EP1036: Why AI-Ready Infrastructure Matters for GCC Banks

I'm Puja Sharma of IBS Intelligence, and you're listening to the IBS IViews podcast. With me is Senna Bahadur, Senior Sales Manager at GCC for Fimple. Welcome to the podcast, Senna. Thank you so much, Puja.

Nice to be in this podcast today. Senna, as AI adoption accelerates across the GCC, how can Fimple's cloud native and API-first banking platform help banks integrate AI capabilities more effectively? I think the AI conversation in banking is starting to change in the GCC region. For the couple of years, everybody was asking, what can AI do?

A very simple and basic question. But now, when we sit with the banks, the decision makers, and tons of the meetings, the discussion is becoming much more practical. How do I actually put this into production? How does it access my banking data?

How does it interact with my core? And probably the most interesting question all of is what happens when AI stops sitting outside the bank and actually starts participating in how the bank operates. So that's where it gets exciting for us at Fimple. Because in Fimple, we are not just built as a traditional core and then later connected to APIs.

It was purely designed from day one as a composable cloud-native API-first banking platform. So every banking module can be accessed through APIs and SDKs. And that sounds a bit technical, but the practical impact is actually very simple. So let me give an example.

Let's say an AI agent analyses a corporate customer's financial statements and comes back with a credit recommendation. That's great. But what happens next is does someone copy that result into Excel, send it to another department, re-enter the numbers into another application, wait for another system to create the limit, because if that's what happens, you've made the analysis just smarter. But you haven't really made the bank smarter.

So the opportunity we see is different. The AI can analyze, it can structure information, it can make a recommendation. That recommendation can enter an actual banking workflow. And finally, a human can review it and approve where required.

And then the core can execute the banking action. So that could mean creating a facility, updating a limit, triggering a payment workflow, and routing an exception or starting another control process. And this is really where our thinking around the agentic bank comes from. We are not talking about putting a chatbot on a mobile application and calling that an IR transformation.

We are thinking about intelligent agents being able to participate in credit, operations, payment, and other banking processes. And also, this is, I think, very important, with still humans are in the loop, especially for the decision-making part. FIMPLAI is designed around that principle. Agent actions are traceable, teams can review and approve them, and manual and agentix steps can exist in the same workflow.

And I think that model fits the GCC particularly well, because banks here are very ambitious when you compare with the other markets. They want speed, they want innovation, they want to launch the new propositions very quickly. But they also operate in highly regulated environments where control, auditability, and data governance matter enormously. So the answer isn't uncontrolled automation, it is controlled intelligence.

And there is another reason why architecture matters. AI is moving unbelievably quickly. Also, probably you are seeing it in the market as well. The model a bank chooses today may not be the model it wants three years from now.

So you don't want to build your entire banking strategy around just one model or just one provider. You want the flexibility to plug in new capabilities as they emerge. That's what an open architecture gives you. So when we talk about AI at Fimples, we are not really asking how do we put AI on top of a bank?

We are asking a more critical and smarter question. How much of the bank itself can become intelligent? Senna, what are the most impactful AI use cases in banking today? And how can banks leverage Fimple's real-time core banking infrastructure to maximize their value?

Actually, there are so many AI use cases being discussed today that I think banks have to be a little bit selective. Because some demos are very impressive, but then when you sit with the CEO, the head of credit, head of risk, chief compliance officers, and the question becomes much simpler. What does this actually change for me? Does it reduce the cost?

Does it improve the risk? Does it create a revenue for me? Or does it materially improve turnaround time? And one area where I think value is very easy to understand is corporate lending.

Anyone who has spent time around corporate credit knows how much manual work can sit behind just one credit decision. A company sends an audited financial report, maybe it's more than 60 pages, 200 pages, and somebody has to find the relevant statements, capture the financial figures, and map them into the bank's chart of accounts. And it's not finishing. Reconcile the balance sheet, calculate the ratios, look at the leverage, cash flow coverage, and collateral.

Then prepare the information for the credit team. One of the agents we are already working with at Fimple tackles exactly this. Our independent auditory report agent can read the financial statements, extract the relevant information, and map the line items into the bank's chart of accounts and reconcile the financials. In the use case we've modeled, an analyst may be reviewing maybe more than 150 pages, and manually entering around 200 line items and mapping roughly 130 lines.

But with the agent, that moves from potentially hours even up to a working day toward just a couple of minutes actually. And to me, the most interesting part isn't that AI can read the PDF. It's a basic OCR capability. That's already becoming quite normal.

