The North American banking sector is moving beyond digital transformation into intelligent transformation. AI is at the core of this shift. While 70% of banks have already adopted traditional AI to streamline operations and reduce costs (Source: American Banker, 2025), the next frontier is Agentic AI, where systems can reason, act, and adapt autonomously.
Banks embracing Agentic AI aren't just streamlining operations, they're driving measurable business impact. While broader financial services organizations are leveraging AI to accelerate decision-making and improve efficiency (Gartner, 2024), banks are leading the charge in applying AI to speed up lending, strengthen fraud detection, and deliver hyper-personalized customer experiences. According to Forrester (2024), 67% of financial services leaders, including banks, plan to increase AI investments in the next year to drive productivity and enhance customer outcomes.
As autonomy becomes the next competitive frontier, banks that delay risk being left behind in an increasingly intelligent and dynamic financial landscape.
Danielle Fugazy, Senior Content Strategist at American Banker, sits down with Newgen Software's senior leaders:
- Rajan Nagina, Global AI Practice Head
- Varun Goswami, Head of Product Management
Watch the full 30-minute video to hear them unpack how banks are transitioning from automation to autonomy, and what it means for competitiveness, compliance, and customer experience.
Transcription:
Transcripts are generated using a combination of speech recognition software and human transcribers, and may contain errors. Please check the corresponding audio for the authoritative record.
Danielle Fugazy (00:08):
North American banks are moving beyond digital, entering the age of intelligent transformation. At the heart of this is ai. Traditional AI has played a significant role in helping banks streamline cooperations across enterprise content management, business process management, and customer communication management, as well as domain specific areas like retail and commercial loan origination systems and digital account openings. However, the next wave of transformation is being led by agentic ai where systems can reason act and adopt autonomously. This transition from automation to autonomy is redefining the future of banking, and Nugent software technologies is right in the thick of it driving AI first transformations for banks across North America. I'm excited to unpack this future with Raja Naina, Nugent's Global AI Practice Head and Varun go Swami, head of product management at Newgen. Thank you both for being here today.
Rajan Nagina (01:12):
Thank you for having us.
Danielle Fugazy (01:14):
So let's just get started here and level set a little bit for our audience. We're hearing a lot about the shift from traditional AI to agentic ai. What's the evolution of agentic AI from traditional ai and what are the capabilities that banks are finding most helpful today?
Rajan Nagina (01:33):
Most of these traditional AI systems are built to analyze data and make decisions based on that data. Agent AI systems are not just analyze and make decisions that region with the data, they also do things so agent, essentially not just making decision, but also executing that decision and learning from the decision and making it better. Agentic AI systems are in a sense, humor like they understand they do things given the feedback, they become better with time. So that's the power of ent. AI underlying, there is an element of AI, of course, where that is where decision making comes from, but from decisioning to execution is the core fundamental construct of ENT AI systems. Where are banks finding value in that? Where are the core capability? We can think of agent AI systems as agents or helpers who are primarily task-based to begin with. So you can outsource other tasks, you can give some tasks, provide some tasks, they can do it for you.
(02:43):
Moving to skill base wherein you can say, I want to do this, and the agents can look at that and solve it for you. Do it for you to even more complex cognition levels such as becoming the research assistant for you. So then you can region with it, oh, I want to do that. How should I do it? So one is, oh, can you go ahead and do it? Which is task list. Second is, oh, there is a problem I want to solve. Can you solve it for me? For example, I want to do a credit decisioning. Can you do credit decisions for me versus that I want to launch a new product? What product should I launch? That's the cognitive spectrum of atrix system that makes it so powerful.
Danielle Fugazy (03:23):
That's so amazing. And I have a few follow up questions. So do you think the banks are understanding that differentiation at this point?
