AI-Driven Lending: Moving Beyond Chatbots to Autonomous Agents

Held on – 22nd April 2025 at 11:00 am IST

Panel Members
Ajit Joshi, Partner Account Manager, Red Hat APAC
Sireesh Patnaik, Chief Product & Technology Officer, Pennant Technologies

Ajit Joshi

Good morning, everybody. It is a pleasure to welcome all of you here on behalf of Red Hat and our partner, Pennant Technologies. Red Hat has been a leader into the overall banking vertical. Most of the banks are our customers, and we will see more about it as coming point.

Today, we are, having an inflection point in our industry. There is a in the overall industry, there is a rise of AI. So similarly, most of the banking institutes, especially during COVID, started working with chatbots. So, since a lot of people found it very difficult to reach out to their offices, lot of them started to, for customer service started offering, chatbots. They created some rules-based systems, what they used to call it as autonomous agents, where you would feed certain rules, and those rules would ensure that how do we work on the on the banking operations and then if it gets stuck somewhere, it would go to a physical agent, and then the agent would take it over. However, that is what not a lot of customers expect. They expect the system to be very intelligent, understand their data, not continuously ask the same information, and if that information exists in the system, provide them with some relevant and transactional information, if that is what is possible. We are living in a situation where if I am spending thousand rupees on my credit card, the banking AI should be able to offer me a loan of 1,000 rupees. We have not reached that level yet of automation. But that is what automation any banker would aspire to have. If you’re there, maybe few years back, whatever I am saying would appear like a science fiction, after some time when pennant will join you, they will prove that most of the things are possible. Some of these things are already in production today. So, it’s not like AI driven lending system. Say it’s like, sort of evolving, sort of autonomous, there are autonomous agents who are able to, there would be a day where those autonomous agents can disburse the loan to some of the people. Let us go through this together.

Just a brief introduction of myself. My name is Ajit Joshi. I head, the partner business for AI vertical in APAC. Along with me, my colleague from Pennant is there, Sireesh Patnaik. He will talk more about himself, within the next seven, eight minutes. He’s the chief product officer and the chief technology officer for Pennant Technologies. Two senior level roles into one person.

Typically, some housekeeping notes, so please be in listen only mode. I think that is disabled on this platform. If you have any questions, use the Q&A tab and for any kind of queries, comments, you can use the chat box if it is available, or you can type in the Q&A tab itself. If you’re not able to hear me, please try to increase your speaker volume. Most of the time in on 24 system, I have seen that the speaker is kept very low on the laptop. We may have to do it. You can close, you can close, your other browser windows. So, some bandwidth is required for the video conference. Close your browser windows, and you’ll be able to you’ll have a better experience.

Very brief agenda. We will talk about what is Red Hat and Pennant. I will explain very briefly what Red Hat AI is, and then we will get into the core of presentation where we are calling, how do we use AI in the overall lending business. Very briefly about Red Hat. By 2019, our revenue was $3 billion. So, in last four years, we have moved, comfortably at double digit, but we are the first company to reach the $3 billion turnover in 2019. We are a way leader. We do more business in open source, maybe three times than our next competitor. India, we have facilities, software development facilities in Bangalore. So, we are as much as an Indian company as anybody else. So almost, 25% of our global staff is in India. This is our customers, landscape in the BFSI industry. And these are all referenceable customers. For every customer, I can give you a case study if I had if I have unlimited time. But essentially, if you look at it in India, we have got, all the public sector banks as our customers. Most of the large private sector banks, most of the new age banks, all of these are our customers. So we have got, great, inroads into the FSI vertical overall. What are we and Pennant doing together? And you need to understand that, whenever somebody asks me that what does Red Hat do, I always say that Red Hat powers your, payments, UPI payments which you do. Or if you make a 5G call, Red Hat fires that 5G call. If you file taxes, Red Hat powers that taxation software. Or if you’re flying, Red Hat powers your tickets. So, essentially, you consume services which are partners like Pennant built on top of our technology.

Today, it is largely about, modern applications, which work on the cloud, on the Internet, on premise hardware and things like that. But eventually, it will be, everywhere and going, beyond, whatever the latest technologies are coming. So, on the right-hand side, I mentioned all other services where we have applications. Slowly and steadily, most of these are becoming intelligent. That is, they are coming into AI.

What is Red Hat AI?

Very briefly, Red Hat AI is about, we have an open shift AI platform. We have a Rhel platform. So Red Hat platform is another add on infrastructure, it is not something different than what do we do. For example, we would have, different platform services, we would have developer services, infrastructure services, AI services are add-on to this service. What does it mean that if a customer is implemented some solution on the Red Hat platform, which is a Rhel or OpenShift, then it is very easy part for that customer to go to AI.

