Everyone bet on PhonePe and CRED. Bajaj Finance got there first.

Namaste Rajji, Rohit Bhollaram, Ghani Bank, aapke liye ek khas Vikal Pnekla hai, ki aap 2 minut mein sunna chaayenge? Hanjanki Bhollaram Aare waa, dhe kiya Rajji, aapka hama de saat bhood acah relationship hai na? aapke financiyal history bhi excellent hai aapke preafroved personal loan aapar kar hai hai aapkha 10.99% hai je interest rate hai aapkha bhi bhi na kise kualitral ke? aapkha hai aapkha? aur abhi to interest me la bhan hai nahi meragul rekuwamand hai nahi vese hai Bilkul saam asakta ho Rajji Aakchali, ye loan urgent need ke liye nahi hai ye to future planning ke liye jaise ke agar aapko kabhi ho? I don't know if you guessed but that's really not a human that's a voice AI bot that's talking to a person and trying to convince him to get a personal loan but why are we talking about this today? Well, it's because we're going to talk not about voice AI but about a company. This is company called Bajaj Finance which has stopped calling itself a lender that uses AI. It now calls itself a quote "fin AI company" which in my mind is like a category of one. I can't think of any other company that calls it a fin AI company. And this isn't something that it sort of created for itself last quarter when AI was the thing that all the analysts everywhere are asking companies about. This is the label that Bajaj Finance put on its numbers on its table on its presentations in December 2024 which is about 18 months ago at an investor day. Now this was a five-year plan for Bajaj Finance which was to become a fin AI company by FY29 and the framing was that Bajaj is going to go from an AI data first to AI first. Now I wanted to stop if you're thinking about this and if you're listening to this and figure out how weird and unusual that is because companies say all the time we are investing in AI and a company that is actually renaming the category that it belongs to and committing to it not for like a couple of quarters but for a five-year plan is actually quite different. And the best part about this gap between announcing a transformation which we have seen many companies do and being in one is that in the most recent call Bajaj Finance said we are not planning we are in the process of implementing. So they finished the whole data architecture for the entire thing last quarter. Three years ago if you ask this question who's going to win Indian finances AI era? There was a very obvious answer and that obvious answer was the digital native company. These are like the app first, you know, cloud-borne. They don't have any call centers. These are fintechs like phone pay and cred and all of these other companies. So the lenders were essentially supposed to be playing catch up to these new age tech companies. But what has really happened now is it looks like the company that is way ahead of the pack in terms of AI and finance are not these fintechs but the company that started financing two vealers and still runs one of the largest outbound calling operations in Indian lending. And the thing that looked like Bajaj's baggage which is a set of people that are on phones a relationship that run through a call and a dealer and it's no app that is exactly the thing that led Bajaj Finance get ahead of everybody else. And the thing that everybody thought was their cost center that nobody wanted to have became the motor. Now I'm going to state some numbers here. Just in a last quarter, Bajaj Finance said they had 31 million customer voice interactions that have been converted into data. They also generated 100,000 new loan offers purely from analyzing that voice. They have 27 autonomous AI agents live today with 118 more planned. And in last Diwali, Bajaj processed 600,000 loans in a single day. And they said in the call that before AI that number their ceiling was somewhere close to like 100,000. Next Diwali, Bajaj Finance is aiming to get that number to a million. So how did this happen? And I have a theory of how this happened. And my theory is that suddenly a perfect storm of three things came together and the first thing was data which is Bajaj already had the richest data record of how Indians behave on the phone. Not on the app, on the phone, on what they do, what they ask for, how they interact, how they respond, what kind of persuasion work. That is stuff that Bajaj Finance already had sitting on them. Second, they also had the incentive because voice for Bajaj is not just a cost center. For many companies, voice becomes a cost center because they are customer support, etc. But for Bajaj, because they are built on top of voice, every call is a chance to sell. And because every call is a chance to sell, which means that a calling first DNA is what gives this company an unfair advantage. And finally, Bajaj has the leverage because voice AI isn't a technology that is locked in with open AI or anthropic or server. There are dozens of companies that are fighting for it. So, a company of Bajaj Finance size gets to look at multiple pilots, multiple companies and try to figure out what works for them and what does not work for them under what circumstances. So, these are the reasons why Bajaj Finance is really killing it. But there are also questions and the questions are, maybe Bajaj is not just ahead. Maybe it is only the one that's saying it's ahead because it's also the only listed lender that's putting out AI numbers on a slide. So, one of my colleagues at the Ken did the story where he said that maybe Bajaj Finance has a disclosure advantage and not just an operating advantage. So, even if they are ahead, voice data is not rare. HDFC, SBI, insurers, all of these companies also have access to voice data. So, what exactly is defensible? That's really what we are here to find out. [Music] And to find this out, I have two wonderful guests with me. My first guest is Sita Ramann, who's the deputy editor of the Ken, who's making a return to 2x2 and he's making a return just after releasing an intermission episode on Bajaj Finance's history. And he's here to answer what does Bajaj Finance past tell us about this future. Welcome, Sita. Hey, PJ, good to be back. I didn't expect to be talking about Bajaj Finance's second time, but here we go. Can you tell us how long did you talk about Bajaj Finance the first time? I think the entire discussion was about 5.5 hours over a day and a half and the final episode is roughly 4.5 hours. So, it's a 4.5 hour episode where you and Rohan, of course, who's the co-host of intermission, are discussing Bajaj Finance history and how they got to where they are today. Yeah, absolutely. And where can listeners of this podcast listen to it? Wherever they get their podcast, it's our first video podcast and it's our first free long form podcast and it's on YouTube. You don't just get to listen to us, you get to see us as well for 4.5 hours. That is a treat. Okay, I would recommend to listeners after this episode. Of course, please go and check out intermission. Intermission is the new video product from the Ken. They've done two episodes. The first episode was on Asian paints. The second episode is on Bajaj Finance and it just actually happened to be a coincidence. I was doing this Bajaj Finance episode because I saw some of their AI numbers and I was like, hmm, who knows a lot about Bajaj Finance that I can call right now? And well, here is Sita. Welcome Sita. We'll talk about all of this. My second guest is Vasitha Agarwal, who is the Chief Revenue Officer at Gnani.ai, which builds voice AI for enterprises. Vasitha was the ex-chief business officer at Inmobi Glance and was also on our panel for a life career event that we did last year at the Bengaluru International Centre where we had a discussion about the future of careers. Vasitha is of course just recently joined Gnani and Vasitha is here to tell us about this technology voice AI, what's really happening and what are the incentives and the financial cost and revenue side of this for all of these companies. Welcome Vasitha and tell us a little bit about Gnani AI. Hey, Hey, Hey, Hey, Hey, Hey, Good to be back. Thank you for having me. Yeah, so, you know, a very interesting topic today. I think voice AI has, you know, obviously it's one of those very interesting AI you know technologies which has taken off in a big way in India because India is a voice heavy market. So interestingly, what we are finding is that voice AI India is actually ahead of the curve compared to a lot of other countries. especially on the enterprise site and enterprise cycles are always long and hard. So, Gyanney has been in the voice AI space for about eight and a half, nine years now, really started with. Really? Eight and a half, nine years. For some reason, I always assumed it was a young, I mean, I say, "Eight and a half, nine years is young. We are nine years old, so we are young too." But I thought it was a much more recent company. No, no. So we've been in the space from the time of building speech on devices back with Samson. Then, you know, pivoted to NLP technology is building actually our first voice agents on NLP back in 2019. And then pivoted to, you know, Gyanney, in the last two and a half, three years. So we've been building in this space for a fairly long time. And that's why, you know, we sort of play across the