Intellect Design Arena Limited (INTELLECT) Earnings Call Transcript & Summary

Thank you for joining us today. The exclusive event for the investors is to perform operations with our enterprise AI platform, that is Purple Fabric. We have with us today Mr. Arun Jain, Chairman and Managing Director; Mr. Banesh Prabhu, CEO of IntellectAI; and then we have Vasudha, she is the CFO; and we have Deepak Dastrala, he is the CTO. So now I hand it over to Arun for his briefing first. But before that, one safe harbor. I would like to remind you that anything, which we say, which refers to our outlook for the future is a forward-looking statement, which must be read in conjunction with the risk the company faces. With that, I hand the mic to Arun. Over to you, Arun.

Good evening all the shareholder investors. Good evening, everyone. Thank you very much for joining this special session on Purple Fabric. This was requested during the investor presentation, where all of you presented it and asked -- asked to present what is the Purple Fabric means. We know it's a multi-agent AI platform, but what does it do? And that was the curiosity you demonstrated in your investor conference. So just to recap what Intellect Design Arena stands for. It stands for the kind of wastages, which happen in the way IT industry works. I was addressing a bank conference 2 weeks back and some of the CEOs are spending that we have a technology budgets are need to increase. Somebody is spending more than double-digit billion dollar technology budgets. And they're saying we have a technology manpower of more than 50,000. To me, is it the right thing to do that you need to have 50,000 people? You need to have -- spend more than $10 billion, $15 billion for technology? Something is wrong somewhere that we are solving the problem and creating a forest and jungles of the knowledge or data. And if you apply first principle thinking, which I was reading a book of What is Life from Paul Nurse, he simplifies the life, which looks so complex to me into just 5 words. There's a cell, there's a gene, there's an evolution, there's a chemistry and there's the information. So 6 words -- 5 words, he simplified life. And similarly, when we talk about Intellect Design Arena, what we had achieved in the last 10 years, yesterday was our 10th anniversary of going listed in Bombay Stock Exchange. On 18th of December, we got listed in 2014. In 10-year journey, what we have achieved is we asked this fundamental question of why bank should exist, why financial institution should exist. Financial institutions should exist because there is a financial event happened in the life of a customer and customers are consumer, SME and corporate. There are 3 categories of customer a bank has. And bank provides just 6 or 7 services, core events they handle. One, first event, is opening a savings account when I get a job first. Then maybe some lending account to buy my scooter or a car or I have surplus money to do the deposit or I have to invest some money, investment, and I need a credit card, and I need to make some payments, a payment. This is the role of a consumer banking. When it comes to SME, then I may add supply chain finance to it or working capital to it. And if it is a corporate, I need guarantees or some instrument of LC and guarantees, which make -- which is important for me to provide my integrity to the -- some of the bids or some of the export I'm doing. So these are the 7, 8 services bank -- events happen in the life of the customer and bank provides the services. And these services when you package into technology, it's called micro services. So the events are finite. The services, which is required to support is also finite if it is revolved around client. But today, if you look at it, many banks say, I have 50 products, I have 60 products, I have 150 products. And when we go to Europe and America, they say we have 400 products. Now these 400 products when they have to support with 400 different databases, you are creating a huge data waste, and we are challenging that initiative. And then similarly, this -- when there are 400 products, they get connected to each other, there will be 10,000 APIs. We simplified this entire architecture to make it finite thinking possible. There are only 386 micro services. They have close to 750 events, even micro events, and there are 2,000-plus APIs. That's all. So now by making everything finite, we are able to go deeper and deeper and providing the configurability to compose the solution and to contextualize the solution. This is what we call black book for us. When we go to a customer, we just show this black book that this is a map of the banking. There are 13 product lines banks operate on. On the top, we have their 8 channels and back-office operations here. Then we say each product line, if it is G1, how this product line map to the detailed processes of the bank and each process over here is numbered, it's ZIP coded. And for each line of business, lending to credit card to wealth to transaction banking, liquidity, all our wealth management is put into eMACH.ai. So this is my core collateral for selling my banking. Our proposition to the banks are stop buying products, start buying customer solutions and which you can do the composing yourself. And this is available in a next level. Each micro service has been codified into a coding ZIP code number. Each API is written over here as an API number. Each technology is a blue book, which is a micro services security for getting an enterprise-grade solutions. So we invested close to 20 million hours of the time in 10 years since we went public. 2 million hours per year at the cost of -- international cost of $100 per hour, it's a $2 billion cost. At the cost of India, 20 million hours still cost $400 million of the investment, which we have made in building this platform together. Today, I want to spend some time on what came out. We have a team of close to 150 research engineers. Since 2016, they are being operating from U.S., U.K. and India. And these people did a -- since they started in 2017, they knew what works in AI, what didn't work in AI. We failed in 2018, we failed in 2020. We did some work. We succeeded to bring the accuracy to 87%, sometimes we bring 89%, sometimes we bring 91%, but we are not able to make it usable AI. The difference is the promise of AI and usability of AI. And that's the biggest gap we find in our journey. But since we had more chances of exploration and experimentation between 2017 to 2021, '22 when generative AI landed us in 2022, we could able to identify the right places where generative AI can create a larger value for the customer. And we -- over the period of time, when we built up these use cases of magic submission in America for insurance underwriting space or we built up -- we sold this technology for ESG for largest investment bank, Norwegian investment bank. Or we use it for magic and voice processing, all were able to build on the same technology, and this technology revolves around principally 8 technologies. First is how you get the data from the various sources, whether it's a structured data sources, which is a data source DS1. You get the data sources from second data source, which is the documents, which are present, which customer is submitting it or document you are crawling and retrieving back from a website of a customer, that is a DS2, data stream 2. There's a data stream 3, which you pick up from websites from various -- not websites, the paid data, you buy out from Dun & Bradstreet or ROC data. Then there could be regulation-related data and there could be operation-related data. This data, we convert into knowledge by using 4 technology. The technology #1 is ingestion technology. Technology # 2 is -- Deepak will highlight to you what these technologies are. Technology # 2 is classification technology. Technology #3 is extraction technology and technology 4 is trusted data technology. Now these 4 technologies, once it is done, we call it Document Intelligence Management System. And this is the 4 technology constitute to create DIMS system, which is parallel to DBMS, what Oracle has promoted in 1985. So today, with AI coming into picture, bank needs a document intelligent management system of their own. It's not a part of LLM. It's a part of the bank -- their own LLMs, which should be sitting in the bank system or whether we call it bank LLM or bank SLM. But these documents must be tagged so that we can do a continuous KYC and continuous risk monitoring can be done on the same documents. From this, we move to technology #5, which is how do you create an agent, which will show -- Deepak will show you how we create agent by feeding in the information by creating almost like a digital expert. So there's a digital expert and there's a human expert. So how do we replicate human expert to digital expert is the technology #5. Once you create a human expert, we simulate an operation room. In operation room, there will be at least 5 or 6 different category of experts are there, which we call roll holder. They could be sales manager role holder. They may be ops manager. They may be compliance manager. They may be a collection manager, who is responsible for collection. They could be risk manager. So all these roles, we create 5 human expert in operation room. Now technology #6 is how these 6 digital experts are talking in the same operation room is a technology #6. Now once the 6 people come together in an operation room, you can understand how much noise they can create because everybody has their own perception and contradictions. To manage the contradiction among the 6 people is the biggest job management deal within a bank. But in AI world, it's much better. We developed a very, very highly sophisticated algorithm called Socrates Dialogue Algorithm. We trained this digital expert to do the prompt engineering among themselves and ask the questions, which makes the difference to their life. And that is -- Socrates Dialogue is technology #7. And once these contradictions are there, we record all the conversation behind it with the traceability. So if regulator wants why you come out a decision or business head wants why operations has decided in particular lending or not, he can look at the actual use case and traceability of it. And all our -- throughout the journey, then this -- the jury of 6 agents will give a final file or final recommendation and that recommendation goes for the human adviser who is there. So I'm converting the operation room of 100 people to just few recommendation advisers are who can take decisions on it. And that recommendation is the technology #8, which should come in the shape of file or it can come in the shape of recommendation letter. So this 8 technology, we wired together and all are wired into a single box with all the security and entitlement embedded into it. And all these technologies are LLM agnostic. So this is the first platform, which is LLM agnostic. It's not that Microsoft is selling OpenAI, which is better than Bedrock. We say for each different use case, we do the benchmarking of the LLMs and this benchmarking of LLMs capability that on 3 AI constraints, which is a speed, cost and accuracy, we benchmark, which is the best LLM there. So this Purple Fabric platform can consume the best. An optimized LLM model for solving a specific problem, not 1 problem. Within 1 problem, we may use multiple expert agent. Deepak will show you how we use for solving 1 problem, more than 13 experts, digital experts to solve that problem. Now you look at it, what is the market size as an investor, how will it translate into the business scenarios. I'll come to you after Deepak present to you Purple Fabric that how we are translating this into the business case through 3 different business models. How Purple Fabric is leveraged for creating a shareholder value. So at this point in time, I want to give the stage to Deepak.