What does Red Hat AI do?

It would tune, deploy, manage, and maximize. So, whatever you have your own enterprise data, it would help it to get tuned with the models which you are wanting to use. It would deploy the model. It would manage the entire life cycle of the model and make it very easy for your team.

Coming back to and this is again, not like an isolation thing. These are all our global partner. We have got a large global partner ecosystem, and then there will be APAC partners like Pennant who will be part of this extended ecosystem. So here we are listing all these partners who are supporting Red Hat AI. But then, of course, partners like Pennant are also part of it who would bring vertical business applications on top of it. As you see, these are all technology applications.

One example here is, we have got a bank in Asia who has already bought OpenShift AI. This bank is owned by Emirates Group. So, it is not like it is only a Western concept, only in the US or Europe. This is very much coming to our place in India. This is what, I have sort of represented how this whole thing work. Currently, if you, as a customer, have implemented something on OpenShift, it is very easy to go on OpenShift AI. I would now hand it over to Sireesh to explain how their system would work on this integrated platform. And then, start with explaining what is brief about Pennant, or how do you do, what is how do you see you working with Red Hat and then take to the core part of the system. Over to you, Sireesh. Thank you.

Sireesh Patnaik

Thank you, Ajit. Very good morning to you, Ajit, and a very good morning to all our, audiences who have joined over here. I see some partners and some interesting, names over there. So welcome aboard and thank you for joining this webinar session. Very quickly about myself, as Ajit mentioned, my name is, Sireesh Patnaik, and I look after the overall product and technology, aspects for Pennant as the chief product and technology officer and before we actually take a complete deep dive into the entire agentic aspects and in terms of what’s happening in the in that particular horizon today, very quickly give a brief of Pennant as an organization. I promise I’ll not take much time over here because I know we are waiting to get into the core over here and then talk about more of the agentic aspects. So, Pennant as an organization was conceptualized in about 2006, primarily present in The Middle East. We were working with banks in The Middle East on core, you know, banking algorithms over there in the payments sector and some of the core systems.

From 2012 to 2017 is when we started our presence in India, and that is when, our flagship product, which is, PenApps Lending Factory came into existence, which kind of, covers the overall lending life cycle. I’ll talk a little bit more about that in a second. From 2018 to 2023 has been our rapid growth, kind of journey where, we started engaging with a lot of customers in India. In the journey, we also signed up with five of the largest, NBFCs in India, the fourth largest bank in the world, and an international digital bank overseas as well. And, in terms of the entire vision that we are having as a core technology product organization, we and as our tagline signifies to be future ready, we have been continuously working on the emerging technologies, and 2024 and beyond focus is also on AI and GenAI.

A lot of the engineering aspects that we are doing to scale and to ensure that the modern aspects for the lending requirements across banks and FIs with the amount of digitisation that is happening is taken care of. There are obviously a lot of tech upgrades and, some investments that we are working on.

Quickly about us. So, I covered some of these. I mentioned we have about 60 plus clients currently, and 80 million number of loans that are running on our platform today, more than about &90 billions of AUM that is being supported. 4 in top 10 largest housing finances are running on the Pennant portfolio today, and 4 in the top 10 leading NBFCs in India are our customers. So, this is just to give you an idea in terms of Pennant as a product is present across few of the largest organizations has been able to scale, has been able to perform. And, if I could just quote one simple, a parameter over here, in one of the largest NBFCs in India, Pennant is running in a portfolio which is having the volume of more than 5 crore active loans today.

Now, in terms of the product suite, so we covered the entire lending, life cycle. So, we’d have the loan origination system, there’s the loan management system, which is the heart of any lending operation and then the collection systems. And all of these are supported by various digital enablers. You know, there are a lot of mobility related aspects that come to picture. There is, customer portals, there is lead management systems, there are CRM support systems, etc, which act as the digital enablers, and which is where the agentic aspects are also coming into the picture.

So now with that as a background, I would like to spend a little bit of time, because I’ve been speaking with a lot of colleagues, peers in various organizations, and I have been participating in a lot of conferences, in India and overseas, etc. While there is one segment of audience who are very clear about what agentic AI is all about, but I do feel that there is still, an amount of mist that is covering in terms of a clear understanding of what agentic AI actually is.  I have had conversations where people ask me, yeah, there was a there was traditional AI, then we said that there is Gen AI, then somewhere in the middle, we said that there are chatbots that are coming in. And now we are talking about agents, and then, now we are talking about autonomous agents.