voice AI stack, not just building the agents, but also the core models, the orchestration platform and so on. So we are looking forward to the conversation today. Can you tell us a little bit more about Gyanney? I think you are also one of the companies that have been selected along with Sarvam for the. Yes. go ahead, tell us. So we're part of the India AI mission. So we're one of the, you know, I think, handful of startups which have been, you know, been selected, which gives us, you know, sort of, you can say, almost the, you know, the privilege to build deep tech AI from India, which is meant for, you know, India, Indian languages, democratizing AI across India. And also, you know, hopefully taking this technology global. So, yeah, so it's interesting times for sure. And yeah, we were able to showcase some of our sort of breakthrough models on voice to voice. At the India AI summit with, you know, PM Modi, share the stage with, you know, all the sort of world AI leaders at the summit. So it's been an interesting journey for sure for the company. Nice. I'm going to start with the first point, which is, I want to start this discussion by because we started with Bajaj Finance and where they are today. And all of these numbers are like quite interesting. And to be very honest with you, when they announced this in like 2024, I mean, I saw this, you know, announcement that, oh, we are going to do this digital transformation, become AI company. I said, sure, I mean, everybody is talking about this. I mean, you are a publicly listed company. I'm sure that there are incentives for you to talk about these things. Good for you. But I am quite surprised by the aggressive disclosures and the numbers. I'm pleasantly surprised. It's like very, very interesting. So like all stories, we have to start at the beginning. And the beginning is not in 2024. The beginning is way before that. So I think the first thing I want to talk about is the data itself because that's really how I began. I began by saying that maybe Bajaj Finance's big asset that they just came to a realization is we have all of this data sitting with us about all these voice conversation that we've had with customers over the last God knows how many decades. And we are going to basically monetize and use this asset in order to build something for the future. So it's like a corpus, which is like collections, cross cell, EMI's. And I think that so the history of it matters. And Sita bring us up to speed. Can you tell us a little bit about how did Bajaj Finance really like get all of these pieces in order? Like how did it begin? Well, it began in 1987 when Bajaj Finance was set up as Bajaj Auto Finance. It was a unit of Bajaj Auto. At that point, incomes were low. Bajaj had to find a way to sell more scooters. So they said why not set up our own financing arm. And cut to 2007-08 when there is a restructuring within the group. There is a succession underway within the Bajaj Group. So Bajaj Auto Finance is spun out from Bajaj Auto. And Sanjeev Bajaj, who is the youngest son of Rahul Bajaj, is given Bajaj Auto Finance. And by then they had started giving out a few consumer durable loans. I think in the early, the late 90s, they had started financing consumer durable. Along with a lot of foreign banks like City. And there were a few Indian lenders as well. Then by 2007, it was still pretty much an auto lender. I think 85% of its loan book was still two dealers. And then slowly they started. And this is around the time when Kroema and reliance digital were set up. So consumer durable were taking off in a big way. So it sort of, you know, and no cost TMI had existed for a while. But Bajaj Finance is the one that made it really, really popular. So if you were to ask someone who came up with no cost TMI in India, they would probably say Bajaj Finance, which is not true at all. So it had been around for at least 10 years. And it just used all these retail outlets of Kroema, Vijay, sales, reliance, digital, someone put it. These became the branches that Bajaj Finance didn't have. That's where they acquired customers. And obviously this came with a lot of data. And they had a lot of constraints at that point because I think, Sybil had launched credit scores in the early 2000s, but not too many lenders were using that. So the point was HDFC or ICICI would just see if you were a depositor with them and then decide whether or not to give you a loan. Bajaj Finance though it was a deposit taking NBFC was not very big. So it had to find other ways. So it started using credit scores. So roughly, I think into 2010, roughly four out of 10 customers who walked into any of these stores for a, you know, to buy a mobile phone already had a credit score. And where they didn't have a credit score, they sort of made up for it by looking for proxies. So they would go and ask Kroema, hey, this guy wants a phone and he wants me to finance it. Have you, has he bought anything from you before and have you delivered anything to his home, which means you have his address. You know where he lives. So if you have, then I'm willing to finance a phone for it. If you haven't, I'm not going to give him money for a phone, but I can give him money for a TV, which you're going to deliver, then I have proof of address, then I'll give him a phone the next time he walks into your store. Oh, very interesting. So the places where they were actually buying in some senses, they made them quasi sales agents, collection, and it's all together. Absolutely. And they said, they didn't want to spend anything on acquiring these customers, right? And then and it became within a few years, they got so good at it, they would go to Kroema and we just say, hey, there are these 70,000 people who bought a refrigerator roughly say seven years ago. And the average ownership of a refrigerator in India is five, six years. So it's quite likely that all these guys are going to be looking to replace the refrigerator in the next six months. Do you want to roll out offers for them? Okay. Anonymous data, 60,000 customers. Now let's figure out a plan and get them to replace the refrigerator right now, right? And these retailers were more than happy. I mean, they had these customers data, but they didn't know what to do with that. Bajaj finance started mining all these, you know, sort of millions of customers for different data points and then they would go and sell them more and more. And one of the most fascinating things which Sanju Bajaj told me was, you know, a customer who had already bought something from them the next time they wanted to buy something, they were why you asking me for all my details again. You have my credit history. I paid all my EMI is on time. I've already bought something from you. So why do I have to go through this whole thing again? So they were like, yeah, that's right. So they wanted to launch a credit card, but being an NBFC, they were not allowed to. So they said, let's launch something which is pretty much a credit card, but you can't call it a credit card. So they call it smart EMI card, right? So this card came with pre-approved financing for 40 or 50 consumer durable products. So you would just take this card somewhere to to which are sales or chroma and then you swipe the card and by then they'd actually work with pine and other POS makers to, you know, if you they said if you're processing city and sorry visa and master card, why can't you process our cards as well, right? So you swipe it would tell you whether or not you were eligible and that's it. So it functioned pretty much like a credit card. But it, you know, being a non-bank, it was able to offer it. So that's how it was able to get customers to consider financing without even thinking about it. And then it moved from that to personal finance, to personal loans, to home loans, all kinds of loans, right? Now, I mean, you've seen the investor presentation, right? I think they offer some 50 products, 50, they call it 50 business units. And one of the things you talked about how aggressively they're disclosed, one of the metrics they are obsessive about is products per customer. How many budget finance products does it does, you know, the average customer use? This could be anything from consumer durable loans to to their their app, their wallet, okay? So they want to keep, you know, increasing that number. I think it's roughly five or six per customer. So they really, really obsessively track that and they actually put out these long range targets. They have said by 2030, we want our overall customer to almost double from 2025. Yeah. And for a company to actually say so much, and you talked about disclosing, and they don't have these, don't seem to have any problems about setting these very high targets, not just internally, but just disclosing that to investors to say, we want to have 200 million customers, and we want to be able to cross sell to say, 150, 170 million of those. And that there is a pyramid in every presentation of this overall customers followed by cross sell customers. That is also something that they track very, very closely. Before we continue, I'll just quickly read out the numbers. You spoke about numbers here. The Q4 financial numbers very high level. They're consolidated. A-O-M, OK, I can't believe I've read it this. Is 500,000 