Rajan Nagina (03:34):
So two things. One by and large there is we talk to a lot of banks. We work with a tremendous NU amount of banks globally, and we talk to our customers quite regularly, ontech these days. We are working, we're implementing for the agent stuff. For them. There is a by and large appreciation for the system because edge consumers first to me use it like in Jack GPT form in other forms of in copilot form and whatnot. So there is a certain level of understanding and a great level of appreciation. While that said, how to bring that level of intelligence and execution power within enterprise not very well understood. It's being figured out and hence we work very closely with the banks to make that journey to start with and move forward.
Danielle Fugazy (04:21):
I feel like you may have answered this a little bit, but I want to make sure we set this for the audience. So what does this shift mean for banks and why do you think now is the right time for an AI first transformation?
Rajan Nagina (04:37):
So this shift means for banks a great deal because it impacts all the core constructs of bank, starting from customer experience to higher productivity, to risk management, and to strategic levels wherein you can region with underlying system to make better products, create competitive advantage for yourself. The shift, it's ubiquitous, it's everywhere. It's just that where to start from it's element of cognition, it's an element of wisdom that works on your data, do things for you sometime fully, sometimes partially, and then hand over to someone else who can do it. So one tremendous productivity gains because it's someone else is doing it for you. You doing it faster certainly means translating into customer experience because you're really coming back and doing a lot of instant gratification with your customers. You creating a lot of operational efficiency and health being a cost advantage position because of this. And finally the ability to look at a large set of data and have an underlying reasoning with them gives that insight, which wasn't even possible I guess one year ago. And hence a great competitive advantage.
Danielle Fugazy (05:48):
It's unbelievable how fast this is all moving.
Rajan Nagina (05:51)
That absolutely
Danielle Fugazy (05:52):
Varun, I want to turn over to you. There are a lot of misconceptions out there there about AI in general, but agentic ai, can you talk through what are some of the biggest misconceptions that banks have when they think about agentic AI and how can banks move past that hesitation? How do they get everybody on board to move forward?
Varun Goswami (06:13):
Yeah, that's a great question. So agent AI systems are built upon generative ai, which is the underlying technology. Now, generative AI by itself does carry some concerns which banks have in terms of hallucinations, in terms of incorrect information. So I think it's very important for banks to have this layer of trust built in between and have trust as a central tenant to any platform. So for example, when we built the whole Agent X system, we kept trust as a central tenant. So we said, what are the different levels of trust that we can build starting from the very beginning, which is a deployment model. So banks have concerns in terms of my data going out or models getting trained on my data so that by devising an architecture which is completely in-house based, where even the large language models are deployed locally. So there is a comfort factor for banks saying that, okay, my data is protected, it's not going out and nobody else is using it.
(07:26):
The second is also how do I trust the output which is coming out of that system? And I think that's a very important point. And what we do over there is we use the bank's own data set and content to train these models. So we do a vectorization or we train small language models. So what happens is all of this is done on the bank's data and hence the reliability of the output increases dramatically, right? And the third is when any agent tech system is taking an action or giving an output, how do you go deeper into it? How do you find out why this particular decision was taken? What was the data that was referenced? Or even if it's a document based, what was the document, which clause was referenced to take this particular decision? So having that explainability and that deep dive into the process of the agent system becomes very, very important. And of course, on top of it we have something called an agent shield prevents, which basically protects users from certain kinds of output like profanity or abusive language. So for banks to set up or build this kind of a system, it's very important for them to go through these four layers, understand how their data is being protected and then start experimenting. But definitely all of these concerns have solutions, so they should not wait and watch or they should not hesitate. They should definitely go ahead and start using these.
Danielle Fugazy (09:01):
That's a good call out. I think that you're right that some banks worry about these things and it delays rather than jumping in and figuring it out and moving forward. Varun, just staying with you for a minute, what bank functions are most ripe for AI disruption today? There's also a perception that AI adds risk, but in reality it can actually reduce risk. Can you just talk about that also a little bit? I think that kind of goes with we were just talking about with the misconceptions.