So, let’s spend a little bit of time to understand what exactly agentic AI all is about, and then we will try to differentiate between what a chatbot is, what agentic AI is about, and what generative AI is about so that we kind of clear this understanding. So, agentic AI is beyond reactive, if I have to just put it in one kind of a phrase over there. What that means is, it goes beyond the reactive AI. So, traditionally AI has been more reactive, more prescriptive, or more predictive. So, these have been the primary fundamental attributes of AI or the combination of AI and ML together. But what agentic AI is bringing up always, it is actively pursuing goals and objectives that have been defined for it. It adapts to the ecosystem that has been defined, and the actions are purely based on feedback and from a learning experience. So, it has the capability to learn, it has the capability of taking feedback, it has the intelligence where it can take certain decisions, which are managed by something that we call as guardrails, which we will talk a little bit later in one of the slides.

And it is proactive and adaptive. So instead of waiting for an instruction, in a passive state, agentic AIs are proactive, and they adapt. So, they can anticipate needs, they can generate solutions and autonomously make decisions with defined parameters. So that’s exactly what an agentic AI is and where an agentic AI comes into picture. So, we have been we have been talking about, you know, the characteristics this is just to expand on that particular thought process. So, an agentic or if I may just use the word agent over here are completely goal oriented. The first thing that we need to understand is it is not something over were there which can just chat with you and then, talk about the weather and talk about different things happening, etc. No. That’s what the that’s what the chat GPTs of the world they’ll do with which are very generic, and then they will talk about absolutely anything that you want to talk. So, agents are not, not the GPTs of the world or the Geminis of the world. They are absolutely goal oriented. So, they know what their goal is. So agentic AI systems are designed with specific goals in mind, driving their actions and decisions. Okay? So that’s the first thing that we need to be absolutely clear about. And they are autonomous.

And this is a word that is buzzing around, as to, you know, what exactly autonomous means. So, basically, they can operate independently. They do not need a manual internet intervention to give another instruction or anything like that. So autonomous basically refers to the agent’s ability to operate independently without continuous or constant human inter intervention. So, it can basically perceive its environment whether it is real time data that is coming in or whether it is coming from an API or whether it is a system event. So, it can know basically from where that particular input is coming. It can decide on the next best action based on the goals in the context. It’s not that it has only one answer or it’s not that it has only one defined action. It has sets of actions that it can pick up, but it has an ability to decide the best action among that. Then it acts using, tools or APIs to execute, execute tasks.

Basically, what I mean to say is it mirrors how a human might work with a high-level objective. So, the agent breaks it down, plans, adjusts, dynamically and executes end to end often, with a learning scope over there, it learns from the outcomes also and it takes that into account for its, future performances. So that’s all about the autonomous, agents that we are talking about.

We talked about, you know, learning aspects, which is the self-learning that, the agent can continuously learn and improve its performance based on the data and the feedback. However, it will not accept anything as feedback, and that is defined by guard rails. So, I cannot give a feedback and make an agent learn something which is wrong. So, there are guardrails that can be set up, there is a knowledge base that exists through which the agents basically work.

And we’ll cover that in in a in a few minutes in terms of how exactly an agent, works. Right? And, you know, the biggest kind of an advantage is, it has a contextual awareness. So, agents can understand and adapt to changing context and making decisions based on real time information. It can go into another line where agents can also, switch their languages based on the context of the input that you are giving.

And agents are not just, chats. It can be a streaming agent. It can be a voice agent. It can be a video agent. So, agents can be integrated in any kind of inputs that are coming in. So and even today, you might be getting phone calls, where, essentially, it’s an agent which is talking to you, and it would say that, you know, I’m calling from so and so, this is the amount that is pending, would you be making the payment today? Please say yes or no. And the moment you say, yes, it will take it, or you ask a question, or you say that I do not understand English, can you please tell me this in Marathi? It will actually do that, and it will have a conversation. Similarly, you know, there are agents that are coming up where it can actually process a loan, while on the phone call. Yeah? So that’s how, agents are coming up, and we’ll talk about some of the use cases, especially in the lending industry and some of the use cases that Pennant has also worked and deployed today.

So, we’ll come to that. But before we go that, let’s again demystify, the difference between a chatbot, agentic AI and a generative AI. So chatbots are completely rule based or scripted. So, it is it is very predictable as to what a chatbot would respond because the intents have to be maintained and there is a set of responses for every intent, and that just picks it up and it reacts. So, they are very reactive. However, agentic AI, as we talked, they are autonomous. And we talked that the whole thing about what autonomous basically means. And generative AI is creative. So generative AI goes into a different context where it can get creative, and then it can respond, in a very different manner.