crores, right? Which is up 22% year on year. It crossed five lacrosse of 500,000 crores for the very first time. The consolidated profit after tax is 5,400 crores, which is also up 22%. There are other numbers, et cetera, which are like you said, there are so many numbers that they're disclosing. I think the real interesting part in this exchange that you just had is, I always thought of Bajaj finance as the people who call, and they get a lot of voice stuff, and I'll bring in Vasotha right now. But I think what you're really telling me is voice is just one part of it. There are these so many other data points that they had underneath through all of these relationships and channel sales and all of these consumer products, and that sort of was the bedrock already. And voice just, I don't know, top of it. Vasotha just jumped in. I mean, specifically, what does voice do for a company like Bajaj finance? I mean, I don't know, don't talk about Bajaj finance specifically, but I'm saying that a viewer company like Bajaj finance has all of this stuff with it. What would voice data do on top of this? Yeah, so I think, if you look at banking and NBFCs, I mean, just the whole sector, right? I think it lends itself very, very well to a voice because if you look at India, right? India is one, it is a country of many, many languages, and a very large non-English speaking customer base, right? People are used to voice first, a phone first kind of interactions. For everything, we want to pick up the phone and call essentially, right? There's also just, I think, a massive, massive ecosystem of outbound calls, whether it's any kind of sales, support, service, right? So I think all of those aspects combined, right? Just add so much of scale to this whole voice AI ecosystem. And then the things in banking specifically, which lends itself very well to AIification, is banking has a lot of forms, approvals. Right? Along with the calls, right? You have a lot of paperwork, workflows. So essentially, there's a lot of data and workflows. And some bit of prediction involved, either assessing risk when it comes to giving loans, right? Or, you know, repetitive workflows, repetitive reminders, if you have to call a customer for, let's say, reminding them on payments, right? So all of these things, right? The fact that India is a very voice, voice heavy market and banking as a segment has a lot of workflows which lend themselves well to automation. I think the combined sort of, you know, power of that is what needs to, you know, some of the maybe metrics are impact that you were talking about or, you know, Sita was talking about that this sector is almost like, you know, it's the number of use cases, the kind of customer touch points and the scale of it, right? Given India's population across languages makes it, makes it very, very AI ready for destruction. I have a question for both of you, right? Maybe you can just, maybe mostly for you or Sita. Suppose you take the example of a company, like, say, I mean, I say this only because like, I have a very consumer tech mindset to this, which is that I would argue that, okay, voice is all very nice, but you know what the richer data is actually in what customers do and not what they say. So in other words, why is it then that a company like, say, I mean, I'm taking as an example, like a phone pay or a PTM? Why do they not have a quote unquote richer data set because they basically have access to every single user on phone pay, every single user on PTM, what screen they go to, what do they see? How long do they stop? They pause. They come back. That data much, you know, more powerful and actionable from an AI standpoint. Then customers saying things like, oh, okay, I understand, but I'm not sure I'll get back to you. Why would this be richer? See, I think from a data point lens, it is about, I mean, of course, the data say on the app or different touch points at even say, a POS and so on or, you know, merchant data that some of these guys would be sitting on all of those touch points and interactions are extremely valuable, right? If they want to do any kind of AI disruption, not just voice, right? It could be coming up with, you know, predicting, you know, customer behavior. It could be coming up with, you know, very customized, personalized customer journeys on the app. So all of those things, obviously, you know, matter and have impact. I just think voice, the thing is that it is very hard to analyze that data at scale, right? Because if you go beyond just the simple, yes, no one interactions of a customer, right? I think the richness of that data over a period of time, right, can lead to very, very sort of, you know, different outcomes. So for example, simply knowing, right, when to contact a customer, when will your connector, it's an answer, it's be best, right? What is the customer disposition and behavior going to be at that time, right? Things like that, right? Knowing some of that over very high scale of data. So for example, we do anywhere between three to five crore calls a day at Ghani, right? Voice AI calls, right? So that gives us insight into simple things like, okay, you know, if it is a collections use case for a customer who's, you know, email is still not due, when is the right time to call to get the best answer rate, right? Visubir customer who's already maybe 30 days past due, how many times and at what times in the day should we call that customer to get the best resolution on that, you know, say payment. So things like that also build with volume of data and obviously seeing it over a period of time. But that's a platform view that, you know, I'm sharing enterprises obviously would have very, very rich data because like, you know, if you just run speech analytics as a product right on a few thousand customer call over a few like customer calls over a month, you can get extremely rich insights about, you know, what your customers are saying and that can be baked in the same way as we were talking about, you know, app based touch points, right? Because customers will share a lot of that feedback and it's not just about what they are saying, but how they are saying it as well, right? But you have, you can analyze those signals. Yeah, so now we're getting into, you know, things like, you know, how do you analyze, you know, disposition of the customer, the intelligence behind it. If a call center agent is talking to a customer, we can even do like, you know, a monitoring of things like risk and compliance. What are the guardrails of that conversation? What should be said, not should not be said, right? So all, all aspects of the conversation can actually be analyzed now. I have a slightly different take and that is to do with what, what all bazaar's finance sells. Okay. I talked about pre approval for consumer jorobos, you know, say from if you border refrigerator, now you have 40 other products for which you can get financing, right? You can extend that to personal loans, which are unsecured. Okay. You have a good credit history, okay, you're eligible for a lack. Okay. It doesn't matter what you wanted for. But as you move up the ladder, you get into secured loans, you're going to want to have a conversation, both you and the lender, say if it's a home loan and bazaar's finance has a mortgage lender, which is now listed bazaar's housing finance. And that is where every, every lender wants to have a huge mortgage book because it's secured and it's long term and the ticket sizes are huge. Say you've been a bazaar's finance customer for a while. You've taken all kinds of unsecured loans and now you want a home loan. And that's why you want to have a conversation for everything else. It's approved on the app. You go, you say, I want this, it's approved, it's done. But for home loans, you want to have a conversation because you know that there are several other large lenders that are that have made their name in housing loans. So you want to have a conversation about VPN period, you know, how much will you give me 75% or 85% of the value of the property. And that's where I think it becomes really, really important. The likes of phone pay and PTM, which obviously originate loans for lenders. You wouldn't really short for personal loans. I would by me go on PTM and look for the best deal possible. But when it comes to home loans, I would think of HDFC, I would think of SBI, I would think of Bajaj. Right. L.I. Say how exactly. So that is the difference in what the two are selling. I mean, FinTechs and Bajaj Finance and which is why and our story, I mean, the one that a former colleague Arun Dothi did on the annoyance that is Bajaj Finance is calling. Correct. There, I think the company mentioned that it accounts for a very, very small chunk of their overall loans. I mean, loans generated or originated through [BLANK_AUDIO] But the thing is they have so many loans, I mean like hundreds of thousands of loans, even that fraction is quite significant, which is why we see all these numbers right now. Right, I mean you've probably answered this to some action, but I will follow up question because you mentioned HDFC as well, right, because one of the things like I was thinking about, what is it about this company that got it here? How did it become like if you asked me who was the first Finnei company