gap between 100% site loss covered two-wheeler book of L&T Finance and the rest of the industry is very significantly high right help us understand two three very salient features of Cyclops how it has helped you in underwriting and where you see this model take you further.

[00:01:28]>> Thanks Samsini and thanks for the opportunity to speak to money control.

[00:01:31]Uh I think uh you have asked a very pertinent question and uh we sort of uh put the data out on our cyclops performance actually granular data out for the first time since we had launched the program. Now as you would recall this is something which was launched in our two wheeler business in and around June 2024.

[00:01:53]So we have traversed close to 20 months period in this particular uh machine.

[00:01:58]And this machine for those of you who or those of your viewers who are coming in new uh I just sort of recap a little bit. This is a machine which uses uh deep learning algorithms to underwrite customers and two wheeler generally is a very aggressive credit product with a high proportion of new to credit.

[00:02:16]Typically what happens is that you know underwriting new to credit is really the holy grail of of underwriting signs primarily because the customer does not have much of tracks. The customer must not have done much of behavior reflecting in the bureau. So how do you underwrite those customers with confidence and also uh you know deliver the results within a short time frame short time frame while keeping your book quality uh consistent. So, Cyclops addresses this challenge by looking at a couple of things and what we call three-dimensional underwriting.

[00:02:49]We obviously look at the bureau. Now, the customer if it's a new might have a bureau track or might not have a bureau track. We look at the customer's banking details that we get through the account aggregator framework. And last but not the least, we look at a host of alternate data. For example, you know, where the customer stays, you know, what has been the customer's density of uh sort of residence at a particular place.

[00:03:10]you know for example the customers thickness of their payment tracks now we work with a couple of fintexs where actually we do not get any PII data we do not get any individual transaction data but what we do is that we get a flag you know whether this customer is UPI active or not whether this customer is UPI active extensively visa v partially UPI active right the also what we have done is that we have divided India into very small grids uh you know 100 m by 100 m grids and we're able to predict loss rates given historical bureau data we able to predict loss rates in that particular grid. So that also goes into this particular sort of sort of machine. So when a customer comes in we underwrite them across these dimensions about in cyclops now about 18 scorecards run parallelly and it gives underwriting output which is of an order of magnitude better and that is the reason because we underwrite the customer across many parameters and not across only the bureau parameter or only the banking parameter. The quality of underwriting that we get is much much higher and these algorithms are calibrated very very quickly. You know every 3 to four months there are algorithms undergo a calibration depending upon the quality of customers that we are bringing in. That has resulted to our sort of uh you know 30 plus numbers at sort of 10 months on book of a book size which is pretty large about 3,000 cr book which is around you know roughly a little higher than 2% while the industry is a little higher than 7%. So the gap between us and the industry is significant and obviously the causal >> this is the causal effect of cyclops where we are able to you know sort of cut between good and bad trade like a fine scalpage using the using cyclops >> that's true I'll stick to the point on uh technology um as we speak so dipto AIdriven collection uh if we look at go back and look at numbers once again has really helped keep a check on costs u and it has also helped improve efficiency.

[00:05:19]Interestingly I found in your uh investor deck that right now you're using AI based collection including voice collection in 10 languages. Do you plan to expand it more because you're definitely present in more than 10 states. So how do you plan to expand this? Allied to that every expansion does come with a certain cost. how how would that get absorbed and uh as a sub question to this if I can ask um AI increasingly there's a view that unless one has the scale of economics to play the cost of adoption of AI and the technology that you're talking about um the beautiful words like cyclops Nostradamus etc they are quite heavy on cost so what's your plan to continue keeping costs in check and expanding technology in the manner that you bought in.

[00:06:11]No, I think this is a very amazing question and the fact is that this is what the conundrum of AI adoption in most organizations is and organizations that are in infancy of AI adoption really do not understand what what sort of vehicle they are getting into till the cost starts hitting right in fact you know just uh before this call I was sitting with my uh chief digital officer where he was saying that you know our engineers are actually uh sort of bursting through their token allocation by the 10th of the months right so I'll come to that a little bit later but uh to answer your question on the first on the languages right right now we are in about 10 to 11 languages depending upon the process uh but you know adding new languages are uh quite easy I would say uh you know the fact is that the sort of the sort of the language model provider that you are working with you know you can patch on their languages as and when their language skills are deployed into their LMS. So as of now we are working with three major uh providers for our voicing AI platform. Uh one is obviously voicing AI which is LTM mind free uh sort of product. Uh then we are working with we have started working with server as well as we are working with 11 labs.