Chatbots are reactive, while agentic AIs are completely goal oriented. So, a chatbot would only react based on what you are asking and as long as what you are asking is in its knowledge base. The moment it is out of its knowledge base, it will say that do you want to talk to a customer service representative, and it will give you the number over there. However, Agentic AI is a goal oriented. So, it knows that what I have to deliver, and then it would work towards delivering that particular goal. Right? And, generative AI, are totally output oriented over there where they focus on entire output based on context and based on the content, which we will also cover. Chatbots have very limited context understanding, and that context understanding also has to be predefined. Okay? However, the agentic AIs, they work on real time environments, and then they use tools and APIs, unlike chatbots which use workflow. So, there is a workflow, I have to go from step one to step two, step two to step three, anything in the middle, I have to call the customer’s representative. Right? So that’s not the case with agentic AI. They work in the real time environment. They understand the context, and then they work accordingly. And generative AI is completely content based, and then there are no external tools, it is completely content based. However, agentic AI requires tools. Agentic AI is not an LLM. Agentic AI is a layer that will sit on top of either your APIs or your service layers or, any of your, integrations that are there. So, it will handle basically you’re in a technical, aspect, if I have to mention, it will manage your cred in a way that, it can do multiple things through the API calls that are happening. Right? So, in chatbots, there is no real time decision making that is happening. However, in agentic AI, there’s a continuous adaptation towards that, and then there are decisions that are happening.

In a generative AI, there is no adaption from real time. It is only from the knowledge base that is there, it will adapt. There are minimal tool integrations in chatbots while agentic AIs are completely autonomous. And, generative AI, the only goal is content creation. So generative AI will only work towards creating the content and giving that content to you. It doesn’t know whether that is a decision. It doesn’t know whether that is an instruction. It only knows that that’s a content that is there. From a chatbot’s perspective, the static responses depend on human inputs. So, all the responses are very static. While agentic AI have independent decision-making capabilities and generative AI, it makes the decision based on input patterns. So, there is a pattern reading that happens in generative AI. The agentic AI, they do decision making basis, how the integration has happened with the APIs around that. While you can also integrate, or, you know, in generally, in agentic AI, you integrate LLM or SLM, in that as well. Right?

So that’s the essential difference between what a chatbot is, what agentic AI is, and what a generative AI is. So how do we set up or how does the ecosystem of an agentic AI basically work? So, there is obviously a user request that is coming, which is from a chat. It could be voice. It could be streaming video. Right? And then that particular input comes to the AI agent. And the AI agent has only three principles. So, there is a perception layer, there is a cognition layer, and then there is an action layer. So, when that particular request comes from the user, again, just to reiterate, the user is either text coming in, part of a chat interface or a web interface or an API, or it is a streaming audio or video that is coming in. The agentic AI receives that particular aspect, then it puts that particular input in the perception layer where there is a multimodal fusion that happens. So as part of the multimodal fusion, it basically segregates the type of media input that is coming, whether it is coming from camera or text or audio or from sensors or any of these aspects. And then it pushes that particular thing to the cognition layer. So in the cognition layer, there are only two critical, I’m not going to go very technical, into this, but just to give you a view as to what a cognition layer does is, there is a lot of memory management that happens, and then there is a knowledge base, that is available. So, memory management is very, very critical, in any AI for that matter. And, you know, there are a lot of technical attributes in terms of how this memory management happens, but that is a key attribute towards a successful agent or an AI output also. The entire success of agent, or AI tool or any kind of even a generic integration totally depends on how effective the cognition layer is there. Because using this memory and using the knowledge base that has been defined, there is a decision making that happens in the cognition layer. And basis that particular decision making, there’s an action that is taken. And these actions are, again in real time are these are provided, and then there’s a continuous monitoring around this action in terms of how these particular aspects have to be taken back into the knowledge base for continuous learning aspects. So that’s typically how an agent basically works.

Having said that, there are risks and challenges in agentic AI. You know, so there’s there can be a bias and fairness issues because, see, at the end of the day, it totally depends on your data. As long as your data is good, your data is, I’ll not say structured. As long as your data is not biased, your AI models that would run on that particular data will also not be biased. But the moment your data itself is biased, the agent will learn only from the data and expect that this is how you want to run your business. So, you cannot expect an agent to come and, work on your data, which is not fair. So that’s a key to understand that you need unbiased data to ensure that the ethical aspects are taken care of.