Thank you, Arun. Good evening, everyone. Let me know if all of you can see the slides.

Yes, go ahead.

Sure. See, one of the biggest opportunity, right? I mean, historically, all the technology companies spend in terms of helping the banks. But if you look into that typical area opportunity, which at this moment we targeted was IT cost of a bank is anywhere between the 11% to 17%. And at least 40% to 46% of the cost goes in terms of an operations. And this is a completely new space. It's actually up for the opportunity for a Purple Fabric. Because right now if you see, right, the nature of a work that can be done in the middle and back office, so far, the technology could not penetrate into that because it requires a human level of a capability in terms of intelligence to do that. Now with where the AI is, where the Purple Fabric is, I think this is the completely new opportunity space, I think where we are working on the various use cases, and I will walk through some of these use cases. So the value of the Purple Fabric is not only to bring in the operational efficiency or reduce the cost, but also in terms of helping the banks to improve their revenue through hyper-personalization, improved customer satisfaction, enhanced risk and compliance and all. So pretty much we help both in terms of banks improving their top line as well as bottom line. And just to give a context, when we say AI agents, right, this is very important for me to set the context because a lot of people, historically, even automation, they might have a view about the bots. But why agents are different and how different they are from bots? At a very high level, I mean, the expert -- for some particular digital agent to become an expert agent, right? One, they should have a capability to leverage the foundation models, which pretty much trained on the universal data and process it. That's what first brings the human level of intelligence, number one. Number two, it should have a memory, just like a human, right? How we have a short-term memory, long-term memory, it should have the memory. Plus, it should have an access to the tools. Because with the kind of a digital transformation that happened over the past decade and so, so a lot of the digital information is available through the APIs or through website and everything. So it should have an access to tool for it to act. But to a little bit further context, right, to go right, at a very high level, you also can look into that, a combination of IQ, EQ and AQ. Why I'm trying to highlight this is because at this moment, what we are trying to do is build a digital twin of a human. It means at a very high level, it should have an understanding about the language. It should be able to synthesize the knowledge. And more importantly, it could reason and reflect because a lot of people whoever is playing with a ChatGPT, they might be using it to create a -- using a text, they can create a text or audio or video. But I think what we are talking right now is about actually further where you can do a human level of reasoning and reflect on certain things before making a decision. Another thing is EQ, which is empathy. And because whenever we are talking about one of the biggest use cases around the customer service, you should really have a right level of an understanding what the customer is talking about, whether customer is sad, angry, happy, any of those things and also adapt based on that. Then the AQ, and this is the most important thing. Because what people haven't leveraged the AI for so far is once you reason, once you understand, once you contextualize what actions that you can take, which typically so far only humans can do. Now as long anything is digitally available, actually, these agents can take an action leveraging the tools and all. And this is just to give you the context about what -- when we say expert agent, what actually qualifies in terms of the cognitive ability. Now at a very high level, Purple Fabric, as Arun highlighted, these are the 8 proprietary technologies that we have built in the last 10 years with a lot of research work that we have done and all. So if you see, right, a lot of people, I know at this moment, they just plug into the LLM and they call it as an AI platform. But what is the fundamental challenge is still the age-old problem of garbage in garbage out exists with the data. Unless you really get the data at the right quality, these models cannot really make sense and all. So what we use these LLMs is largely for -- purely for thinking like a brain actually, the doing part is what we have the unique capabilities. And what also we do is, for example, while we have this capability, the biggest opportunity what we have is, how we can leverage the knowledge that exists in the enterprise. That knowledge could be enterprise structured data. And here, right, and still a lot of critical information from a decision perspective, it helps in structured data. So we have ability to really tap into structured data. Then the unstructured data. And this is where we have spent a lot of time in the last 10 years. And because everyone who focuses on the data quality, all the AI programs, they have a challenge. And when they say challenge, they still they're struggling with structured data. And the 10 years' work that we have done in unstructured data, that is what gives us a significant differentiation. And we see the unstructured data, it could be a text data or it could be an image data. And right now everyone is comfortable with multimodality, but that's what we spent the last 10 years, how we can get this data right at a scale. And also, we can tap into the knowledge of enterprise policies and practices because most of the banks have their own, let's say, from a lending perspective, what are the guidelines, what are the process policies that they need to apply for a different line of business, so we can tap into all of the data. And along with the market data, it could be from website, it could be the third-party providers, it could be anything at any frequency of any format and then also the regulators kind of a thing. So these are the 5 different knowledges that we can tap into a structured data form or unstructured data form. Now a little bit going into the 8 technologies that we have developed. The first is a data ingestion technology. Here, we can actually ingest the data at a huge volume and basically at any scale that is possible, whether that is structured or unstructured and all. So -- and also for most of the use cases that we need to do, we probably we have to do it at a near real time across the various sources. That's what we can do. And then especially in enterprise, right, people will just leave the data wherever it is. And it's very, very important to have an ability to classify the data, whether it's structured or a semi-structured or an unstructured form because their classification is so important in terms of getting the context right and all. And normally, you might see, right, in the lending, someone just submit all the documents, all of that be in under 1 PDF. So how do you know actually which one is a PAN card, which one is a passport or which one is anything else, right? And sometimes they do it within a 1 page itself. So we'll be able to classify the data at a various level at a page level, the document level, it could be anything and all. Then how do we contextually extract and all. Because it's not just an extraction. This is where the major differences and all. We need to really understand what I meant by contextual extraction is, which use case we are looking into that? Within that use case also what sub use case that you are looking into? And how do you really get, right, and all. Because the word policy might be very different depending upon when you're looking into the insurance versus some other context and all. So how do you understand the context? This is where our deep domain knowledge in the BFSI will come into the thing. And that's what differentiates us from anyone else and all because it is not just about -- just applying an AI, right? Unless you understand the domain, unless you understand the data, you really can't get that last 2% right. And for us, this contextual extraction, where our deep domain knowledge helps in terms of how do we apply to the various banking, financial services and the insurance use case. And ultimately, it is about a trusted data. So because especially when you're taking the data from so many sources, how do you triangulate it? And how do you know which one is trustworthy, especially when you're combining your in-house enterprise data versus external market data. Even external market data, you don't depend upon just 1 source, you get from multiple sources. Each of them may be best at some particular entities and might not be good at other entities. Some might be good for MSME, some might be good for the large enterprises. So how do you bring all of together and contextualize where actually you can make a critical decision in your bank on the particular data. So that's what this trusted data helps us. Then the expert agent, as I mentioned, the expert agent, what qualifies, right? Now there are the 3 major user journeys what we focus on. While you might see so many copilots out there, so many agent-related platforms out there, we see it as 3 major journeys. It is basically Assist, Augment and Autonomous. Assist could be any help that will help you. It will give you only that 10% to 15% incremental productivity. Augment is, we are talking about a very deep domain expertise persona within the bank. It could be an operations person. It could be compliance person and all. So that's where we focus a lot of our UCaaS because that's where our domain expertise helps us to really make a difference and all. So how we build deep expert agents and all, that is the Augment. Then Autonomous could be typically any critical decisions never happen by the 1 person within the division or sometimes even 1 division itself. So how do you have these expert agents within the divisions and across the divisions can work together to actually make the business decisions happen. So a lot of our focus is around the Augment and Autonomous journey. That's where our domain knowledge and data understanding that we have worked on for the last 10 years makes a difference. And finally, the decision and action. So even the decision, right, how do you actually make these agents collaborate with each other? Because a lot of people are talking about agents there. Our differentiation is how these agents collaborate each other and how they actually come together to the decision and all. And then finally, what actions and all, when probably they are happy for someone to bring into the loop and when they are comfortable to make a decision. So this is -- these are the critical capabilities our Purple Fabric provide. Again, we are completely cloud provider agnostic and the model provider agnostic, and our deep knowledge has really helped us to really solve the last 2 questions better than anyone else. So what I will do is just to contextualize it further, right? I'll focus on the 1 major use case. By the way, I think to the best of our understanding we have, after talking to the many analyst firms, we are the first one to put out there the multi-agent use case in the world. While everyone is talking about the platforms, and I think we already went live in a highly regulated financial services firm, and I'll walk through that particular use case now. So if you look into the use case, right, right now the complaints or claims, while they are a universal problem. Right now if you just take U.K. alone at this moment, it's actually -- it's costing around $9.24 billion a month and handling the complaints across the various businesses and all. And this is the data from the Institute of Customer Service in U.K. So if I just take the use case that we have operationalized, especially after the consumer duty came into the picture last year, a lot of people are complaining around the various financial services. So what we are talking about a use case with one of the largest financial services firm where they are dealing with so many complaints. And these complaints are actually sometimes it will cost them anywhere a few thousand pounds to the -- sometimes millions of dollars kind of a thing. And they are not that easy for someone within their department to really investigate in a short span of a time. So the reality out there is at this moment, while they have almost more than 150 people, they're able to actually meet only the less than 30% of the SLA. And because of the complexity of the