Varun Goswami (09:36):
Yeah, yeah. I think any process which is content heavy is absolutely ripe for an agent kind of a system. Traditional ai, which is machine learning based AI was really great at going through vast sets of data and getting meaningful output out of it, but it had some limitations in terms of it needed structured data, it needed all the data to be in one place with generative ai and now with agent ai, that restriction is no longer there. In fact, generative AI excels at going through a large amount of content, which is documents could be residence proofs, ID proofs, account statements, could be income statements, tax returns. It can go through a large amount of data and generate some very real insights out of that data. So any process like a retail lending, commercial lending, trade finance, wealth management, all of these, if you look at it, their processes are very content heavy.
(10:42):
So banks can definitely look at this and they can get some real business insights and they can get real productivity gains out of it. On the risk front, I'll say that okay, risk is something which is of course for banks, it's a big deal, but there are certain types of risks which actually an agent AI system produces in terms of you've got biases, you've got got human errors, which and agent tech system would never make. So actually you are reducing your risk in your operational process by deploying AI agents which are transparent, which are explainable and which can get a lot of done work done much faster.
Danielle Fugazy (11:22):
Yeah, when you say it that way makes perfect sense. Je, I want to go back to you. Can you share some real world examples you sort of did when we started, but can you share some real world examples of where agentic AI is already making a tangible difference? Maybe in the alone origination, there's so many places onboarding, loan origination compliance for North American banks.
Rajan Nagina (11:47):
Sure, sure, absolutely. So working with a large credit union wherein the agent platform is deployed and we are in a sense transforming the entire landscape of the lending by solving a typical trade off of growth versus risk. So if you grow very fast, you have to manage the risk of that growth because you have to underwrite that fast or otherwise you break the customer experience. Or if you don't put the underwriting levers that strong, then you basically get into the higher risk territory. So our ENT platform essentially look at this both one, how do you keep going growing faster, and then how do you keep managing the risk of the additional application that you get because of the fast growth? The agent solution essentially goes as a life cycle of lending, starting with the first as prospecting, there is an agent for prospecting among the base who are the ones who are likely to take which lending product.
(12:54):
So you look at this, the agent, look at the data, go through the data, analyze the data, and come back and create high value prospects of a particular product. Now that's first part. The second part of it is if you figure out these are the prospects, can you also create personalized offer for the prospects in terms of limit setting the offers and whatnot? So you move not just from prospect, but to rather offer driven prospects, which certainly takes, let's say get closer to the actual conversion. So these two were the first, let's say part of the agent tech system that focuses on growth even for lending product, essentially moving the lending business ahead. Then the other part of this is the underwriting or create decisioning agents which look at, well, I have got this data, I have got this applications, how do I underwrite better by looking at data that's there, looking at data that can fetch from integrations, but also look at the data that is available within my system, not just the traditional, let's say relational data, but also the content that Garron spoke about. There's a lot of data available in form of let's say bank statements in form of your tax filing in form of your credit decision reports.
Danielle Fugazy (14:02):
Those
Rajan Nagina (14:02):
Data is not something available at a table in A CSV, but it's available. So we put all that data so that you can have a deeper credit decisioning strength. So all of this put together the system essentially look at both, it moves your business identifi faster, also manage the rest of this. But that's one aspect which essentially increase lets the productivity gains and the velocity of the business. The other aspect of EnTec is the reasoning capability, and that's like a little thinking fast and slow. One way is, okay, the system maker decision, this is a good application. The credit decision person can ask why it is a good application and it can answer you why it is a good application. It can also refer you to different points. It's a good application. Well, that's one thing. The other thing is the person who is running the business is not looking at one application at a time.