Security threats, obviously, you know, they keep coming in, and that is a continuous kind of a threat in any kind of a technology, let alone the agents. There is also a perceived loss of human control that is there. So, there’s this concept of human in the loop that is being talked about a lot right now where, the hype that, not too sure if hype is the right word over there, but the amount of disruption that, that AI and especially agentic AI is bringing in the world today, there’s a perception that there is going to be a loss of human control. So, it’s very important that there is a human in the loop. So, that’s an important aspect of the risk that has to be always kept in mind.

Challenges we covered. So, the ethical concerns, like, in terms of, how your data is there, whether it is biased, and, what are your guardrails towards, managing that, what is the governance model that you are putting on top of your, of your AI framework, which is very important. The regulatory and compliance issues, so there are, you know, obviously, there from an Indian ecosystem point of view also RBI is, is actively working on, AI related stress models, and then, there is a mandate also, to you know, kind of ensure that any kind of an artificial intelligence based critical decisioning that is being put into any of the core banking aspects has to perform a stress test on that. So that there are regulatory and compliance related aspects that have to be kept in mind, and there’s obviously accountability and transparency that has to be provided because, a lot of again, there’s a strong perception over here that AI is a magic wand and then you just do abracadabra and then magic comes into picture. No. There’s a lot of integration that is there’s a lot of data that is there over there. So, it’s purely technology. There is no there’s absolutely no magic over there.

And with that as a background, I guess we can just move to our first, poll question just to, you know, ensure that we are we are with the flow and then take some inputs from you as well. So, if I can request the users to select, any of the options pertaining to the question is where is your organization today in its AI journey, especially for the lending operations? And this is this. Let me try to pitch, my next conversation, with you. So, I’ll give about ten seconds over here, for the poll to be taken.

Two more seconds, and we’ll move on. Alright. So 42.9%, which is good, saying that actively scaling AI across lending life cycle. This is very, very pleasing to, note, because, from my reading, is, with the customers and then, with our partners that, I’m working with is, we are setting up the ecosystem to kind of bring AI right now and most of the banks or most of the organizations and when I say ecosystem, it comprises with, the infrastructure, the skill set. And, then the third thing, unfortunately, which I’m seeing is, lack of use cases. So, we are still hunting, I think, in a lot of organizations as to where exactly do I put an agent or where exactly do I create an AI use case and how do we bring that level of effectiveness from that particular AI. Right? So, we have 28.6% who are still exploring new AI use cases, which is a fair kind of a poll assessment in terms of what we have been seeing in the industry as well. And piloting AI in limited areas, 21%, it is it is very pleasing to see that as well, and we’ll be happy to connect and then see if, there are any areas of collaborations on or common synergies there.

So, with that, let’s just, quickly talk about, you know, summarize what we spoke about right now. So, we have moved from chatbots to autonomous agents right now. There is an increased complexity, that autonomous agents are addressing. And then there’s higher efficiency that is, the agents are bringing, it could be credit task, or it could be cutting down your manual efforts, or operational inefficiencies or improving the efficiencies over there. Real time decision making in a lot of areas that is coming in. Faster approvals, these are some of the numbers that we have taken from Forrester. 67% of AI enabled lenders so a 25 – 30% drop-in loan approval times. And, it is obviously at the end of the day, what everyone is looking for is cost efficiency. And one thing that sits on top of all these attributes is customer satisfaction. So, with AI, I think these are the two objectives that every organization is looking for is increased customer satisfaction and improved cost efficiency. So, these are the fundamental two areas or goals for every organization where they are looking to implement AI.

And in Pennant also, we have we have a strategy, from an AI perspective deployment where we are working in three areas from AI or Gen AI implementation point of view. So, one of the areas is, our own internal efficiency. So, we are a technology company. So, there’s obviously a lot of work that happens in the background in terms of how we build products from an engineering point of view, from a testing point of view, from a solutioning point of view. So, we are building agents which can improve those efficiencies. So, we have we have set up, our own data centre with our own tools, and then we are running on our own models, which can work, on our core factories. And then we have trained these models in a way that, you know, the data is structured. We have created the vector databases in a lot of areas. We have implemented agents in our support framework which helps our support analysts to quickly resolve any kind of production issues that are coming in. So that’s one area on improving the internal efficiencies, within the organization.