The interesting part is what happens after the PDF. The structured, validated financial information can flow directly into spreading and credit decisioning. No exporting and less manual error. And suddenly your credit analyst is spending less time about typing the numbers and the repetitive tasks and start to ask more strategic questions.

Do we actually want to lend this company? That's much more better use of a human being. That's what we believe actually. And also, we are building further in that direction as well.

Our roadmap is including corporate lending limit allocation agent that can bring together financial statements, audit reports, external bureau information and everything in the same page. That's where it starts becoming really interesting actually, because now you're not automating just one task. You're beginning to rethink the entire credit workflow. And another area that I can give an example is for the early warning risk.

Imagine a borrower is not yet in default, okay? And nothing dramatic has happened. But over three months, account turnover is declining, balances are weakening, repayment behavior is starting to change. There may be a negative bureau information as well.

But maybe their sector is also under the pressure. A human may eventually notice all of these things, and AI can continue to look at them together. FIMPOL's roadmap includes an early warning agent that combines repayment behavior, overdue patterns, credit intelligence, collateral top-up, and limit freeze. And if you are relationship manager, the timing matters enormously as well.

You don't want the system to tell you the customer is in trouble once the customer has already defaulted. You want to know when there is still time to do something about it. And also, there is the revenue generation part as an example. Let's say that a customer may already be banking with you, maybe they are using deposits and payments, and their turnover is growing.

Their transaction profile is changing. And companies with similar behavior may typically use trade finance or working capital. So why wait for the customer to figure that out and come to you? So FIMPOL's roadmap also includes the next best product agent looking at current product holdings, transaction and channel behavior, lifecycle information, and product eligibility.

So that for me is where AI becomes commercially interesting for the banks because they start to act earlier, earlier about the opportunity or about the risk or about the customer journey. And when the AI sits close to a real-time banking platform, the distance between recognizing something and doing something becomes much shorter. How can GCC banks combine AI-driven customer experience with Fimple's composable banking approach to deliver personalized financial services at scale?

I think personalization is about to mean something very different in banking. Today, quite often, personalization means marketing actually. Let's say Senna is an affluent customer. Let's show her an affiliate credit card.

Or this customer traveled recently, and a frequent flyer, maybe. Let's just label them and let's show them a travel product. That's useful in a certain level, of course, but this is not really what we understand as the personalized banking. That's just the personalized communication, actually.

The much more interesting question is: can the actual financial proposition become personalized? And that's where AI and composable banking start to work really well together. Let me give an example on it. Take two SME customers, okay?

And both came to the bank asking for, let's say, 500,000 diram. On the surface, that's the same requirement. But one company has a very predictable monthly revenue, very stable supplier payments, very little cash flow volatility. But the other one is highly seasonal, let's say.

Maybe their industry is hospitality, retail, or tourism. They are very common industries in the region, especially in the GCC. And its cash comes very differently during the year. So, all in all, these two companies they need the same amount of money, but do they really need exactly the same product?

The same pricing? The same repayment structure? No. AI can help the bank understand that context.

But when you hit the next question, can the banking platform actually respond to it? And this is where composability is very important. So Fimple lets institutions work with modular banking capabilities rather than treating the core as one large fixed and boring block. The platform is designed around configurable business flows and product capabilities, with tools such as the process designer and the transaction composer intended to accelerate the creation and modification of the workflows.

In a more intelligent banking model, AI can help answer what the customer actually needs and the platform helps the answer. How do I deliver that in a controlled way? Controlled part is highly important in here, Puja, because AI doesn't suddenly replace credit policy. It doesn't replace pricing rules.

It doesn't replace the risk appetite. And particularly in this region, it doesn't replace Sharia requirements where you are operating Islamic banking. So Fimple supports both conventional and Islamic banking alongside multi-currency and multi-entity models. And I think the GCC is a particularly interesting place for this because the customer base is so diverse.

Retail customers, SMEs, large corporates, Islamic customers, different nationalities, cross-border needs. So the idea that you're going to serve all of that complexity using a handful of rigid products becomes harder and harder. We also see this in the architectures institutions are beginning to build. In one of our UAE banking as a service implementations, the institution can operate its own direct banking channels while also enabling fintech partners through the same underlying banking platform.

Each fintech proposition can operate on regulated banking rails without requiring an entirely separate core stack. Now, combine that type of composable architecture with AI. Suddenly the bank can become much more contextual. Which product for which customer, at what moment, under which pricing and risk boundaries?

So that's much more interesting to me than putting a customer's first name at the top of the app. I think today we personalize this message. And increasingly, we are going to personalize the financial proposal itself. And I believe that's a much bigger shift.