case, it is taking so much time. And the number of complaints are just increasing, especially because some of the legal firms are just recruiting the customers whoever has a concern, and they are doing everything on behalf of the customer. Why it is so complex is typically these complaints could be anywhere between last year to the last 14 years. And each of the time, the policies or the regulations might be very different. And there is no single source. The data might be across the various sources. And also because there is no standardized rubric for the nature of a complaint, it will be extremely difficult for people to really investigate and do that within the SLA of less than 50 days. So just for you to understand what it looks like, this is the typical operations team that looks like. If you look into the -- if you have to deal with the capacity of anywhere between the 200 to 400 complaints per week, at this moment, they need anywhere between 5 to 10 FTE just to look into the complaints, log those complaints, and they need 5 to 10 FTE just to classify and understand whether the complaint is right or not. And they actually need a 50 to 100 FTE if they have to meet a capacity of 700 per week to just do the evidence gathering. And then finally, to the case handling, they need 100 FTE. So if you look into the whole thing, right, everything is manual in nature because of the complexity and the nature of the challenge. Only thing that digital is they have a complaint system where they log and they use it as a system of record. That's all. That is what it is. Now what we have done is, this is where we build the multi-agents who actually replicate this particular personas and pretty much each agent replicate those persona, you have a compliance manager, just like how in any division you have a manager or a supervisor and then you will have the complaints checkers who does the various roles, someone focused on extracting the complaints, someone focused on creation of reports, someone focused on verification and all. So what we have done -- how these agents will work together is, they work together in terms of tracking the complaints, understanding and classifying them, creating the whole dossier, which shortly I'll do a demo and, finally, actually recommending the decision. So the most important thing that you need to understand here is these sources could be both structured and semi-structured and could be unstructured, number one. Number two, we never just use the 1 LLM to do that. And this is where Arun is referring, right? We benchmark what is purpose -- like what is the right LLM for the right agent for that particular job. So for this particular use case that I'm referring, we use more than 6 LLMs from the different providers and 13-plus agents actually came together to really deliver the outcome. So the outcome that we have provided is what took 5 weeks, we are able to deliver the outcome for -- in 20 minutes for 80% of the plus cases, which has brought in the significant operational efficiency and customer satisfaction. The beauty of this work is now what we have delivered to the customer actually doing in a very reactive work of doing the complaints because we have built agents in a modular and reusable form. Now the customer is using this to proactively look into the who are the other vulnerable clients who might complaint and what are the risk they have. So this is where our first principle thinking actually helped us to really build the whole thing in a very modular way. So what I will do is I'll quickly get into the demo, and then I will probably walk through what is the -- where exactly it is headed. So this is our -- this is our Purple Fabric platform. And what you can see is so many agents that we have here. So let me go through the agent who's actually the complaints manager, who pretty much orchestrates all the work that -- across the various agents. So what I will do is, because this takes some time, I'll just probably file the case in front of you, and then I'll go through the previous case in the interest of time. So before I go through that, what you see -- what you can see right now is just -- again, for the interest of the demo, we are showing it more as a conversation, but the whole thing works as an API at the back end. So what you can see here is just I ask -- actually investigate this particular complaints of a case reference something. And what you can see is, there a complaint manager immediately kind of called the complaint logging agent. And in fact, this is where you can see live what is happening and all. So the complaint manager, I gave in a very natural language, investigate the complaint for case reference. So this is where the decision logic because this is where the explainability and everything will come into the picture. What the complaint manager has done is, first, I need to extract log and verify the complaint using the complaint logging agent. And then it gave what is the action. This is where what I referred as what is the decision, what is the action. You can see we made it completely visually available for the people to actually understand what happened because tomorrow, regulator want to come and see, okay, how these agents had made a decision. What exactly happened? Can you prove me? We have a complete audit trail at any point of a time how this happened, okay? Now if you see, right, I mean, this is just what happened is that it has kind of got that particular complaint. This is actually the letter, which it has read and it got completely that information, which is very unstructured here in nature. And then it has logged this particular complaint in the complaint system. You can see the input for that. What I will also do is like just to see, it is also verifying whether the complaint has mandatory details or not. And it is also clearing, giving the output, it is actually successful. For example, it is saying that all the required -- if you can see what I highlighted, all required information present, which is the first name, last name, date of birth, address and everything. And also it has the letter of authority and all. So just like how the human will do, the agents are doing it. Now what I will also show is, it is actually -- it took that particular text, it is classifying the complaint. This is where like a human, right? When we read it, we can always see there might be more than one category might be qualified for a complaint. Sometimes when a human who is not trained, they might miss this. This is where agent has really read the whole thing, which is a text. And if you can see, it has kind of understand the complaint category. For example, if you see here, it might not be that visible. It has 2 categories. One is inappropriate product transfer and the partner service. And it also kind of validated according to the FCA category, what are those particular categories they come under. This is a pure human intelligence, which is actually replicated by the agents. Now if you can see, as I said, right, complaints gathering. This is where you can see just these agents. It is actually gathering the data from the various client systems, and it is gathering the data now even from the data warehouse, which is, as I said, right, we have a capability to really get the data from the structured data. And all these things it is doing in parallel. Now what it also doing is once it gathered all the data, it is saying that the next step is, for example, in this case, okay, first -- the next step is to -- sorry, it started investigating that and all. So basically, it has identified that I need to investigate 2 complaint reasons, inappropriate product transfer and partner service. First, I will handle, just like human. First, I'll handle it inappropriate. So it is doing all this particular investigation, going through the various sources, which could be structured, semi-structured and everything. And then what it is also doing is, now I have an outcome of an inappropriate product transfer complaint, it will proceed to investigate the partner service. So it has completed 1 investigation, and it is going through that. And because it is happening live, some of the work is in progress. But just let me show what's already done. Just before this meeting, I had a demo. What it will do, I'll go through the traces. What it will do is once the investigation is completed, it says, I have all the necessary details to generate a final report. Now I will use the complaints report agent to generate the report. And what you can see is complaint agent has taken all the information that it has received so far, and it has created a simple, easily understandable report. So let me show that in the actual this one. If you look into this particular report, let me -- where you can see this. Just like how the human at the end of the investigation prepares, right? Dear case handler, I have completed the investigation, so on and so forth. What is the client information? What are the case details and what are the complaint categories? And what are the reasons and how these are against the FCA and what is the final verification status? The status is successful. And it got whole complaint summary and it got all the -- in this case, this is the context is a wealth adviser. So these are the various products. It has all the particular product details, the codes and everything, the client name and all the advice details. And then finally, what is the investigation. So in case, the first thing is, first is an inappropriate product transfer. The outcome is the complaint is rejected. The reason is -- it has identified actually within this particular record where there's an evidence that client has agreed to the recommended portfolio and also accepted the higher ongoing advice fee. So I'll show you where it is. And then similarly, the second one is a partner service. Basically, for something to be qualified under the advice service, it has to prove that actually within the 12 months, the adviser has met the clients at least once. So from all the kind of meeting details, it got all the dates, and it has calculated that actually the adviser has met the clients multiple times in less than a year, and it has rejected and it has given evidence for that. Now people might -- still you don't believe it, just go further. This is where that information are available in the document. What we got is only like 1 reference. And similarly, where this is like some 28 pages document, the information available on the 26th page. We show exactly where that information came from. For example, if someone want to see. Here, there is a note like agreed client has a medium attitude risk and given SRA willing to play high to high equity content. So all this is recorded. So everyone knows exactly where that information is. And just if you think about a humans, how many people for how many hours they need to coordinate across them, across various systems to achieve that. And this is -- we are able to do that in less than 20 minutes. Now going back to the -- what was the outcome look like after we implemented the system, this is the reference that I showed you, right? This was a state before it was everything is manual. Now everything that is manual, now we replace that with an AI agent. And in fact, we have an AI agent, which is also kind of recommending the decision, but we are leaving the decision to the human expert to look into it. And what was where we -- they have some number of an FTE for each of this task. Now they really don't need any FTE for that. And actually, the capacity could be infinite. It's up to them because we can scale the compute. Similarly for triage, they don't need anyone and actually, the capacity could be actually scaled to the infinite. And then for the case handling, this is where we left the final decision with the human. Actually, with the same capacity, they can do the 10x of the work. So this is the kind of an impact that we have provided to our customers. And this is just 1 use case. And we are already many use cases, as Arun has mentioned, in the insurance underwriting space and the ESG or nonfinancial risk. In fact, what we have done for ESG now in the governance risk and the compliance space can be applied for any company. It is basically a corporate intelligence, then account payable, the trade finance. These are the various use cases where we are already making a huge difference in terms of an operational efficiency and the newer opportunities. This is just a start. We are applying it across the various Intellect products. We have identified hundreds of agents that can be implemented within our Intellect products. Over to you, Arun.