[00:07:30]So these are the sort of the main three sort of sort of parallel horses that we are testing right mind you it is not a one-stop or a one uh you know one attempt journey it took us almost 7 to eight tries to optimize the entire voicing platform because it is not easy you have to make sure that uh the voice latency is not much the voice is natural it has to able to you know the the AI tool or the AI bot has to be able to carry on a conversation that the human on the other side thinks that it is he or she is talking to a human. So that's the first thing. The optimization took us about four five months and as you see in our analyst results we have given out the results almost 80% plus of our PDM calling uh as well as our early bucket calling especially the selfare calling is now done by Ibots. It has given us good efficiency in terms of operating cost. It has given us good efficiency. I would say overall you know in the last three to four quarters I maybe you know maybe not three to four quarters last six months I would have saved about you know anywhere between 50 to 70 crores of cost in terms of uh outbound telecoming cost you know given the intens you are able to pick up intensity without >> on monthly basis >> no this would be across about five six months you know our average cost per AI call is right now about rupees 10 you can say you know bul rupees 10 right uh now uh on your question on the cost on the AI journey right now. One of the things which organizations will realize very soon is that AI is not cheap. Now there is one hand that common perception is that AI will lead to a lot of uh people efficiency and you know you will not need to hire so much of people which will lead to automatically cost efficiency but uh it is not such a onetoone relation. In fact, what happens is that yes, AI will help you to do the job without people. But the question is that the expenses that you add to is basically the cost of compute that you add to right and the entire the cost of development, maintenance, licensing, all that framework cost that you drop into and cost of compute actually can be significant, >> right? Especially if you're working on cloud uh suddenly you realize that your cost of comput has shot. I'll give you an example. You know, between in our AI adoption journey and which we started initially with Cyclops and Nostradamus, we saw our cost of compute go up by 50 to 60% every year, >> right? So then we realized that if the processes start producing more and more AI AI AI subutines as well as AI AI front ends, you know, you will have a cost of compute issue. And that is why uh about about four to five months back we took the decision saying that let us try to build our own private cloud.

[00:10:29]>> Right now probably we are uh among the very few organizations in India who have started the sort of job of building our own private cloud where we said we will deploy the development uh sort of compute on our private cloud.

[00:10:45]That means you know we will unplug ourselves from the hyperscalers and we will plug ourselves onto our private cloud. So the compute will come from our private cloud and when we do the costbenefit analysis the cost of compute on our private cloud is exactly one/ird of that what is available on the hyperscalers >> right so that means that you can actually run your AI development processes and AI process especially when your developers are working on you know building solutions using AI tools. If I use my private cloud then I'm able to deliver that at a much lower cost than I were to sort of incur if I were to use a you know commonl commercial hyperscaler environment. So we are going through the baby steps of this. We know what is the sort of cost efficiency that come out of it. But you know running a private cloud also is a you know within in-house is also a big task because you have to have 99.99% you know availability at all times. So that is why we will first migrate our tier three and tier 2 applications. tier one applications will continue to ride on on the commercial hyperscalers. However, over a period of you know 18 months to 24 months, we have enough experience, we have enough reliability built on our private cloud.

[00:11:52]We will not shy away from switching some of our tier one applications onto the private cloud as well. So the only way to keep cost of compute down in the AI world is also to build some compute capacity of its own. And most of the organizations which work this journey will realize this sooner or later.

[00:12:12]I have to tell you this uh Sudto you're not you're sounding less of a lender and more of a you know technology provider or a technology builder. Uh so should we expect that your private cloud is the next project that you're working on after uh seeing success with Cyclops and Nostradamus? Is is is is my understanding right or is there anything else on the product interface that you're working on apart from private cloud as well?