random you had asked me like two years back, we would all put bets, I may not have put a bet on this right. So the question that I have is sure we talked about data and of course even was with a you said look you know voice data is actually rich in its own way it gives you all of these other signals in addition to all of these signals that Bajajas. I think the final thing that I have been trying to grapple with is that okay if this data is the mode. I mean others also have such huge call archives right like HDFC has it SBI I would imagine SBI's call archives would be massive LIC housing finance Bajaj Alliance. So why is in this why is in this corpus in and of itself clearly it's not just that there is more to it or there is some nature to this corpus that works in Bajaj finances favor. I think again going back to the call center operation right we said even if it's a fraction they don't mind getting that customer on board okay and like someone said one of the you know one someone who worked with the chroma said they're always looking to up their approval rate. So at one point I believe total capital which is which was an in house finance here had an approval rate of 45 50% at chroma and Bajaj finance had an approval rate of 70%. Okay and they said we're not happy with this we're going to take it to 95 98% and they did that and they came to this guy and said we've run these checks on your customers or and we've pre-approved I don't know it was something like two lakhs or five lakhs for all these customers. Whereas in house finance here we're not willing to go beyond this because we trust we don't have anything but the credits go to go by. And so I think it's this sort of DNA of the organization where like we said on intermission always be selling right they're like okay now what we said to this guy he's got like five loans from us now can we these an HTFC customer now if he plans to wire second home why should it be with HTFC why can't it be with us. I think that's that's baked into how the company functions and that's what really has your absolutely right as we have more data than you know any lender in the country right correct and probably more than everyone else. Absolutely. Absolutely. Absolutely. And you have the positive right you have people who pack their money with you. Exactly. It's a bank. But I think they there is a certain sense of complacency it's that if I look at my loan book and home loans are 30% I don't have to sell a lot of home loans right because each home loan is one crore two crore three crore right and I sell and I'm done. Where is it you look at by judge finance one of the metrics again they disclose is. I think they say I think they just close this last I think in in the third quarter they said with 12.5% of loans booked an India but only 2.8% of retail a U.M. Okay so they very clear that we sell a lot of loans but right now because of our legacy as a consumer durable finance here these are all low value. So they mean that we have we can be 10% of retail 8 a.m. at some point right and this is I mean and this is why they obsessively you know sell higher value loans to their existing customers which is why I think the data that they have even if others have way more data they're not making a good use of it. So let me just bring you in by the way before I bring was with and let me just say because you said ABC always be closing recommendation of a movie if you haven't watched this movie called Glenn Gary Glenn Ross which is specifically about selling and it's specifically about leads and selling. I don't think to enterprise maybe it's to enterprise is just it's a phenomenal movie it's an old movie you'll probably not find it anywhere that easily but it's a great movie I recommended was with. So I asked you broadly speaking if you looked at say what makes if I said oh you have access hypothetically for data set a data set B data set C data set D data set E belonging to five different banks and BFCs etc so so I asked you to detach yourself from the story of baraj finance and all of this DNA and all which is all like see that story which I get. I mean if I asked you very objectively what makes certain data better than others not obviously not all voice data is equal or not yeah not all of this is equal now you know that all of these people have data clearly some data is better than other data. So basically I think if so if you're looking at enough you're talking about voice AI and the impact or the efficacy of it right I mean sure data that you know the enterprise bank ABCD is one part of it right so that's what feeds into your the LLM or what we now call the SLM's right the small language models so the data actually feeds in more into that right really but if you look at voice AI and what makes it work and creates. The impact right and why large enterprises are adopting it is because of two things it is either improving top line right it's a revenue growth lever or it is improving bottom line right it's reducing operational costs and that's I think fundamentally what voice AI has started to prove right and the data sets the quality of the data the scale of the data is one part of it but what I do want to highlight what makes voice AI effective overall. Is that actually if you see the stack has three components to it you have the you know the ASR or what detects the audio converts it to text right the text is then processed by the SLM or the LLM like the Gani for example uses SLM small language models that we have fine to to you know work in India latency conditions because if you use LLM's for large length for real world conversations customer conversations it's just the latency is too much right so we actually have SLM's which are tweaked for different industries different use cases and so on and then that text is converted back to speech through the TTS model right or so these are three components the ASR the SLM and the TTS which really come together to deliver the impact that voice AI you know is delivering what we're hearing about all of those things so so bias say this is because you know the data set is important because yes it is you know used for you know fine tuning the accuracy of the data you know multilingual data sets all of those things are important we at least like we've trained our data on almost you know and our 14 million hours of data available from you know different different to kind of conditions but if these three components don't come together well then you cannot replicate a real life customer conversation and realize the gains either on top line and bottom line that you're looking at right because I get it I got I got understood all that was with the I'm still going to hold you to the data thing which is that let's say for instance you're working with I mean you work with enterprises right so again let's say you're naming you're working with say company a company B company see they're all banks or NBC's let's say one of the message DFC one of the SBA whatever let's say it's one of these kind of companies and even before you start I imagine there are a bunch of two three things you look at to say you know is this going to is this coating code promising like in the sense that is it going to work what's our likelihood of it is succeeding what's the likelihood of not succeeding and I imagine that comes down to attributes of these individual of course there are softer attributes like or do they really care do they invest forget about that I'm just talking about objective attributes of a data itself see what what we typically look at in terms of you know whether it will be promising is that you know will voice AI make a dent right and for that what we will look at is okay if if we look at the whole customer journey right from right from a new customer on boarding to you know engaging the customer supports service collections you know cross sells up sells the whole customer journey we typically one will look at how many customer touch points or interactions in a day or in a month does that enterprise have right because if they have that volume and scale then we know that you know voice AI will definitely make that dent right so got it and those touch points can be voice or non voice voice chat maybe what's that app right any conversational touch points right is what we tend to at least look at right by that sorry what's it that you mean how many times the lender interacts with a particular customer or the aggregate number of conversations total conversations see the total conversations because see when we are talking about voice AI right we are not looking at necessarily like we don't consume PII data right so for me it is ultimately masked or anonymized data that I am dealing with right so it's not so much about unique conversations per customer but you know total conversations and touch points and then you know and then again is like things like complexity of workflows that we will look at right so for example how many languages do they operate in is it you know tier one customers is it a lot of let's say for NBFCs a lot of, let's say, rural and SME lending, things like that, tier through tier 4, because there the complexity of operation goes up many fold, because they have to do it in multiple languages. They have to manage between the ground workforce and an AI agent, all sorts