Thank you. Perfect timing. I think 4:42 and exactly less than 25 minutes to explain such a complex topic was quite interesting, Deepak, for looking at it. People may have many questions. I'll put on the chat box if you can ask that question on it. In the meantime, I'll ask Banesh, how are you trying to monetize, Banesh? What are the business model through which Purple Fabric can be giving the value to the shareholders?

Yes, Arun, I think I'm just going to touch on a couple of different thoughts and also the areas that we are focusing on for our clients. So firstly, I think as we've already seen the platform capability, the operations area of a financial institution spends a lot of money. And the work we've been doing and investing over the last many years has actually helped us identify certain specific ways in which AI can be implemented for our customers. Clearly, we focused a lot on the U.K., U.S. and the developed markets because their cost-income ratios are very high in operations over there. And the other important thing about the platform, I just wanted to state one more thing is that we bring a lot of experience in BFSI. I think it's very critical because when we use AI, we have to ensure a lot of focus on trusted data, the audit trail of what we do from an AI perspective, all of which Deepak has covered and something we have considered very carefully when we built the platform. So I think it's very important to understand because a lot of people are doing investments in AI in very generic cases, but I think our capability very much focused on ensuring that financial services is well managed from an AI perspective, I think is very critical. Now to Arun's question, we have 3 broad categories of how we are monetizing the system as such, Purple Fabric. So Purple Fabric is one where we basically offer the whole platform that you saw so that the client who's already been working on existing various kind of use cases and AI initiatives can actually bring those initiatives, can take some of our use cases, can build their own use cases for which we actually train them through an academy of Purple Fabric, where they can get certified to actually build use cases. We provide them the support required. They can take various capabilities and tools that you saw on the platform that we have built. And these tools are continuously evolving as AI is evolving in the market. So we'll keep looking at new capabilities and enhancing existing capabilities. Deepak gave an example of how we use multiple LLMs. We benchmark, which LLMs is most ideal and also various sort of AI ethical focus around data because data is at the heart of it. And I think all the technologies, the first 4 or 5 of the technologies focus a lot on data. And I think that's very important for us to consider. So we actually offer the platform for them so that they can govern multiple AI initiatives. So if you were a CIO or you were a business manager or a CEO, when you have multiple use cases across financial services and across the various platforms that Arun mentioned as part of our eMACH.ai frame, the books that he was talking about, whether it's wealth, whether it's insurance, whether it's banking, whether it's lending or any of those areas. As a CEO, there's a lot of pressure by organizations to capture AI capabilities, which are very generic today. What we bring is BFSI-focused platform that they can actually govern multiple AI initiatives, ensuring they use the best LLMs, they ensure that the data is trusted. It is very well trailed from an audit trail perspective so that we can follow all the risk, compliance and regulatory requirements for the client. So this is what we basically offer as a platform is 1 construct. The second construct is the example Deepak gave of use cases. We have multiple use cases in different functional areas. We take those use cases and we offer those use cases for a customer to be able to actually implement it in their technology environment or their technology ecosystem in the bank. And we, therefore, have these use cases, which we normally charge per use case for a customer. And the third is where we actually embed some of the AI capabilities that you heard today into various services. For example, in wealth from a relationship manager perspective, you could have actually embedded certain capabilities for the financial adviser to be more efficient in the way they would deal with customers. So there is a lot of embedded AI and various solutions for our various platforms as eMACH.ai. So there are these 3 constructs. Most of our customers come and offer this as a subscription type of pricing rather than the traditional license type of pricing that we've had in some of our products. So this is what we actually, I think, right now are focused on. And this is evolving very rapidly. And as both Arun and Deepak mentioned, a lot of people are doing stuff in GenAI asking assistance questions of data, but our ability to replicate and human operation setup through multi-agents from an AI perspective, I think is a unique positioning. And I think it's effectively going to be -- going to grow into be the sort of a replication of a different workforce within the bank that will complement the human and change the future of the way work is done.