[00:12:40]>> Yeah, on your first comment, you know, I'll just spend about 2 minutes on it because it's a favorite topic of mine. I do believe that the technology content of financial services will continue to grow with every passing year to the extent that it is my estimate that for senior management in financial services firm technology will become 40 to 50% of their job content exactly 3 to four years from now right and frankly uh many organizations are not prepared for this leap in the country uh so uh in fact what we are doing is that this year we are embarking on an organization wide exercise or what I called massive tech DNA upgrade. It is not about upgrading the tech DNA of our technology teams or pim them with the latest tools right. it is across the organization right from the guy in the field right the you know the frontline sales guy and trying to teach them how to use AI or AI assisted AI tools that either might be built inhouse or might be available on any of the chat platforms uh which which are there or any of the assisted platforms which are there so that their job becomes much much more easier. So that is what we are doing because we sincerely believe that the tech content of a financial services manager's job will continuously increase exponentially over the next couple of years. Those who do not know or do not or will not make this change will find themselves obsolete very very soon right and obviously that will organizations who work this step will also become more competitive right so that will create new leaders in the financial services space >> the second thing is that the second part of your question as to what we are working on uh private cloud obviously is a is a large uh object of our interest right now but it's an infrastructure problem right so you know which I call you know uh and obviously cloud and you know data center etc. As you know a massive infrastructure has is is being built and as you know L&P as a group has been focusing a large on largely on building the data center infrastructure in the country. So we are fortunate in that saying that within our parent we have this infrastructure available you know to deploy our private cloud facilities right. So we are working with the group especially infrastructure of the group especially in terms of their data center infra to build our private cloud sort of infra. So that is in terms of hard infra build which is basically what will juice our systems you know if I call at a much lower cost space than if I were to use commercial hyperscalers. But this year we have taken a massive if you look at the technology architecture that we had declared for LNT finance in the day of raise in in in November 2024 was that is a three-part exercise or a three-part modular build. The first build was that the customer intelligence the second build was credit intelligence. The third build was portfolio intelligence and the fourth build was service intelligence.

[00:15:32]So the customer intelligence as well as the credit intelligence part is basically Cyclops which was built.

[00:15:37]Nostrade is portfolio intelligence it has been built this year. We are building the service intelligence block >> right. So the service intelligence block is being built this year. We have just have had the blueprint of the service intelligence block and the service intelligence block contain contains two parts right. It sits next to Nostradamus but it contains cross cell service and collections in a combined >> because we do believe that once the customer comes in on board uh you know the handholding the customer through the various touch points that we have is also important and also that should be done using technology so that we're able to leverage the various customer insights much much better. So that build has started. There's a new software tool which is being built internally which will launch on the day of raise of race 25 which will be on the 15th of December. We will raise it. We'll launch it to the market.

[00:16:28]>> So the date is also set right. So we exactly about uh you know 8 9 months to build it right we have already started architecting it >> and that is one major thing. The other thing is that we are going to launch our payments business this year. you know we uh announced to the markets that as an organization we are starting up we will be launching our payments business but we will take a slightly different take on payments uh we will look at payments again an AI enabled payments and we'll try to leapfrog the sort of uh from the current payment scenario to what we are trying to imagine as to how AI enabled payments would be and that is what we will try to leap frog so that is also one of our major projects this year so hopefully in the intervening period between September to December you will see launches of some of this uh new initiatives and that is what's also keeping us our technology team and probably some part of my time is also going in.

[00:17:22]>> Absolutely. Um let's let's start delving into a bit on your uh business and your numbers. Um, again an interesting uh uh note in your uh uh results deck that that caught my attention is that L&T Finance now onboards customers only with zero DPD. Doesn't this at some level shrink the pool of um attractable customers for L&T Finance? Is this a decision you took conscience consciously and if so why? If you could help us with the logic.