of things. Complexity of workflows, because even there, we know it will be promising and AI will make a dent, things like multilingual, things like complexity of workflows, volume of data across use cases. And yeah, then I think ultimately we do look at sample calls, right, or calls that they are doing, let's say, if they have large call center operations for support or service or selling and so on, we look at what are some of their best customer calls, what are some of the best call center agents doing, right, because the conversation design, the kind of nudges that you're doing, those then become very important for us to also train the voice agents, right, and saying these are some of the best conversations in terms of customer you know success rates, in terms of collections rate, all those kind of things, and then feed that into training the voice agents as well. So those are other things that we eventually look at. So got it. Those are your last question. Do you, sorry, I was just going to ask that does use cases also come into it? I imagine use cases does, because at some level, if, let's say one enterprise comes and says, oh, we want to use voice AI to basically, you know, reduce our customer support time and get a resolution faster, you might say, yeah, that I'll do. But if someone comes and we want you to sell refrigerators to our people, I imagine that might be, oh, okay, that's a harder problem. So imagine some problems are harder than others. Like, you spoke about complexity of workflows, but I imagine the objective itself can be either simple or complex. Actually, interestingly, for us, you know, right now, support and services probably like 20, 25% of our agents. Interestingly, now, you know, sales and marketing use cases have gone up to almost 30% or so, 30, 35% because people are wanting to use voice agent for things like, you know, sales, right? Obviously selling a refrigerator may not be, you know, but then there can be an assisted flow. So for example, if a customer, you know, you can do the first level of lead qualification, saying, you know, okay, is this person even interested when was the last purchase done, right? What is the propensity to buy your purchase, right? And if you're able to pass that to, you know, a seller or a human agent, then you've already probably eliminated, say, 30 to 40% of the effort of the human sales. So we have situations like that, right? Where it's, think of it as assisted selling, right? Where the first stage of qualification is done by the voice agent, and then passed to the human agent. Got it. And this is where the workflows come in, right? Because what you really are then telling the enterprises that look, if you want to sell whether it is these apartments or home loans or whatever it is, right? We can do it with this assisted way, but the caveat is that once we begin and when an interest comes, you should be prepared at any point in time to take over immediately. And if, so if they have that, quote, unquote, capacity to do it and the willingness to it, that is when you realize that, okay, this is probably going to work out. Yeah. So it could be, you know, a live handover, if they have the capacity and preparedness, it could also be feeding that data back to their CRM. So it could be picked up by the team in not real time, but let's say in a phased way, right? So all those options are possible, and that's where the workflows and the integrations become important because this is not just like a silo agent is talking to people, right? That data needs to feed back to the enterprise into their system into, you know, either a ticketing system, into a CRM, into a email workflow, something, right? Which makes that call valuable to them to be used further. Yeah, I was on, I mean, I was on a customer support call. I don't know what it was for. For the first 10, were you the customer? The first 10 seconds, I actually thought it was the person I wasn't being attention. And then I said something. Then there's another person. That's when I realized, oh, the first one was an agent. Okay, I didn't realize that, I mean, I answered a couple of questions, obviously very stock questions, but I answered them and then came, I mean, I hope that was an actual person and not the agent. But I mean, how often are people able to tell an agent from a human these days? I guess, not very often. Yeah, I mean, it is becoming harder and harder because now you have the ability to have very natural and human-like voices as well, right? Or in the last component, which is the TTS. So it is becoming much, much harder. And the ability of the agent to also respond like a human word to the customer's questions, right? Which is what, I mean, which is what, you know, confuse me a little bit, right? You know, two questions, absolutely, you know, proper responses. And then came a person, oh, that was an agent. Yeah. Yeah. I think, yeah, interesting. I think we spent a lot of time talking about the mode or like what exactly is this asset that Bajaj finances access to, which got to where it is today. And I think we've gotten a good sense of the mode with respect to data and everything else. I just want to switch to the incentive. And this is the second part of it, which is what Bajaj finance also has a very strong incentive to come and say, oh, we are this FinAI company, we are this, we are that, et cetera, also because, you know, so we have to like not just let that pass unexamined, right? So let me just talk about this. So Bajaj announced this transformation like I said in December, 2024, it would become a FinAI company by 2029. So you're literally like maybe around a third of the way through. So I think the question is, are you ahead of or behind this roadmap, right? And if you look at the, whatever the Diwali 6X jumped that I said, they also say that 99% of their business requirement documents are AI generated. I have seen business requirement documents in my long career, Vassutan, I have worked together. Some of them should be AI generated. They should not be written by humans. 100% of marketing videos. I think they have also said is also like AI generated. Their customer communication is moving to a boss, bot interface, et cetera. So I think now I feel like it is going beyond just the standard workflows of collections, KYC, renewals, et cetera. So I don't know. I just wanted like very quickly bring Vassutan and talk about, you spoke about those two things, which is that, you know, it's either top line or it's bottom line. Could you like straight, you know, expand on that a little bit for a bank or for a NBFC when AI can touch a call, which is both reducing cost and increasing revenue. Like what changes? Is that revenue change like fixed? Is it marginal? I mean, whatever. Talk about the financial incentives for a company like Bajaj Finance. Yeah. So I can share, you know, maybe some use cases and then some metrics, right, which are actual sort of impact metrics associated with those use cases. So for example, I think earlier I was talking about marketing and sales. So typically, you know, for a new customer acquisition or even for let's say, you know, loan top-ups, you know, new product upsells, right? Like someone has a type of loan, you want to upsell a new loan or a credit card, things like that. So marketing and sales use cases, typically I think now the whole process around at least lead enrichment, lead qualification, a lot of, you know, banks and NBFCs are starting to use voice agents for that and see multiple benefits of it. One, you don't have a capacity constraint because it doesn't have to be done sequentially, right? In a single second, you can like do, you know, ex-thousand calls, right? So like for us, we can do about, you know, go up to almost 20,000 calls a second if we chose to. So the capacity increases just tremendous in that way, right? For any kind of, you know, outbound use case. And then, you know, the ability to do that and then integrate it back with your CRM or ticketing and so on in a very consistent standardized way is very, very high, right? Because if it's a human, you know, agent or, you know, salesperson calling, will they put that data back in, you know, what did they listen to all of those things, right? So the sort of the capacity to skate and the standardization of response, right? These are things which, yeah. What are you really saying? Sorry to interrupt. You're basically saying that it has solved the thing that everybody has been trying to solve for a long time. It's a word that you and I know very well from our days that in movie, it's called attribution. You're able to like literally attribute. You can see that this sale that I got was because of this agent. Yes, yes. So, you know, that really, you can say brings a lot of value on the marketing and sales use cases and it's applicable to any, any kind of product in the, you know, the BFS space, insurance space, the full journey, right? If you have customers coming on your website or app, for example, and they drop off, right? It can also, you can just trigger a call to that person saying, can I help you close this? What happened? Etc. Right. You don't need to wait for again, you know, someone to go back to that person. So, all of those things. So, the third aspect is real time, right? So, the capacity, the consistency