Thank you, Banesh. So in summary, what you're saying, there are 3 business models to monetize business model 1. BM1 model is around Purple Fabric as a platform like Palantir offer a platform, they charge $3 million per annum per customer for subscription. We are looking this to be in the range of $1 million a year. A customer can purchase it. So that's the first model, which is there. Second model, which is there is, build up an application, specific application like invoice processing application, AI-based invoice processing application, which is iAPX; underwriting application like Magic Submission, which we have launched in the U.S. We have more than 20 clients using the complete underwriting platform, which is an AI-driven platform where we are charging from them on basis point. For insurance premium they are underwriting on our platform or we are using claim management, which is a per transaction -- per application file or per complaint, which could be -- even if it is a per complaint close to GBP 100 per complaint. It's a kind of if 10,000 complaints or 100,000 complaints are there, looking at the market complaints in the marketplace of misselling, there is a third business -- second business small case. So these are the cases within business model 2, there are 4 business cases or 5 business cases we have already put in the market. Each use case, we are looking at it that each use case should be a minimum of INR 100 crore revenue in the next 3 to 5 years, each use case can bring. And business model Canvas model 1, which is around Purple Fabric as a platform, if we can reach out to 50 to 100 clients in the next 4 years, that's the kind of objective that we are looking at it. And third business case, which is there, which is embedding our eMACH.ai in all our products. So our product value, which we are selling, we can sell at least 20% at higher premium to other market players using the model Canvas 3. So these are 3 models, BM1, BM2, BM3 to capitalize on it. So with this, we are quite bullish about this entire IntellectAI platform. Investments are quite high from the R&D perspective and POC perspective. So there are 2 kind of investments we are making. One is on research. Second on doing the proof of concept for our clients. So as of now, that investments are going in proof of concepts because whenever we want to implement certain thing, as Deepak puts it, the claim investigation system is the first of its kind, which has gone live in the world, where it is using enterprise structured data from multiple databases and unstructured data from the market perspective. Nobody has currently using the 2 systems of structured data and unstructured data and multiple system -- for multiple systems to solve a particular problem. What he is showing using 13 digital agents -- digital experts. And I don't call it agent because they are experts. So there are 13 digital experts plus 6 LLM models. The interesting thing is 6 LLM model. So for one claim management system, we are using 6 LLM model and system is going -- is smoothly there. It's diversing and saying, what is the billing for OpenAI, what is the billing for Bedrock, what is the billing for different LLM. And these are the important things which we are very uniquely defining and getting a good traction from our partners and partners are now started. So we have one partner who started to doing the academy, he wants to train many people in his organization. So we have a line from January, February, March, we'll be training our partners who can sell to their customers to reach extend our outcome. So at this point in time, maybe I will ask the questions what you have?

Now we open for the questions. [Operator Instructions] Arun questions are there in the chat box also, let me take that first. [indiscernible], please read the question.

The first question is, what is the unique value proposition of multi-agent platform against any other Gen AI platform?

That's what we explained. So I think this has been explained very well. That the multi-agent is 13 agents to solve one problem. We go to next question.

How to manage trust and explainability in platform?

I think so there are at multiple levels, not just as one dimension thing because throughout that, what is required for typically for data privacy perspective we will ensure that, I mean, from a there's enough -- there is a masking capability for any PII data and there is a toxicity capability to ensure that the questions are the right and sensible kind of a thing to the -- actually adhering to the standards of GDPR and everything is one part. Explainability, you have seen that, I mean, every single thing, what the agent does is absolutely available and to the very -- we ensured that actually the trust is visually can be felt also. We spend a lot of time to really understand what should be the human AI interaction should look like and how they should be able to interpret what agents are doing. That's the reason we have decision logic and everything, action and everything. So this helps in terms of having the right traceability, right expandability, not only in the interest of an organization, but also an auditor and the regulator. So this is what we have put in place.

Okay. So the next question is, are the target clients even ready to adapt to AI agents actively. A lot of current adoption is let's do a small pilot and we'll see later.

Exactly, that's what we have demonstrated in U.S. there are 20 clients are already using it for full-fledged underwriting submission these agents are being used. We started this turning into 2021 today. And we have a pipeline of more than 100 leads are there who are on the pipelines where we are doing the POCs with them. So that they can onboard. In claimant, one customer is fully on live, ESG, customer is fully on live. So that's what our uniqueness is we are in action mode, not only in POC mode.

So just to add, right, most of the clients go through a POC. But once a particular use case in the BM2 use case sort of model is available, then they can actually start consuming this much faster than before. But what really happens is the client has to -- is still going through the stage of accepting AI into their environment. And I think that takes a POC, which we do, we do -- we are doing multiple POCs across product lines today. And I think that's the way the model is still a point where everybody will accept it, and then you can scale it a lot faster thereafter.

Deepak can regulatory requirement on how AI is implemented in BFSI sector detail some of the solutions you have built?

I think one other thing is we are very actively working with various regulators through the sandboxes in U.K. and other places and also trying to ensure that the ISO standards like a 42001 and everything are applied. So that it is completely in the context of regulatory, for example, in both U.K. and U.S., a lot of regulators are helping with innovators like us in terms of how to operationalize. So at least where we are operating what we are doing, we are kind of seeing some help. And we also constantly focusing on what are the new framework they're releasing new regulations, they're calling it out for us to adopt.