[00:17:56]See the logic HSN is that uh is it a permanent thought process? Let me tell you it is not a permanent thought process right because I do believe that there are some customers who might have missed payment here and there right uh uh are on the long run can turn out to be good customers right >> however you know the the sort of the framework that we came from and you know we are an NBFC who are in who are in relatively riskier segments or segments what I call sensitive segments right you micro finance a sensitive segment you have you know >> economic overhang you have climate IC overhang, you have political overhang, all sorts of overhang of that business, right? You have two wheeler again bottom of the pyramid. No, some part of it is bottom of the pyramid. Again, a sensitive business. Tractors, farmers, again the same sort of overhangs as the micr finance business, >> right? Personal loans, personal loans tend to be cyclical, right? So, you are in a business which is which which has got lot of sort of I would say soft underbellies, right? So uh you know uh to borrow the sort of euphemism as that when a when a when you are when you are building your physique and when you're exercising you know trying to get much more fitter you have to be very very disciplined about it otherwise you will not get your uh end and one thing which I believe is that every business every lending business has to be have something which I call lowrisk balance let me tell you what do I mean by lowrisk balance it has to be a critical mask of low-risk customers which keeps the ship stable like balance keeps the ship stable right similarly you know the lowrisk balance keeps the lending organization stable and while you are building this lowrisk balance there cannot be any shortcuts so our focus on no zero DPD customers is to make sure that the speed at which we build this lowrisk ballast is quite accelerated because you know if the customer has been quite pristine you know the lowrisisk book builds quite fast and once Once you have enough of lowrisk build book, you have the wherewithal to start doing experiments, right? In a little bit of risky alco. So that's what we will start doing. But the fact is that you know we were doing lots of stuff. We were doing you know hitherto running into uncharted territories. We employed cyclops now you know almost almost two years later we know that cyclops is a success but when we implemented it we did not know it was a success right? It was a you know sort of a leap of faith that we took >> of course. Now as it gives us more confidence and as this lowrisk grows the confidence of experimenting on the margin will also increase right and that is why we have now started working on the next version of backlogs where we will start an attempt at underwriting this sort of customers because I do believe that among this sort of customers who might have exhibited some sort of delinquency here and there some sort of noisy patterns here and there but overall on the long run they might be good credit at a particular device right now. Can a human being differentiate? No. Uh can the machine differentiate looking at multiple sort of nodes of data? Yes, the machine can differentiate. So that build is will be also our next project. So obviously over a period of time you will see us experimenting on the margins on that.

[00:21:13]But the focus on prime good quality customer will always remain. So that is will be our middle path and we will always follow our middle path.

[00:21:20]>> How are you looking at gold loans at this moment? So, you had an acquisition in the form of uh Paul Merchant last year and ever since that happened escalatedly in the last um 12 months or so. We've seen massive volatilities in the gold prices. Given this background, how are you looking at gold loan portfolio today? Would you be a little more cautious and also possibly guide that the value growth that we saw last year could possibly be well behind us?

[00:21:52]I think you know last part of the question I think the value growth will be tempered a bit this year because primarily the new RBI guidelines have kicked in >> uh the market is taking some time to adjust to the guidelines and and obviously uh in April for example when we adopted those guidelines and push on from a systemic basis all those guidelines were adopted we saw some hidden volumes right so you're absolutely right in that you know some of the some of the heady growth across the industry that we saw for the last 24 months you know also fuel to a large extent to the heady growth of gold prices also we'll get tempered a bit >> right uh but for us you know again you know as I said you know we are middle of the road right so you know we just don't want to chew off more than we can buy and for us right now the focus on the gold loans business is to build our distribution >> when we took Paul merchants we took 130 branches we took roughly about 1100 of balance sheet guys now we have roughly about 340 branches right so we've added almost one branch a day in into it >> right and uh we are the the branches are scaling up quite well. Uh so one of the things as what we have done in our gold loans business is that we have heavily digitized our gold loan business in our gold loans branches you will not find a single piece of it >> right. So from the beginning and the fact is that the gold loans platform we were using an uh you know we have built our own platform >> we have built our own platform on gold loans especially origination platform completely inhouse built right with a with a lot of customized features we have built >> and we are working on certain tools wherein you know one of the thesis of our gold loan business or our getting into the gold loan business is that many of our customers especially of rural customers have borrowed a lot of gold loans between our tractor customers or two customers customers and our micro finance customers between them they borrowed anywhere I think between 40 to 45,000 crores worth of gold loans and they are our customers so we thought that maybe we can take a share of that pie you know try balance transfers try crossell you know and try to sell that gold loans so our focus on the gold loan business is you know part cross-ell to our own customers as well as part new customer acquisition so in a cross-ell scenario we have known those customers for a period of time so we are more comfortable in taking our calls on those customers customers and already that crossell engine has started delivering some results especially so you know some of our catchments are delivering funneling customers into our gold loan branches where we are able to disperse though it is still in its infancy and still you know has time to gain to maturity and obviously we also have a strong external work force salesforce which is building new to organization customers as well. This year we plan to deploy anywhere between 400 to 450 new gold branches for the full year. Again almost one to one and a half gold branches every but again you know we our trade policies are tighter. Our we do not deviate from any LTV norms. Our checks are much more stringent. Our audit processes are extremely exhaustive and we are deploying a lot of tech on ground especially in terms of good quality audit just to make sure that we don't get blindsided by any sort of unto incidents right so in a way we are building this organiz this business faster but we are building it with a lot of caution as well >> uh are you happy with the current split of secure and unsecured loans would you want your book to align more towards unsecured loans and in that plan would seeking an NHP license uh be a part of your uh um you know proposals as well because increasing the amount of mortgages could help for you in terms of uh uh the secured portfolio and also on the liability side of the balance sheet >> again you know it's about the secured and unsecured balance is about also maintaining our ins and fees and balancing them right so uh >> but the unsecured business also brings in higher risk and the fact is that we are a rated identity we are now internationally rated so you know there is certain commitments on that side as well >> so having said that the ideal ratio I think is 60/40 for us 60% secured 40% unsecured we will try to maintain within that right >> we are marginally here and there you know depending upon the quarter >> but it's really and one of the reason of acquiring the gold loan business was that we wanted a secured high yield portfolio in fact what we have started doing in our portfolio we have launched something called partially collateralized business That means you know uh we do not take a full collateral but we take a partial collateral which is also easier for the uh you know the customer and also builds a little bit of security cover as well you know I call maybe that that portion >> goes into the territory of quasi unsecured not fully unsecured right so that is what we are trying to do balance it >> obviously uh so but 6040 is where we are uh and on mortgage yes see We are not in the affordable mortgage space. You know, if you look at, you know, uh our mortgage focus, it has been prime and near prime mortgage. That has always been our focus with a higher share of loan against property that we have been over the last couple of quarters primarily because of some of the rate inefficiencies.