(14:52):
The person, the business head is looking at thousands of application that has gone processed in the month and now you want to know what my customers are looking for. Why do I reject applications? Why do I accept application? If I accept application that missed, that's my target segment, that's a target segment, I can reach out better because that seems to be higher acceptance of my products. So look at both the strategic layers that you can put this all data around and then have a regioning capability and the underlying data to figure out better what my customers ask for, what they're looking for. Can I tailor my products? Can I create new targeted segment product? All of it. So while it increased the productivity layers, also it translate into better customer experience of course, but it also gives you that strategic lift because you can now region with underlying system talk to it, figure out what goes inside. It's just incredible.
Danielle Fugazy (15:44):
It is it, it's just so powerful
Rajan Nagina (15:47):
Indeed.
Danielle Fugazy (15:49):
But you can't have this conversation without talking about the data stack, right? I mean if you don't have the data, you're not in a good place. So can you talk to me a little bit about what's required? Do you need a new data stack or is there new infrastructure and what advantage does this bring over traditional AI implementations?
Rajan Nagina (16:18):
Yes. So it certainly require new data stack. Why? Because the underlying data changing in earlier system, most of analysis would happen on data, which is a table like structure. Now you are going beyond that. You are trying to look at what is there in documents. You're trying to look at what is there in events. You're trying to look at what is there in call, like normal phone calls you're trying to see a lot of customer service has gone shifted on chat, conversations, chat.
(16:46):
And because it's so synchronous, you don't have to go to the phone, you chat, you do something else, you chat. The asynchronous nature of customer service creates data in chats, and that means the underlying information landscape has changed. It's not just data. It's data in many, many forms to leverage that one for decision making and more so for reasoning, you certainly need a different kind of data stack. And as Warren mentioned, vector DB or vectorization of that is one element to it, but there are many, many forms of such. So you certainly need it, but it's totally worthwhile. It's worthwhile in a sense. It's worthwhile because you can do so much out of this data and hence it's worthwhile. Some of it is very complex and hence we put it as a part of our platform only because if one to restriction, if you want to get to generative AI applications and if you have to do so much on data infrastructure, probably there will be a lot of friction. So the platform that we built has this new data stack already interwoven, so let's reduce the friction. You get the platform, it sorts the data out and sort of start deploying. The agents have the goodness of agents in a very short period of time.
Danielle Fugazy (17:59):
So what would you say to a bank that is like, this is their hesitation, right? They let these roadblocks get in their way. What would you say to them about not letting this be their hesitation point, like building the new data stack?
Rajan Nagina (18:16):
I think the value that it brings in out runs the friction that comes along is as simple as that. But while you say that in abstract term, it makes sense when you put an implementation plan, it becomes very intimidating. So to say, it's sort of very frictional, and that's what I said, our job as a technology provider is to ensure that the complexities are solved for but remain within the platform so that the friction become relatively less. While there is hesitation, platforms like ours, Newgen one solves this problem to a great deal. We also work very closely with customers in a form that because it's a new thing, it requires
Danielle Fugazy (19:02):
That it's so important. That piece is probably so important. Varun, you touched on this already talking about the safeguards, but I want to go a little bit deeper into that. What mechanisms do companies like Newgen have in place to mitigate AI bias to ensure fair explainable decision making? And we've talked a lot about the idea that this is all explainable, but yeah, if you could just dive into that a little bit deeper.
Varun Goswami (19:33):
And if you look at it, that's at the heart of any ethnic system, right? Because the ability of being able to consume different types of data and make me take out meaningful insights out of it or take decisions on top of it is what is differentiating the agent AI framework from the previous traditional AI framework? One thing in particular is that when we talk about age intake, we are not letting go of traditional ai. Traditional AI becomes one of the sources of information. So there's still machine learning models running, which are doing maybe content classification, which are doing probability of default. Those models are still running. Now what age intake does is it brings all of that into a single system with which you can interact. I think that interaction is the biggest difference, which has not come. Now I talked about that trusted data, but it's also very important to understand the trusted models because this is your data.