Sorry. Arun, just to add to what Deepak is saying, right? I think what we also do, not just all the discussions with the regulators, but we are actually finding ways in which we supplement the human beings. Sometimes people believe AI is completely autonomous. Not only do we ensure that the data of the bank is something that we keep secure within all the security regulations of data that exists, but also the work that we do actually enhances the human's capability to be able to do processing. And therefore, it supplements or augments the human's decision a lot faster. And it's not something that, from an AI perspective is completely doing things which eliminate the control of the human in this case or the operations process where it will end up doing the wrong things of the wrong financial transactions for the customer.

And also, Banesh, just to add to you, right, the use cases that we have operationalized, the biggest challenge for the companies is they don't have enough expertise available to deal with the demand that is coming on their way. And if you look into the use case, I said the biggest challenge they have is they're not able to meet SLA. They were adding more and more people that is not helping them. So if you especially look into the risk and compliance space. There is a huge shortage of an expertise, and this is where the agent helps in terms of giving the right coverage and also reducing the cost or optimizing the cost.

Yes. There's a question, how is it to some of your leading competitor to replicate this kind of solution? I think we have -- almost 10 years, we are working on a replication is possible, but nothing can be seeing that we have a unique solution, which cannot be replicated. We have applied a few patents on it, which can protect us from the replication. But more than that, the I call it basement 1, basement 2, basement 3, basement 4. If our solutions are as defined as 97% accuracy or 98% accuracy, somebody has to reach to 98% accuracy takes almost months and months of jobs. So at least it is a 2-year, 3-year mode available to us. On applying in banking financial sector, which I was talking about Black Book. This Black Book gives us a uniqueness because within each area process, we know but difference which is there. What are the cumulative investments? Cumulative investments in this business is over a few hundred crores, which we have made the cumulative investment in this area, something is available in our balance sheet capitalization, something is available in U.K. subsidiary. So it's a few hundred crores of the investment is already made in this area. With platform potentially replacing human operations, are we facing resistance from the employee from our customer? Deepak highlighting very beautifully that we are not able to solve the problem fully. The regulatory questions we are not able to solve. Today, it's not a question of employee, the people are paying fine in U.S. and U.K., they are being fine to the government because they are not able to solve problem. This is a problem solving they are using it with both hands. So there will be some situation where employees may feel constrained or they may feel then with it. The question number two, Rahul Bhansali, you are asking what is the biggest risk we see in our ability to monetize the Purple Fabric. I think the risk or opportunity, we look at opportunity whether this opportunity is a INR 1,000 crore opportunity or INR 5,000 crore opportunity, that's the dilemma we have. INR 1,000 crores, we are sure of it in the next 4 years, we were able to capitalize on it. INR 10,000 crores could be INR 5,000 crore opportunity for us at a margin of 50%. That's what is our current challenge right now, which Banesh, me or Deepak spending a lot of time in the flight to look at it, how do we make it INR 1,000 crores to INR 5,000 crores journey anywhere in between. In terms of complexity of implementation, I think this is where the beauty of the platform, Deepak has done a phenomenal job with his R&D team, you can do it. So sometimes if you want to take it. We can give a session to you. You can build in on it. So this is our...

Rajesh Sharma, which client does not use any of the core products has signed up with this platform? Which customers does not use any of the company's product has signed for this platform?

We have one which is right now doing a POC with us. And I think we've got a big pipeline of POCs going on with clients. Many of them are not clients of ours at all.

Manish's question, a beautiful question. How well does Purple Fabric integrated with legacy infra of the banks in U.S.? This is a beautiful question because that exactly the structured data access which we are able to provide with generative AI of many players are not able to do it. That's what he explained.

Yes. And there was this question the other day, also somebody asked me, all the platforms that we have built and clients have our platforms. This platform actually is an operations transformation platform that connects to those platforms to pull data along with the unstructured data from documents puts it together through the agents and the multi agents to do the actual decisioning involved in supplementing the human to achieve an outcome. So it is actually complementing existing platforms by pulling data just like a human would log into those systems and pull information and take decisions. And AI agent is helping the human doing it more efficiently and faster.

There's another interesting question is that content creator, the cost of building software using. So how the software engineering cost has come down? I think we are doing it. We have stopped hiring the people from October onwards. Our headcount is constant, and we are now -- will be in the phase of using the Purple Fabric for our internal automation substantially. How much is incremental investment is needed? I think we need to continue the investments. The change is every week changes there. So we have -- our more investment is in GTM investments will be much higher. We are creating a good marketing department in U.S. and U.K. These are 2 countries we picked up and the first phase of going there. Does Purple Fabric improve conversion ratios? I think the question is just a question of each lead generation, it's improving a lot. I think each quarter, we are signing around 8 to 10 deals a quarter. They may be small POCs, they may big POCs, which may not reflect on the revenue as much in the -- each quarter if we're looking for the revenues. But each sign up customer has the potential to grow the revenue based on testing the our products. So each customer -- some of the customer can become almost like ARR of $1 million ARR to $3 million ARR. Those are the possibilities of these large customers who are under pressure for regulators to solve their volumes. $3 million for them is a peanut for the kind of a penalty they have been charged by the regulator. Any other question? Praveen, you can look at it even raising the hands if somebody wants to ask a question.

There's one gentleman Neel Chhabra from SOIC. Neel Chhabra, please unmute yourself? Naresh, please unmute him.

Neel, you may ask questions, you are unmuted. I can see it is not mute from your side. So you can ask the question. Okay, Praveen, you maybe some technical issue you may go on to the next one.

Okay. I think -- Vimal Kumar already has put it in the chat also. There no other hands there.

Okay. Yes. 2 participants raised hands.

Yes. Vipul, you may ask the question.

Yes. Am I audible?

Yes.

Yes. Yes, Vipul. Please go on.

So sir, you said that the revenue potential is INR 1,000 to INR 5,000 crores. So that is spread over how many years?

5-year. 5-year period.

5-year period. So can we safely say that from next quarter onwards, we'll be clocking revenues in hundreds of crores for these products.

Hundreds of crores, we are saying it is accelerating revenue, not a quarter-to-quarter revenue. So please don't expect anything significantly change in quarterly revenue because when we are doing a POC, we are doing a first, 100 complaints to be handled, not 1,000 complaints. But as soon as it's accepted, it can become 1,000 complaints a month. So that is the time the revenue buildup will start.

And what will be the margin profile for these products as compared to our other products.

These are completely different lines of business that other products are in transformed enterprise space. This is in the transformed operation space. So they are not contradicting to each other.

No. I mean margin will be higher or low?

The margin will be -- margin definitely will be higher, and this is a very distinctively different AI product today. AI margins are still on a higher side. So that will be higher than the existing products.

There is no other hand raised.

The time is also over. We had a 1 hour session. Okay. So we can close the call. Any other questions there?

Any other questions are there, please raise your hand. So that we can...

There 2 questions, Praveen from Pratap and Nishant. 3 now.

Pratap you can speak.

Yes. Am I audible?

Yes.

Yes. So can you explain a few other use cases that can be like one is that customer resolution you have explained, like what are the other areas or what are the other use cases you have currently developed or in future, what are the other areas where you will be -- you think you can have developed application basically?

There are 68 use cases we identified in banking. And we are going step by step to look at like trade finance business, the entire use cases is in trade finance business. The entire use case in lending underwriting solutions, the entire use case in regulatory management of the reporting, which is there. So there is a 68 list of expert -- digital expert use cases, we have a book available, which we are using for selling in entire black book space.