and standardization and then the ability to get back to a customer real time. I think we all know how much of an impact that makes, right? It's like the cart abandonment use case in e-commerce, right? That if you get a call back immediately, your propensity to maybe close the transaction is higher, right? So, these are, you can say, impacts of using it in the marketing and sales context. And we see, we see, you know, anywhere in this, like, at least from whatever that we get in formerly and formerly from customers that there is almost, you know, the interest rate that we see is 10 to 15% and this I'm talking specifically in banking and NBFCs for different kinds of loans. It could be a goal loan, a personal loan, any kind of loan, but 10 to 15% interest rate, which at least from an industry, benchmark lens from what I know is almost three to four times, right? Typically that you would get from a human agent kind of scenario. So, so extremely sort of interesting there from an impact lens. The other I think customer even engagement, right? So like we all know of relationship managers, right? From our world of, you know, savings accounts and so on, but customer engagement at scale can be a challenge, right? You cannot have a relationship manager for every customer. So if you want to do things like, you know, welcome a new customer, you know, give them information just about the bank, it's products, right? Just engaging with the customer about, you know, new offers which may have come, right? Something new about their either account or, you know, loan that they have, that is now possible, right? Without having an RM for everyone. So the whole customer engagement, but again, becomes an interesting one. Obviously the impact of that, we look at at least things like C-SATs, cause and so on and how that improves. Because ultimately that should link to the top line for the, you know, for the institution. - Got it. So okay, I see how sales and marketing is attributed because it can be attributed, because it can be attributed. It obviously links back to top line. And of course anything that you're doing in customer support cut-stown costs, I imagine that voice air companies like Nani, I mean, there are so many. I believe that what they would probably charge, or the way they would charge banks and imbibes the billing side would be on a per call basis, is it fair to say? - Yeah, per minute per call, basically kind of that kind of pricing structures exist today. - Fine, so is that kind of a pricing structure? And obviously that is like, you know, I imagine it's like a several notches below what you would basically pay for humans. So obviously you have cost savings coming over there as well. How do you do, I had only one question. How do you do attribution for things like collections? - Sick, yeah, sick collections again is typically it is, you know, okay, so you can understand is one of the biggest use cases. - Yeah, so the collections was like so marketing sales, customer engagement, third use case, yes, is a big one is collections and collections, you know, can be for so many buckets across, right? Because you have a pre-dub bucket, you have a self-cure, you have a bucket X, right? So the quality of the conversation, the nature of the conversation, right? Changes with every customer segment. And the voice agent has the ability to do that, right? Let's say someone's, you know, past the, you know, 60 day mark that conversation will be a bit harder, right? Whereas someone who's in the pre-dub bucket, it's just a friendly reminder, right? So those things, obviously collections covers the entire spectrum of customer segments based on when, you know, the renewal or their EMI, et cetera, is due. And that is measured based on the fact that, you know, whether the customer on the call said that, you know, they have a propensity to pay or they will pay. And finally, when the bank collects on that customer, that gets attributed back to that particular voice agent. So collections, we are able to, because we get that data back from our customers, able to track that, you know, what is the improvement in collection efficiency or collection percentage of the voice agent versus say a human agent, because typically, you know, they will also be slain and benchmark based on that. - Correct. So it's not just a thing of the, which voice agent the customer says, I will pay. There's suddenly that voice agent company is like, oh, I have the one who has to get credit for this with this person paid. Sita, I just want to bring you in, because when you're talking about incentives, right? And something that I wanted to connect to what you had said earlier, you said that Bajaj finance has this propensity and history of making these like big, ambitious claims and investor side. So we have to like really, and we, and FinAI, I think falls in that bucket, okay? So we have to really talk about this, because here is a company that's coming and saying, well, we are a FinAI company. Of course, the numbers are impressive, and we'll talk about this, but we need to look at it slightly skeptically as well, which is that one, the really, really skeptical lens to look at it is that look, this is just a normal automation program that they're doing, and they just have like better PR and better storytelling. And here are a couple of things, okay? Which is that, so one of the things that I've seen is that Bajaj's cost line really hasn't moved and this is what Vasuta spoke about. Vasuta spoke about, oh, you can deploy all of these agents and you can make yourself like FinAI, you can create all of these use cases, but if you, and they have done some great stuff, but they're cost of, if you really look at the financials, has not particularly changed. And of course, if you ask Bajaj, I imagine they would say, look, these benefits are going to come within 12 to 18 months, et cetera. So I think my question to you is that knowing Bajaj's history and with all of what they have done so far, how do you read the whole FinAI company? Do you think it's credible or is it something like a public market label and sort of like, even investors know and they reward them for this ambition rather than punish them for falling short. Because for instance, if you look at companies, actual companies who came out and said, we are now in AI company, look at, I mean, most famous example is Klarna sales first to some extent, who came out and said, we are no longer whatever we are, we are now in AI category of one company and they got destroyed for it, right? Because investors took them at their word. Now I'm not saying that Bajaj is in doing that, but is there a sense that because of the history of the company, investors are like treating it slightly differently? - Yeah, investors don't really care for all these terms and labels over here. I mean, especially in the context of Bajaj finance. And this is a company that tends to nickel and dime. Okay, and it has to because it's not a bank and it's an NBFC. So banks have the luxury of having access to CASA, which is current account and same. These are the lowest cost deposits funds that you can possibly have access to as a lender. And Bajaj finance because it's an NBFC can only have fixed deposits and not savings account deposits. Okay, so whatever it can do on the cost front, even if it's a little bit, it goes a long way given the volume of loans that Bajaj finance processes. And we have this very interesting anecdote in the episode and intermission, where I think somewhere in 2012, 2013, they have obviously a lot of lenders do this, there's something called loans against property. So where you pledge your property for a loan, I think the cost of processing of one crore loan was something like 45,000 rupees for Bajaj finance. Okay, because this property, there are multiple levels of legal verification and whatnot. So Rajeev Jan said, why is it so high? Okay, we need to bring this down. And finally the guy who was heading that set, I'll bring it down by 10%. He said, no, we have to bring it down by at least 50%. And so he sits with this guy, he says, okay, I'm gonna sit with you every Saturday and we're gonna break down the cost and see why we spend so much on each of these categories. And I think 18 months down the line, they're carried by not 50% by 80%. So they have to, they have to be really particular about how much they spend on processing, originating or processing each loan, because at the end of the day, even if they are top tier among NBFCs, they're still an NBFC and they don't have these low cost funds. So I think what investors would look at is not these claims, but what these claims amount to in the company's bottom line, say 18 months or 24 months down the line. All right, if that's made a difference, then it doesn't matter, I mean, you can call it whatever you want, right? And also, Bajaj Finance has shown in the last 10, 15 years that it's considerably better at a lot of its competitors and getting into newer categories and upselling loans. But at some point others will also start doing the run. They already started doing that, right? So now how do you differentiate? So if everyone and as a customer, as a potential borrower, I'm gonna go where I'm gonna get the lowest interest rate possible, right? Maybe like with home loans, I'll be like, oh, I'm