(20:34):
The models have been trained on your data, so these models are also not trusted. So you've got trusted deployments, trusted data, trusted models, and the ability to audit and interpret everything. What happens in our system is typically whenever a decision is made, all the Q and a, which a bank employee would do with the Gentech system or any reasoning that the Gentech system made is all audited. And this was very important because if you want to go back and find out why this decision was made, or if something happens and this and you want to go back and find out who did what, our system already has that detailed audit trade. So you can even find out that, okay, an underwriter spoke to the AI agent, understood these four things and then took this particular decision. So beyond the trust layer, it's the auditability also, which becomes very, very important in any agent system.
Danielle Fugazy (21:35):
Yeah, I would say. So let's talk a little bit about compliance and the regulators. That's been a constant piece of this conversation throughout. How do you ensure your AI models and platforms adhere to regulatory standards, and what is your sense of how the regulators are feeling about AI based decision making in banking these days?
Varun Goswami (22:02):
That's a tricky question to answer, but not because of the regulations itself. It's just that the field is changing so rapidly.
(22:11):
Every two to three weeks, there's some new development which is happening. In fact, even if you look at most of these meta lama, Claude Anthropic systems, all of them are also moving into a chain of thought models and a chain of thought model works very differently from a traditional large language model. So there are so many new developments happening that regulations or also taking time to catch up. But definitely at NuGen, we are committed to ensuring that our customers stay safe. So whether it's even before this, whether it was PCI DSS or whether it was HIPAA compliance or GDPR, we have always ensured that whatever regulatory compliances there is becomes part of our system itself. So our customers don't have to worry about whether I need to build another layer on top of it to ensure compliance to regulatory systems. We have all of those capabilities built into the product itself.
Danielle Fugazy (23:12):
So we talked a little earlier about onboarding and lending. Can you talk a little bit about the key benefits banks are getting when they leverage the AI capabilities?
Varun Goswami (23:28):
So I will say the benefits come under two broad categories, the tangible and the intangible benefits. Intangible benefits like your risk is reducing or your overall growth is increasing. But in terms of tangible benefits, what we are seeing is we are seeing a drastic reduction in the turnaround time. So the kind of work that people were earlier doing, going through large set of, there's so many fields on the screen, so many documents that they need to go through, and with agent X systems coming in, it has made this job much, much easier. So we are seeing that a lot of heavy lifting gets done by the AI agent and which gives or presents the data to the user and the user's job has become much more easier and faster. So what they're ultimately able to do is they're able to process a lot more applications in smaller amount of time.
(24:23):
So we are seeing a drastic reduction in turnaround times. We are seeing the accuracy rates improve, we are seeing the approval rates becoming much better because now there was always a pressure saying that, okay, I need to finish this application in so much time, and there was certain nuances which were getting missed. But by having agen tk, I do a lot of heavy lifting. A lot of this information comes upfront to the users, so they're able to more confidently process applications. So I think across all of these three areas, we are seeing a major benefit being given to banks. So
Danielle Fugazy (24:57):
I want to turn back to you with ECM and BPM being core to how banks manage unstructured data and workflows. How can AI be integrated into these areas and how does Agen AI change the game for content led decisions and keeping in mind that speed and compliance are both critical here?
Rajan Nagina (25:20):
Sure, sure. I think workflow are the execution arm of any bank. You process scale through workflows, they create consistency and through which the users basically take the decision and move from one place to another. The content flows through the workflow. And hence, if you look at the content repository till now, it has been by and large used as a repository. In that repository lies vast amount of previous decisions, vast amount of information that is now being leveraged both independently because that repository suddenly has now become a knowledge hub in that repository. Knowledge lies. So you have a repository, you agent to create knowledge out of the repository, and second, this knowledge now flows back in the workflows to help users to gain productivity. And the productivity is coming because there is some wisdom in it, isn't it? So the cognition and the knowledge powers the workflow to create higher order productivity and hence you can do faster.