So out of the 68 like for all of this has developed or like how confident we are with how many cases and like what is our plan to add up basically.

Yes. So that's what our -- depend on our capacity to go to market. So we are doing 1 -- at least 2 use cases per quarter to go live. So we are just picking 2 at a time and then making it perfect and then go to market on that. We need to set up a full team around that particular per sales team to sell this product like claim product, which is for the U.K. market, now it can go to U.S. So now we are setting up a U.S. team for claim investigation team. So 1 quarter, it will take to set up a U.S. team. U.K. team is ready. U.S. team will be set up in next quarter. So it's a 3-step process. First is building the agent. Second is do a first live sites. And then third cycle is for go-to-market.

And just to add to that, this is not like a coding platform. We're actually configuring this. So the third option is to get the partners trained so that they can actually create the use cases for their customers or cases that the customer already has and enabled them on this platform. So it's not a traditional coding type platform as the earlier platform.

And sir, what is our plan for the go-to-market like we want to have some partners or mainly we want to have your own?

Yes. So this is -- we will be doing with the partners. So some of the use cases are taken by consulting companies. So these consulting companies of Big 4, which I mentioned, or big 5 we have mentioned, not big 4. We added another company called BDO also, along with the 4 partners. So 5 companies which are there, we are going -- they are taking us to the customers.

Okay. Sir, just to understand like traditional IT software companies, like can they be our partner or there is some concern...

They are our partners. The traditional IT companies are definitely our partners, and there are 2 IT companies which are training their workforce using Purple Fabric.

So there is not much conflict of interest with them basically.

Initially, we felt is a conflict of interest when we launched it. And they also launched it. But when they saw our platform, they themselves backed out that they don't have capacity to do such a deep R&D on these kind of products, particularly in the multi-agent side.

Next, we have a question from Nishant Gupta. Nishant, you can ask.

Am I audible?

Yes, Nishant we can hear you. Yes, you can ask your questions.

Thank you for this interactive and deep date session. Sir, I had actually 2 questions. One is this product definitely has a lot of use cases, right? But -- and you're targeting the back office in the operations team. But will this actually lead to termination of the workforce on that front? I mean, you briefly touched on that, but I just wanted to get a more realistic sense that will people lose on jobs when you start implementing because eventually replacing the agents with technology. So more from a fairness point of view, right? If you can comment on...

So I'm not a trump to comment on this fairness point of view, but I think AI will definitely take away a lot of routine jobs. People have to reengineer themselves within our software engineering business, I believe that 30% of our engineering service industry will lose their jobs. In 2026, our software engineering industry from India has a serious risk of losing the business as well as jobs. So it's a fair question. There's no answer to it. By not doing it, it doesn't solve the problem. If technologies have come in, those technologies have to -- you need to create a new kind of a job, more fair jobs, more sustainability element. So if the job can be put on ESG or compliant, more sustainable thing. I think world will be better placed using AI.

This is a bigger moment than the Internet when the Internet started and new kind of jobs will arrive, but there is going to be a transition of skills and changes and different kind of jobs. But this is not like -- Arun said this is not something you can stop. This is already in motion. If changes are going to happen, and it is going to change the way we operate.

Got it, sir. Sir, my next question is that Google Gemini, they are also innovating a lot. They are also conducting sessions where they're telling about what all capabilities they are developing. You mentioned that there are 4 levels of basements and anyone would take 2 to 3 years kind of a time line. But considering Google is there, Perplexity, which has recently received a fund raise of $500 million, valued at $9 billion. So is it fair to come in, in light of these events that you still have a 2- to 3-year kind of a moat in this?

The difference is we are in applied research. We are not in a foundational research. But Google and Gemini is spending money is in a foundational research. We are leveraging them to solve the specific business use case which is not their competency. So what they work on is technology competency, we work on applied banking competitors. That's what Banesh tried to highlight in the morning -- in the beginning of the session that our differentiation is which particular smaller space when you are issuing an LC, where is in LC, which is a particular context of LC and which business rule need to be applied for, that is not known to Gemini or any of these products. So that's where the moat of 2 years is there.

Just to add to what Arun said, right? If you see pretty much from the level of intelligence, this narrowing down between the top players. Now even if you see OpenAI, right, most of their recent work, they're all trying to figure out how to verticalize their intelligence and what kind of user journeys, what kind of a problems they want to solve. So that's where being in a BFS space for more than 3 decades gives us that very unique insights on how this can be applied and how it can be solved to the last mile better than anyone else. That's where our differentiation is.

Got it. Got it. Final question from my side. This is more like from a long-term view, you are targeting to capture let's say INR 1,000 crores to INR 5,000 crores of a market right? But let's say, if a bigger player comes in offers to buy this particular product segment from you, would you be open to selling it at a particularly higher value? Or you want to keep it in-house and keep kind of getting a recurring revenue.

Was -- you have seen the Intellect, these investors do come in with always some purchase options. We may do some action on this company, some corporate action on this business. We can think about something of that nature where we can leverage it better subsidize it something. We don't know as of now. We're just discussing it, how do you really capture the value of INR 1,000 crores to INR 5,000 crores journey. That's what we are discussing in the board meetings that how do we capitalize INR 1,000 crores, we are sure of. But INR 5,000 crores, if you have to take some action, we need to -- we are exploring.

Got it. Got it. And you're targeting not the Indian market right now, only the U.S. and the U.K. and all market right now?

Yes, that's right.

And any particular [indiscernible]?

Pardon? The question?

So why not also target the Indian market where the BFSI sector...

There are Intellect tactical clients which are critical to us, where we are talking in other geographies to those clients. But our main focus of scaling the business is going to be in the U.S. and in the U.K. to begin with.

We are doing an APX model like voice, magic invoice, some small way Indian market. We are using it. So I mean, big customer for India. We want to look at it Central Bank and as of now, India rules are not clear where they can use where they cannot use Central Bank has not issued any notification. How can they use AI or not. So we tried doing in India. I think we'll take it up when some clarity will emerge from Central Bank on the other AI and which use cases can be there. So as of now, we are waiting for the AI policy to come from India and then we can use in India.

Thanks, Nishant, We take from [indiscernible].

No, I have a question. See, this human and agent is a wonderful collaboration. So this produces a tremendous output. What -- can you hear me?

Yes, we can hear you.

So it's been -- it's delivered fantastic optimization. It's an optimization tool, it's an enhancement tool. It's done a fantastic job. So how do you measure this going forward from use case to use case from application to application, the productivity of this in combination when they work, how do they work? And how do they get technology and domain specificity together. And then how they'll get the predictive ability of the platform because I don't see much of a productivity in this application?

No. So I'll just take that question, Arun. So I think you got the example from Deepak on actually showing you the outcome from 50 days to 20 minutes, 30 minutes for an investigation. You got the accuracy, you've got the cost benefit of that speed. So all the quality dimensions that would be there any kind of quality management on accuracy, speed, cost, and efficiency is all there to actually measure use case by use case. So the client actually can see that. And that's what we work with the client as part of the POCs to do.

My question is if you -- what you're saying is right. If, let's say, you would spread it across 10 use cases. So is there a productivity improvement? Is there an output efficiency coming up in each and every -- I'm asking -- I'm trying to understand let's say, it does 10 applications, 10 projects. At every stage, because of the agent involvement, the combined productivity should get better. The output can get -- more importantly, the output get better. The result that you are seeking it gets better. So does that happen?