gonna have a 20 year relationship with this lender. So I'm gonna spend a little more time. Otherwise, which one offers me the lowest interest rate possible, right? So if it's become super commoditized, then what is your edge, right? If you're, if someone who's borrowed from you five times is gonna treat you the same as a lender that they've never interacted with, then you do everything in your capacity to bring your cost down. And that makes Bajaj Finance successful. Then it could already its collection efficiency is really, really good. I think one in every 100 customers tends to default, which is insane, okay? So now, - So they not only have, sorry, they're not really able to ridiculously high approval rate, which you already said, they also are really high quality, which is automatically points to the fact that they just know home to lend to very accurately, and they know how to get it back from them very well. - And all the things that Bajaj was just talking about, the use of voice air in collection, right? Now imagine a company which is already very good at getting their money back, right? They put this to use as well, right? So then you don't have to say you bring down your, your NPAs or non-performing assets a little bit, then you don't have to prove it. for those, right? Again, that just, you know, improves your bottom line. So I think all these little, little sort of improvements go a long way for a company of Bajaj Finance's skill. I think that's what it is. It doesn't matter whether it's a NAI or it continues to be called whatever, like a big NBC order. It doesn't matter. I'll add one interesting wrinkle to the whole Finneia this thing because you spoke about Bajaj Finance, whatever they want to talk and what they say. If you actually look at the quarterly numbers, there is actually one part, one place where they do a bit of a walk back. And I don't think people have really spotted this, but I was doing some research and I figured this out. So in Q3, we spoke about Q3 and Q4, the last two quarters, right? So in Q3, Bajaj Finance actually said that their guidance that they gave in the investor earnings call was that we're going to deploy 800 plus autonomous agents. Now, agents is the next big thing, autonomous agents, which include sales. They're very specifically included sales into this. Now in Q4, which was last quarter, that number became 600. And sales was dropped out. So they didn't put sales in it anymore. So of course, the number has reduced fine and things change. We understand that, but I think the part that I found very interesting is that the highest takes, the revenue-facing function. They just said, you know what? This thing autonomous, let's not really like put it there. So I think there is something interesting here about the places where they're deploying it and they're sort of realizing maybe not this. Let's just come back. That's higher order than even collection, I suppose, right? I mean, you're trying to, right? So I guess the agents will have to get as good as Bajaj Finance sales agents for the company to actually use it for that function. I'm sure Grannie and Vasitha are working on that. Now, finally, I'll come to the last section. So the last thing I really want to say, I was, okay, I thought about what makes Bajaj Finance, and I spoke about the data and the moting. We spoke about that. Second is we spoke about the incentive. It's a public company. It has this history. It has this thing that it has to say, wants to say, can say, and whether investors reward it, don't reward it. How it connects to top line bottom. So it has the incentive. But the third thing that I want, I realized is it actually has leverage, because Voice AI is not like any other forms of AI. It's not the same. Right now, if you look at company after company who are, and we've done this in our other podcast, ZeroShotvis AI podcast, they're quaking in their boots when someone like a Claude, which is of course, one-band tropic, someone like an open AI, where they bring out products for say to make coding better. Claude has come out with Claude for legal. They've also come up with Claude for financial services. So apart from this, Voice AI kind of is a little distinct, because it's not like everybody who wants Voice AI or believes in Voice AI or uses Voice AI has to go and get a Claude subscription. There are like, I mean, Voice AI was actually quite good even before, like, Gnani of course was doing it for quite some time. But even before that, you had a whisper and open whisper, which you can literally run on your own laptop. And in fact, at the can we do that sometimes? And we have interviews that you want to transcribe, etc. You can run it locally. It's pretty decent. And on top of that, you can do other things like Ospeaker diarization, making voice better, adding some context in the multilingual, etc., etc. Fine. But even in multilingual, like, I would argue, in fact, until recently, people were testing out like, service multilingual with like, Gemini's and they were like, this is not very far apart. They are almost more or less the same only. So I feel like that is interesting because from Bajaj finances perspective, they are, oh, you know what? We can just deploy multiple pilots, multiple, companies we can get on board and say, okay, you do a pilot, you do a pilot and watch everybody fight it out, play one against the other. They can basically say, no, no, this is cheaper than this. So I'm not saying Voice AI is a commodity, but I'm pretty sure Vasotha will vigorously disagree. It's a commodity and I want to hear from her. But I'm saying that does it feel that way? Is that something that gives them an edge? Vasotha tell us, Voice AI is clearly not a commodity because obviously there are so many companies doing it. What exactly gives one company the edge over the other? When you're going in pitching to an enterprise and enterprise, I'm sure we'll tell you why you, I can pick 10 of these people out there, including like the 11 labs of the world, the servants of the world, etc. Why you? I'm sure you have a very clear answer. So, I think couple of things, yes, there are a lot of companies in this space, especially I think in the last probably 12 months, a lot of companies have emerged. I think what differentiates or what will continue to differentiate is whether you are just building a wrapper on top of another platform and model or whether you are actually owning the whole stack. So for example, you could use anyones ASR speech models, anyone's LLM, stitch it together, then you need an orchestration platform on top of that to actually code the agent, build the voice agent, and then finally you build the agent on top. So there are different components to the voice AI stack, which a lot of companies were just building the agent layer on top. They don't own, they license it from others, which leads to two or three things. One is of course, you're cost of delivering that goza, because if you don't own the underlying models or the underlying stack, you're cost to deliver an agent, voice agent, goza. So obviously that impacts your pricing. Second is I think when you talk about actual conversations in India, they are not happening on the web, on high, Wi-Fi connections like we are speaking now, they happen on a telephone line in a very noisy environment, where you hear horns and kitchen noises and also things in the background. And that's when a lot of things break, because you're talking about telephony frequency, which is much lower than these web conversations. So lower frequency conversations in real world noisy conditions. And I think that's where it emerges from just being just a nice to have just voice AI technology to what you can really deliver in a real world noisy environment at production grade scale. Because what we talked about earlier, if you're trying to do like three to five-corrored calls a day for 200 enterprises and deliver them at low latency. Because let's say when I'm speaking, you will understand if the latency is more than like one one and a half seconds between when the agent speaks and when the customer responds, you will understand something is up, right? Or something's not right in the conversation, right? So the noisy conditions, the low latency and being able to do it at production grade scale, I think those are things which become very, very important when you're trying to really deliver voice AI impact in business metrics, in the top line, in the bottom line, metrics that we were talking about earlier. Because that's when it like rubber hits the road kind of thing. So you know, it's very, yeah. From the customer's point of view, you want the agent to be almost indistinguishable from a human. Yes, indistinguishable from the agent, but not just in terms of how human like the voice is, right? But the latency of the conversation, being able to, you know, cut, like if the customer is sitting in, you know, in in a village somewhere and you know, there's all, you know, in near the kitchen and there's a mixy running in the background, being able to cut that out and still recognize what the customer is saying and be able to, you know, respond to it on a telephone connection, right? In Tamil or in Telugu, right? So multi-lingual languages is the other part of it that with how much accuracy can you