The other area is embedding AI in our existing products in itself. So be it the loan origination system, the loan management system, collections, the lead management systems. So, we are we are finding, areas where we can embed AI, AI agents, or analytics in any of those areas, which could provide better efficiencies to the entire, loan life cycle in itself. And that is another area that we have adopted AI. And then the third area that we are working in is creating AI first product itself. So, this year, we launched our, new product, which is called as the PenApps Studio, which is a low code, no code, and a pro code, platform, through which you can create your own applications to the scale of actually creating an LOS in itself. And it comes with AI and Gen AI integrations where you can create your own journeys, workflows can be created by the agents. And even, there is a pro code attribute where you can write your complex code or a Copilot integrated, which can help you providing that particular code and generating it for itself. So, the reason I just draw the tangent is this is exactly what every organization is doing. There are internal efficiencies, there are business use cases, and then there is new products that are coming in.

So, from a business use cases, let’s just quickly go in some of the use cases that we have also implemented. So personal loans is a simple use case, and then we, implemented this, with one of our agents. So, all the agents that we are producing, we call this as Pennfinity Garage. So as part of this garage, we are producing different agents and to help the various loan processes. So, we’ve launched the personal loan agent which is something that is available today where, where it can help a lender uses agentic AI to autonomously evaluate personal loan applications and analysing credit scores, income data, and spending. So, this can this has two attributes. It can be used B2C and B2B. So, the end customer can also, you know, interact with the agent directly or, if a branch can also use this particular agent and then basically interact. So, it basically does end to end loan processing. It guides the applicants, collects documents, validates the authenticity of the documents. It has integrations with the vaults and e-signs and the likes of all the integrations that are required in a LOS kind of a journey.

And then it qualifies through the documents that are submitted because it has the intelligence to read the documents. So, it can read, for example, a PAN card or an Aadhar card or a license or the payslips that are there. It has intelligence to evaluate the overall income, aspects around that, what is the eligibility around that, and then it can have this conversation real time with the with the end user in terms of giving that personalization impact, and then it can recommend the best fit loan products or it can if it is a personal loan, the best fit personal loan, you know, kind of, options that are available with, you know, it can be a varying EMI, there could be different kind of, attributes that can be integrated. And you can integrate with your own BRE in terms of how your rule engine attributes have to come in and, comes in with, multilingual kind of a support to help different communities across.

So, what’s the impact of this? The impact is fast, tailored, approach. See, one of the things around, why the agents or for that matter, even the chatbot got so famous is, it is a private conversation, and I can ask anything which, in general, there is a tendency not to ask, you know, certain things, what is the other person going to feel? Do I say that? Can I get this instead of this? So, all these dilemmas are actually taken out with a one-to-one chat, interaction where the user is asking absolutely anything and telling anything and is very clear in terms of what he or she is expecting.

And then the chatbot is not a human. They also know it’s not a human, and as long as it is able to integrate with an LLM and then it is having a content and a context-based interaction, they’re happy with it. And the measured results are approved; time just goes into minutes from days, there’s no manual intervention that comes into picture, the accuracy, and the in-loan default predictions, which come in as an input, it improves. And then, obviously, the biggest thing is the customer satisfaction.

There’s another you know, in the interest of time, I’ll just go quickly around, you know, another kind of a use case in the asset finance world powered by the agentic AI. So, asset finance from a docs generative, agentic framework to automate and streamline the loan approvals for SMBs, replacing manual data entry and fragmented process. So, this is the efficiency part that we are talking about. So, end to end application automation happens. The agent guides the borrowers through the applications, collects documents, validates data, prequalify the applicants in real time.

There is also an attribute of fraud tolerance and fraud management that can be integrated either with an external API call or any kind of an agent that can work on that as well. Right? So, there is dynamic risk assessment that happens. The agents fetch financial credit and asset data via the API framework, and then they run the scoring models and provide risk-based recommendations. So, this is, actually, you know, coming up very, very successfully in a lot of industries today.

And there are tools and system integrations that can be kept. So, the agent, what we need to understand over here slightly from our technical and logical view is it’s not one agent that works in the background. While we are interacting with one interface, there are multi agents that work in the background. That’s the reason we call it as an agentic framework. So, every agent has got a goal, and these agents work in their own silos, however, integrating with the master agent, which is controlling the input. So that’s how the entire agentic framework basically works. So, this is also bringing in this particular use case, this proactive engagement that the agent can, you know, give, like, you know, follow ups for missing documents or status updates, this and this continuous learning that is there here as well. So, impact, smarter, faster loan approvals with minimal manual effort. So, efficiency improvement, and this is where most of the organizations are going today.