(26:21):
Second is you're able to take decisions based on what is there in those repository, which was not the case before either somebody would have manually gone through it or somebody would have done some sort of, let's say buy and large summary basin stuff from the documents and then move on with that. So either your quality of decisions were relatively low or your time of decision was relatively high. With this, your quality of decision goes really amazing with really in the fast time possible, and you embed this in the workflow so that people become really empowered. That's the beauty of agent care. It's really magical.
Danielle Fugazy (27:00):
You make it sound so perfect. So just a few wrap up questions here. First of all, I want to thank you both for sharing all your insights today. It's been great. So you both spoke about the transformation, but I guess I have a few follow-ups. One, how do banks get started? Why don't we just start with that? How do banks get started?
Varun Goswami (27:27):
Yeah, I think the starting is always the most important part. I would say that any new technology brings in certain improvements to existing processes and certain transformative experiences, which were earlier not possible because the technology wasn't there. What I would suggest is that banks should start looking at their existing processes first, look at all their core processes that they're running, whether it's account opening or lending or trade finance, and start looking at how can generative AI add value to these processes? How can I help my own employees make faster, more accurate decisions? That would be a good starting point. And then also at the same time then start looking at what new experiences can I build for my customers and now I expose my customer's data to them so that they can also start interacting with my backend systems. So those kinds of transformative experiences I believe should follow. But first, whatever is running, how do I automate, how do I get agent A, KI into those processes is always a good starting point.
Danielle Fugazy (28:45):
I want to give you a chance to answer that question too, if you have anything to add there. Sure.
Rajan Nagina (28:49):
I think completely what Warren said in a sense that the starting point, essentially what we do to help banks is to solve this dilemma of time, value, and cost. This is something that everybody is excited about, there's no doubt about it because it brings real value. The fact is how do we go about taking the first initial steps that has value, how do you do it fastest and how do it in the minimum possible cost if possible? This is what we solve for. What we have done is we have built deep vertical agents so that we can really hand over initial few agents. Obviously they need to be customized for bank use cases, but we know these are high impact use cases. They're built as vertical agents. They're called Lumen Harper and Marvin, and those agents are the first point to begin with. We also, as I said, we also work with banks because there is some evolving technology. So we also help them figure out what's the best way forward, what are the use case you want to start with? But this is how basically by building the vertical reasons to reduce friction in getting started by solving a trade off of this thing. And second is helping them help work with them to help getting started.
Danielle Fugazy (30:04):
So just in wrap up, I would say, what advice would you give a bank that's looking to drive AI transformations across their institution?
Rajan Nagina (30:21):
Do you want to go? Sure, sure. Okay. So I think first is the first advice is it is very, very important to start because everything has a learning curve. Things like this, which are changing so fast, there is a path, different answer. So certainly it's important to get started. I just said starting with a value wherein the agents are almost prebuilt is a good way to begin with, or at least working with a partner that brings technology and knowhow. Both is a good way to get started. I think that's what I'll say. Get started. If you like us, please work with us to get started will help you out.
Varun Goswami (30:56):
Arun, and I'm a big fan of a platform based approach. I'm always like, don't go in for, let me just do this bit and this bit, it always ends up creating a mess in terms of maintainability, always first get a platform which has all of those capabilities built in which you can trust and then start building on top of that platform. So I'm a big fan of a platform based approach. What it ensures is it ensures your compliances, it ensures your security, it ensures your auditability, and then it also ensures that your time to market becomes much faster. So if you've got a platform, start building on top of the platform, start launching it, study it, learn from it, iterate. I think that's the best approach I would say share.
Danielle Fugazy (31:41):
Well, thank you both so much for joining me today and thank you to Newgen. Like I said, I really think that this is going to be very insightful for our audience. I've learned a lot today and I'm sure they did too.
Rajan Nagina (31:54):
Thank you so much. Thank you so much.
Danielle Fugazy (31:55):
Have a great day.
Rajan Nagina (31:56):
Thanks.
Danielle Fugazy (31:57):
Thank you everyone for joining us.