[00:26:58]>> I uh in HB license, we had a home loan uh entity. Uh we gave up the license. So uh as of now there is no thought process within the organization to go and ask for a license once again because one of the reason we simplified the structure of the organization into one organization was to do multiplicity of lending entities and and make sure that you know from an administration perspective we are a much more simpler uh sort of organization. Uh so as of now there is no such thought >> uh but we we are working on you know certain mortgage ticket sizes which which gives us a little bit of higher yields right especially between ticket sizes of maybe 25 lakhs to 50 lakhs that is where the focus is. It is not in the not exactly in the affordable housing segment what we something which we call economical housing right that's a terminology we use internally but it is again in in in the normal mortgage space right so that is what our focus is but frankly balancing of secured and unsecured high yield is primarily will be through a gold loans vehicle and through a microlab vehicle to understand that the microlab business is fully secured but comes as a very high yield very small tickets but that business is scaling very very well >> right so that is where also we are we are focusing Mhm.

[00:28:14]>> Mhm.

[00:28:14]>> So it's a it's an exercise that will continue for the next couple of years but 6040 is our goal.

[00:28:22]>> Excellent. Last question to you sir. Uh you uh last time when we spoke you said till FY28 you do not envisage a capital race. I know we're at least two fiscals away from that or maybe a fiscal away from that. Um but given that there's been a lot of interest from foreign players in India, would you be open to an equity partnership with a foreign player as and when you start thinking about your capital race?

[00:28:50]>> See the speed at which you're going a capital race might be uh in latter part of FI27 early FI28 as well. So you know so uh so so the preparation might start from latter part of FI27 right and probably with the capital raise finally culminating in FIK >> right uh but uh >> see we had Bane as a strategic investor right in LNT finance for quite a long time >> right and uh so we we uh Bane exited and obviously now we do not have what I call uh anchor market tenant investor we don't have someone like that. Uh >> the group is the group is quite uh uh uh you know has a very high holding in L finance and the group probably will continue that holding you know as we go into the capital race. So the the sort of the elbow room or wiggle space for having someone large come in uh you know uh as as we stand right now might be limited but who knows right this is something you know given valuations given uh sort of interest you know all these things can develop over a period of time so I'm saying I'm not saying no I'm not saying yes you know we will see as it comes right it is slightly premature to talk about it and probably we will sort of evaluate these options.

[00:30:14]I'm sure with some options will come to us and we'll evaluate these options uh when we come closer to the capital.

[00:30:24]>> Thank you so much. We leave it at that.

[00:30:26]It was pleasure talking to you Mr. Roy.

[00:30:28]Uh look forward to another interesting conversation very soon. Um maybe when we're ready to talk a little more on the payments and uh your uh private cloud business. Uh thank you so much.

[00:30:40]>> Thank you. Thank you for the opportunity.