So Deepak, what is the benchmarks you are looking at, he's asking the question that you have a benchmark metrics which you're showing. You're not showing it. Otherwise, there was the accuracy.

I think if you see, right, we are in a very interesting phase where we are actually initially, right, typically because the people see initially as a more of an automation journey. That is a reason a lot of metrics are around the operational efficiency and the productivity and everything. That was the first part. But if you see right, this is where also we are working with our customers to also see the different dimension because automation is largely the FTE saving and those kind of metrics. But what we are also doing is right, like the use case that we developed for the complaints, right? Now we are using it for a compliance means what is the cost of getting the decision right and how we are reducing it? It means, for example, if you take the banks, right, the real value is how you can reduce your NPA. How you can deliver the loan so that your the chances of you're paying and all. So that is where what we said is we are not only helping in terms of improving the bottom line. We are also helping in terms of improving the top line. So that the misselling can be stopped and all those things and all. So we are actually -- because that's the reason we are also helping and educating our clients to not to see it as an automation way, but also see it an AI way. That was the number 1. Second thing is, in fact, we are working with some advanced research universities to how we can benchmark the competency of an agent across the human because right now, from a human perspective, right, when you onboard someone, this is the competency, this is the salary and all. Now we will be continuously increasing the competency, and that also will give a differentiation in terms of the price point and also the quality of decisions and all. So this is the various ways we are helping the banks also kind of creating a new ways to benchmark on this.

So I'm using Puranik for my software testing and software compliance testing of security. I'm using the Purple Fabric for doing it, which was -- it used to take 4 weeks' time to have a security leakage in a software or something. Today, I'm using Purple Fabric to check gates of entire my software quality. So we are going to leverage almost 20% headcount reduction for the same job in next 6 months.

Wonderful. So Arun, Banesh, can you give us some use cases on underwriting because underwriting, it's everybody is concerned that well lend loan is well recovered. So initially, you give the right lender, you don't have to put in too much effort the borrower. You don't have to put in too much effort in collection. So how will you help in that process?

Banesh is running Citibank operations here.

Underwriting, whether it's loan underwriting or insurance underwriting, for underwriting, you normally get unstructured documents that are -- that we call Magic. So we have a product called Magic Submissions, where people submit documents and this can be very complicated documents with great data of detail. So we actually, through our technology that you just heard today, you can take that data, you can ingest it, whether it's unstructured or structured. We can -- we also have a data platform of our own for underwriting, where you can triangulate this combination of data. Now what are you doing in this model? You're actually going after BPOs, right? BPOs do their jobs operational. Documents hand to them, they data enter, a few fields in the document and then that underwriter uses those fields. But what we are actually doing is through AI, not just collecting all that information into a knowledge database of information for the underwriter to ask questions, but we will also be creating underwriter agents to help the underwriter become more efficient as we go forward. Those are examples on underwriting, and this can apply across the board for underwriting.

Can you give some examples of that? So what are the fields that you have eliminated? What fields you got more efficient than what it is?

So if you -- I mean, we don't need to go to the field level, but there are a lot of data that you do not capture from submissions, right? Data enter a few of things, but actually the other data is kept on document folders. And then the operations people actually look at document folders and log into systems and then do the decisioning, you're actually by extracting all the information from these unstructured documents, along with the transaction systems, you can actually get the AI agent to help the underwriter become faster at taking complex underwriting decisions. For example, in the U.S. you wanted to promote insurance on 50 Marriott properties, or something like that at world time. That is very difficult to do as a human underwriter today.

But how does the predictive underwriting engine work?

No. So it's not about predictive underwriting only. This is actually about helping the human with information to help them manage underwriting and give them things that they will have missed as humans with more information to triangulate and take decisions faster.

Just -- Puranik, just to add to what Banesh said, right? What we do is extracting the data is world thing, but what we do is we really enrich the data also with the multiple sources and then triangulate it. So that way because as a human, right, the challenge is most of the time with all due respect, depending upon the time in a day, they end up taking a different decisions. And most of the time, they are biased with the historical view of that particular entity or the particular organization or whatever it is. So what we do is we completely irrespective of the size of the data or volume of the data, we triangulate the whole thing, and then we will identify the blind spots at every single thing. So if you see, right, what is the value that we are creating because if you see the one that I showed, we also kind of a recommended the decision on the various things so that we kind of do the red flags. At this moment, right, it all depends upon the underwriter he or she, what they view about. There is -- the biggest challenge in underwriting is the standardization. Every underwriter has a particular view irrespective of the clear guidelines within the organization. So now what we do is we actually ensure that those guidelines are properly applied irrespective of the volume of the data and also when there are gaps, we also have an ability to enrich that by connecting to the public or private sources, and fill the gaps also. The coverage also will increase. So the biggest value that we bring in is ensuring that we're able to aggregate triangulate. And then also recommend.

Enrich. And one more process is continuous monitoring. Every 3 months, I can go to website and check it on the status of the loan, which is a continuous KYC and continuous loan. So I can -- I go every month to the website of the customer and see whether there -- or any results there, any new results have come in, embed and along with it, triangulate it again.

Exactly as Arun beautifully said, right. The biggest challenge people do this only when they disburse the loan. They never have a capability to do it on a continuous basis. So that way, the predictive part what you are saying, right? This is what we can bring in terms of predicting the chances of a default and a frequency that you can configure for that particular corporate.

One other question I have is how do you measure an agent's performance with reference to its augmenting the human skill.

Yes. So for example, what the use case that we have done, right, initially, that was the exact dilemma. So we kind of actually led the same use case because in a model office, that's how they benchmark. And what is the beauty is, right? It kind of opened up a lot of gaps because typically, when humans make a decision, very rarely people will check and all. Now we have a way to compare on the performances. We -- the way they got the acceptance is actually and when they ran on the use cases, the case is where already the decisions were made, we identified how many cases actually they wrongly paid back for the wrong reason because the human hasn't really validated them. So the sign off on this happened again as the human benchmarks, Puranik.

Thank you, Puranik. I think we need to just close [indiscernible].

Thank you, Puranik. Arun, we have a last question from Mr. Vivekkumar from Bestpals. This is the last one. Vivekji please unmute yourself.

Can you hear me now? Can you hear me?

Yes, Vivek go ahead.

Thanks for making us understand how AI agents can be a big opportunity for Intellect, but in terms of changing of profit pools of the industry, like what happened in the content creation has moved to aggregators like that? Are you seeing any trends in IT Arunji? Like I understand how AI is helping Intellect, but in general.

It will happen. It will happen. If you're going first principle thinking, a lot of aggregation will happen. I think Indian IT industry is still sleeping and too much disruption will happen on the service company. It's going to happen. It's bound to happen. As of now, we are still not acknowledging the capacity of AI, what has done. I am pursuing it from last 10 years, I know what the power is and how it accelerates. Completely, it's a tectonic shift. It's a tsunami. It's no less than tsunami right now. But tsunami takes 3 years to come to a surface when it happened from the -- where the attack happened. It will take 3 years for India to understand.

Thank you, everybody, for joining the call today. Still, in case you have any more questions, please do write to us. Thank you. Now you can log off.