Efficiency and customer. Right? So, these are the two things where agents or AI is coming in a big, big way. Measured results, we just spoke about the loan approval time, data entry, errors, and operational costs coming down, customer satisfaction, and conversation rates are on the up.

Quickly on the third use case is a commercial lending area. A commercial lender just uses agentic AI to automate large business loan evaluations, analysing financial transactions, and market data across system. So automated application handling, the agents can guide applicants, collect documents, prefill forms via API, real time risk insights. So, this is the interesting part where the agent can perform or give a score in terms of a risk rating that can be configured or that can be measured by the underwriters or the credit department, whichever is there. So whether, you know, that particular business entity, you can define your own scoring algorithm and feed that as a data input to the agent, and then the agent will work on that and it will go to the Internet as well, and it will bring, the relevant, data from the Internet. It will put it in its vector database. It will create the chunks around that particular database, which will again go as an input to the entire scorecard and provide a relevant score over there, and it can flag the risk accordingly. Right? So what it does is it helps in smarter underwriting support where for a business entity, you have a score indicated by the agent already, and you can actually click on that particular flag, and it will give you more details in terms of why, or where, that particular input is coming from, which is flagging it in red, amber, or green. So those are the capabilities that it can provide. And then there’s a compliance monitoring in terms of, a lot of reviews around the documentations, and ensuring that there is regulatory alignment. So, impact again, faster decisions, better risk accuracy, and lower operational effort. Approval time is you know is up by about 50% to 60%. Credit assessment errors are down. Operational costs improvement by about 30% to 40%, and loan processing volume goes up.

So, with that, again, we’ll just go to a quick poll question just to take some, feedback from you. So, what is the biggest barrier, that is there today in terms of adopting AI driven decisioning in your lending operation? So, if I can get this input, we’ll probably have a conversation around that.

Just in giving it about ten to twelve seconds. Right. So just giving me a couple of more seconds, and then we’ll take the output. Alright. We have a pretty distributed response, which makes it interesting. So, 27% is, legacy systems and infrastructure limitations. Again, 27% is regulatory or risk concerns. 27% is unclear ROI or business case, which I feel is a lot, more prominent, from the interactions that I am having. Data quality and availability issues. Again, a fair kind of a count over there. But having said that, I’m a bit surprised that lack of internal AI skills or expertise are not a factor today, but from various conversations that I’m having and if you ask me, that is probably the most critical thing to be done as well is to build an AI skill set. Anyways, we’ll see how we cover some of these things in terms of how do we plan the implementation of AI driven lending platform. So phase one of the planning is let us clearly define the objectives, scope, success metrices, identify key AI models, whether it is going to be credit scoring or risk assessment or fraud or operational efficiencies or you know, from a collections point of view or predictable default predictability or telephony integrations or customer support. So there has to be a very clear view where what is the scope. Usually, in most of the organizations or a lot of, CIOs and CTOs that I interact with, there is a budget that my company or my organization has allocated to implement AI, now tell me the use case. So that’s where most of the conversations are starting and stopping today. So, I think we just have to take that conversation couple of steps further and then define the scope because from a lending point of view trust me, there are a lot of areas in which AI or agentic AI can be implemented. Let me also add analytics as part of this because there is an underlying AI ML attribute over there and there’s a lot of areas where it can be done. And let us be very clear in terms of what are we expecting in terms of the goals or KPIs, for AI implementation. So, this is the first kind of a homework that needs to be done. Then phase two is we prototype. We develop the pilot models, and then we test it with limited datasets. And this is where the skill sets come into picture where we need to, we need to define the framework, we have to define the architecture, the POC, and then the initial testing and feedback loop. Phase three, we scale. So once phase two is successful, then we start scaling. We integrate the AI models into either the core systems or the workflows or the BREs or the, the data lakes today that we have. And then we do a full deployment or automation across the business. And four is four is, the optimization aspects where we start fine tuning these models, in the real world, data. So, this is a simple way of starting an AI, kind of implementation in your respective businesses, and we are happy to collaborate and help you, plan and implement any of, any of these use cases that you might have in scope.

And then how this is going to kick in. Right? So, integration with existing technical stacks is the key aspect. So, from a LMS point of view, an API based integration, is required for data exchange, and, you know, you can use any of the restful APIs or any other APIs that you have as long as you have. If it’s not restful, then, if you have, an open stack APIs like Swagger or, if you have an ESB, then you can integrate it through that as well. So that is for an LMS integrations that is required. So, you need to have an API capability in your LMS to integrate with any of the agents or AI framework that you are looking to build.