deliver in all Indian languages? So possibly, be a joke about what is being made, right? Yes, possibly possibly when you get to the, you know, human emotion and human part of it. So I think those are things which I think differentiate P.T.K. back to your question, right? It's it's not just about, you know, just building the agent, right? It's when you get past the demo and the, you know, the POC phase and you have to do it for real customer conversations in multiple languages in a noisy environment. I think that's when the true impact comes. I think you're making an argument that I think most, uh, SaaS organizations make, which is fair. I understand the argument, which is that, oh, you know what? I mean, if you talk about SaaS as an example, people always say, oh, you can wipe code, like I can wipe code slack in a day. Sure. Good for you. You can. But what it can, what you cannot do is deploy slack in these 20, 30, 50, 100 other use cases in these kind of enterprise-grade security with this kind of complex, well, that stuff, the whole workflows and all of these other things, that stuff that we know. So there's a big difference between exactly what you said. Oh, maybe somebody can come and wow, someone with a pilot. But for a company like Bajaj Finance, I imagine and other enterprise companies, I imagine a big part of your conversation is not about the technology itself. But whether, can you guys, how big can you go? How much can you scale? How much can you take? How much capacity do I believe that is the clincher at the end of the day. See, then I'm going to throw it to you to close this because we've spoken about all the things that made, or gives Bajaj Finance this edge for this finneye claim and how they've gotten here and what is it that they had that nobody else had, right? And we spoke about three things. I said, first, there is this mode of data, second, there is this incentive. It's a public list company has to say these things. And third is, of course, it has leverage to some extent on all of these voice AI companies that it has. Of course, Vasotha has explained how they all have their own differentiating point. as well, but these three things apart. I think there is a fourth and I think we have touched upon it very lightly, but maybe you should end here, which is on, I believe there is something about the DNA of the company itself. And I feel like that is really what we have not gotten to and I can be all very MBA about it and say, oh, this is the three reasons why they got here, et cetera. But I think there is an aspect of this company itself, which is not replicable, which is something that an SBI with even more data can or maybe at least at yet cannot do like an HDFC like all of these other companies you spoke about with which have access to all of these things. Something is there about this company itself? Yeah, I think it's always looking for these sort of wedges where there are opportunities. Like, after Bajaj Finance showed every other lender how you can just go after each borrower and sell the multiple loans. Everyone else, everyone else, everyone else, all started doing the same thing. Now, Bajaj Finance may be the first to deploy agents at scale. In three months or six months, you'll have every other lender doing the same. So now going back to the period when Bajaj Finance rivals also started calling up people repeatedly to sell them loans. Despite that, Bajaj Finance ended up growing year after year, ended up becoming more and more profitable. How was it able to do that? Now, similarly, after everyone deploys voice agents and all of us are talking to agents all the time, what makes Bajaj Finance better than everyone else? So I think that, like I said, it's not willing to settle at all. Like, okay, I may be selling 12% of all loans in the country, but why is my value share only 3%, why can't it be 10%. Now, what can I do to get to 10%? And if a company tracks something like how many of my products, a customer uses and tells investors repeatedly, I want that number to go up. I don't care if you already have a discount-broken account, I want you to have that account with me. I don't care if you bought your last insurance policy on policy buzzer. I have my own marketplace. You come there and buy. Now, what can I do to get you to do that? You look at all these companies are considerably smaller than the largest discount brokers, the largest insurance marketplaces, right? But for the company as a whole, these little things add up. I think their discount broker brings in something like 800 crore and revenue and it's profitable. That's nothing to laugh at. So what differentiates Bajaj Finance from everyone else is, you know, it doesn't take customers for granted. Even if there is a little bit of money to be made somewhere, it's willing to do that. Even if only one out of 100 calls converts, they're happy doing that one call, right? I think that I don't think, I mean, maybe other lenders could also pick this up, but so far, I haven't seen this play out in any other bank. You can just look at what they disclose, how their CEOs approach this and interviews, and I did a fair bit of research for the episode. You don't see the certainly not SBI, but not even, you know, HDFC or COTAQ or ICIC, you don't see that at all. And I mean, some of these lenders have also had multiple CEOs over these years and that could be one reason why they've not been able to do what Bajaj Finance has done with just one CEO between 2007 and now, right? So I think that is why I don't think anyone else will be able to do what these, what these guys do at Bajaj Finance. I feel like what you're fundamentally saying is that they're willing to go and do this battle of inches where every day I will gain one inch and one inch and one inch and all those and that is a different mindset actually. It's a very different mindset from sitting at one day and saying, okay, we're doing a great business. What's the next big thing? Right? And let's build towards that. So I feel like this hyper optimization kind of a mindset, I think has that's really what you're suggesting is the thing that differentiates it. Yeah, being an on-bank lender, you know, you begin with constraints on funding. So when you have these constraints, then how do you work around those constraints? So what can you control? I think they've always been about that. I can't do anything with these constraints. You know, the country's banking regulator is not going to let me do certain things. That's a given. Now, what is within my control? And how can I save a little more every quarter? Interesting. Thank you so much, Sita and Vasuta. I'm just going to end with what where I began. I began by basically asking this question. Bajaj finance has all of these clearly is doing something interesting and whether this whole thing is like, I think really I wanted to just ask the question, is this thing that they're doing and is all of the stuff that's happening? Is it a moat or is it a slide on a presentation? That's really where I wanted to understand. And so either it has found something extremely durable. And we have spoken about all the stuff that is durable. It's data and sorry, it's data, it's incentives and the way it works, it's DNA, it's reaching and trying to find inches and wedges everywhere else. And this is all stuff that maybe everyone else has access to. But it has the machinery to act on it inside a single customer relationship and everything that it has done in its past has all compounded and added to where it is today. And fundamentally, now that has just like allowed it to like explode and go out into this brave world where AI has just undecubined or else what really may be happening is that like we said that it's probably something that is Bajaj is now aggressively disclosing. It's talking about it is created a very good, it is surely doubtless, deployed all of these things and is doing a lot of stuff. It's doing really aggressive and a better job of talking about it than actually finding out whether it leads to actual revenue, cost impact and all of these other things. I think we'll find out, I suspect we'll find out in about 12 to 18 months, that's really 12 to 18 months is obviously the right time frame when you have to discuss anything, whether it is with acquisitions or IPOs or finding out things is what you tell your boards, it's what you tell your bosses, give us 12 to 18 months and we'll see where it takes from there. Thank you, thank you so much, Vasotha Sita for a wonderful episode. Thank you, Peeja, wonderful conversation. Thank you. Thank you for listening to this episode of 2x2. Today's episode was produced and hosted by me, Raveen Gopal Krishna and the audio production was done by our technical producer Rajiv CN. If you feel like there's something that we missed out in this discussion or just want to share some thoughts and some feedback, just write to us at TWOBYTWO at the rate of the hyphen Ken.com. Also, if you're listening to this episode on Apple Podcasts or Spotify, just show us some love and leave us a rating or even better leave us a review. Finally, if you love this episode, do share it with your friends, family or colleagues who might find it fun to listen to as well. As always, we'll be back again next week with another great discussion and of course, a 2x2. you