From a CRM system, again, AI driven customer segmentation will be the integration focus. And, you know, from a real time data sync, and machine learning capabilities that have to a worse setting, that is where you can deliver. Risk assessment engine, you have to have an advanced AI decisioning model, which is complicated, and it totally depends on your risk policies and your risks scoring dynamics. And that has to be fed in and then created on the AI/ML layer. And that is on top of that, the agent can come and sit. So that is, again, to implement predictive models using historical data. Compliance tools, automating compliance checks using AI, again, building AI power tools. And this is, trust me, this is not very complicated, and it can be done because compliance is a finite set of aspects that we are looking at. So, there is clear instructions. There is clear expectation that is available and basis that it can be done.

So, these are various kind of an integrations that’s there. So, what are we doing with Red Hat, and then how are we taking this initiative forward? I can see that this is one of the Q&A questions also, so let me just answer that straight away. So historically, you know, between software and hardware oblique infrastructure, there has always been a challenge. It was like a Batman vs Twins story; While the software wanted to give more output, faster output, etc, we never had the hardware or the infrastructure to kind of give that kind of a processing or a compute engine for the software. But now what has happened is with a lot of advancements that has happened in the overall infrastructure industry and the focus on specific areas like AI and related aspects, Red Hat has also come up with a lot of AI services which Ajit was mentioning earlier. And, Pennant is collaborating with Red Hat in terms using that particular infrastructure or using those services around that and fuelling our solutions in a way that the lending related use cases that I just spoke about and then, you know, that we have we have integrated, they get better optimized and, you know, powered with the required compute and then the service that is required.

And one thing that we need to remember is, especially when the likes of AI and GenAI, there’s a lot of infrastructure compute that is required, and there’s a lot of, processing, and services that are required with, which the likes of AI Red Hat is providing today and which we are leveraging to, provide the solutions in the lending space. And this is just a very quick representation of that. So, there is a cloud layer that comes, there are multiple, you know, partners that Red Hat has over there. There’s a Red Hat AI platform that sits over on top of it.

There’s a lot of services that Red Hat is basically providing today. And on top of that is the Pennant lending factory, and all the digital enablers and then the AI agents are basically sticking in that particular scope, which are fuelling our, overall journeys of loan origination, loan management, collections, and the other products that are available to ensure that we are giving an effective output of our solution and an optimal output of our solutions, which can scale, to a larger set of organizations, in an effective manner. So, with that, let’s go to our last question, and then accordingly, I think we are close coming towards the closure of this session as well.

So, what lending functions do you see as the most promising for AI enabled autonomous agents in the next twelve to eighteen months of our time frame?

Just giving it about ten to twelve seconds. Couple of seconds more. Right. Credit decisioning and underwriting. Undoubtedly, I think that’s one area where there’s a huge, kind of requirement and implementation as well. Having said that, there’s a lot of scrutiny as well because that is the most critical part of any lending life cycle. So, yeah, so that is one area where there are multiple things happening in terms of making it more efficient through agents, getting an alternative credit score through the agents, and then giving a risk assessment profiling, use case that we talked about. So, all these areas, you can bring in the agents to work for you. Great. So, we have just come towards the end of the clock.

So, we will just take some Q&A, if there are there are any, and then, we will look to close the session. So, Ajit, unless you have anything to add, we’ll go to the Q&A.

Ajit Joshi

Yeah. So, most of the questions I have answered, online, and, you have taken one of the questions. So, one question was that, are these products ready?

So, I said we are ready by the customers. So, it is not like it is some theoretical thing. Initially, Shirish had mentioned. Then, then we had a question on value proposition. You have taken it.

Somebody said that what is the name of the product that you covered in the last slide again, Pennant Lending Factory? Somebody said our slides would be given. I said, yes. We’ll be sharing the slides. So, I think with that, it is also time up. We are at the top of the hour. So, thank you very much, to for attending our session. There was some confusion with the invite. Apologies on that. But thank you very much. Hope as we are developing, you will see a lot of communication from us. Hope to be engaged with us on our solution. In case, you need further detail, we are all available. So, you could connect, with the email IDs which are available, and we would take it forward. Thank you very much. Thank you, Great. Thank you.

Sireesh Patnaik

Thank you, Ajit, for facilitating this webinar. Thank you, Red Hat. So, it was a pleasure discussing this. So, our coordinates are available. Please, feel free to reach out to us, for any further questions you may have on Pennant point of view, you can reach out to me directly. And anything on Red Hat, you can reach out to, Ajit. Thank you so much for being a patient audience over here.

Ajit Joshi

Thank you so much. Thank you. Bye.