What role does AI play in fraud detection, risk management, and compliance? How important is a modern platform in enabling these capabilities? I actually think this is one of the areas where AI will become impossible for banks to ignore because fraud itself is changing as well. It's becoming unfortunately faster, unfortunately more sophisticated and more behavioral.

And increasingly, we are talking about things like deep fakes, account takeover, and social engineering attacks. So eventually, uh fraud defense has to operate at a similar speed. Traditional rule angles still matter, of course. Let's say if a transaction exceeds a limit, we are flagging it.

If it's coming from a certain geography, you're definitely looking at it. Those controls are not disappearing. But anyone who has spent time around fraud or AML operations knows what happens when you rely too heavily on static rules. The answer is the false positives, uh actually, a lot of them.

And a false positive is not free because someone has to review it, someone has to investigate it, operations get in gets involved, compliance gets involved. So sometimes the customer gets contacted as well. So the detection problem quickly becomes an operating cost problem as well. In that point, AI gives you the ability to look at the behavior rather than the only thresholds.

Let me give an example again. Let's say a corporate customer routinely sends 3 million Dijham payments, okay? And this transaction might be completely normal. But suppose this one happens at 2 a.

m. from a device we have never seen before to a beneficiary the company has never paid, and maybe it's preceded by an unusual logging pattern. So none of those things on its own necessarily means fraud, of course. But altogether, the context is a bit interesting.

And that's exactly the sort of pattern machine learning can look at very effectively. Our AI roadmap includes the fraud detection and action recommendation agent looking across payment activity, digital banking sessions, device and location patterns, beneficiary history, and behavioral indicators. But here is the bit I think gets overlooked. Detection is only the half of the job.

Because after this, will you block the transaction? Will you hold it? Will you send it to fraud operations? Or will you allow it but increase the monitoring?

So that's where you need the banking workflow underneath the intelligence. And you need to understand why the agent has made that recommendation. Because a bank cannot realistically say the AI thought it looked suspicious. There has to be a reason.

There has to be a traceability, accountability. So that's why we are deliberately building Fimple AI around human in-the-loop processes. Of course, the automation where it's it adds value, but the human review where the judgment and accountability matter. We have another example already in our AI work that shows this nicely.

Banks receive court and agency notices that may contain some liens, debtor lists, and multiple individual records. Today, someone may still need to read that document, extract the records and key them into the system one by one. But the FIMPL customer intelligence agent can extract the records from a document, generate confidence scores, protect the sensitive information, and prepare the data. But every row remains human-approved before it's committed.

That's the kind of AI I think the banks will increasingly be comfortable with. So it's not the uncontrolled autonomy, it's the controlled autonomy. How can banks ensure data readiness and operational agility for AI initiatives? And what advantage does FIMPA's cloud native architecture provide in this journey?

I think this is the slightly less glamorous side of AI, but maybe the most important one. Because everyone wants to talk about the models. Very quickly, though, almost every serious AI discussion becomes a data and architecture discussion. Because most banks don't have a shortage of the data.

They have an enormous amount of data. The problem is that it's everywhere and it's not very controllable. Customer information may be in one system, lending is somewhere else, CRM is somewhere else, trade finance is somewhere else, uh, documents is somewhere else. So technically the information exists, but the context, the operationally, that part is still very difficult to see.

And AI exposes that problem very quickly, actually, to the surface. This is one reason FIMPL's architecture is relevant. It's microservice-based, API-first, modular, and cloud native. So we are working with banking capabilities that are designed to integrate rather than being locked inside one tightly coupled environment.

But there is another point here that I think is particularly important for established GCC banks. I don't believe that every institution in here needs a complete massive core transformation before it can start becoming more intelligent. Because uh, in the field, sometimes people hear core modernization and immediately imagine, you know, up to three years, huge migrations, hundreds of integrations, and very high operational risk. We can understand, we can make the empathy with the customer in that part.

Because uh, of course, core transformations can be complex time to time. But banks can't simply switch off a system that has been running for their business, let's say, minimum 20 years. There is the customer account, payments, interfaces, operational processes, regulatory reporting, and everything. But it doesn't always have to be a big bank for this core modernization.

So FIMPL supports coexistence with the legacy environments and parallel running capabilities and phase migration, which means product by product or branch by branch migration. And I think that's very important because maybe the bank doesn't want to replace everything today. Maybe it wants to modernize only lending, only trade finance, or maybe wants to build a banking as a service based proposition. Actually, they can start from there because we have this modularity and composability.

We also have a good example in our wider customer base where a financial institution integrated FIMPL's trade finance capability into an existing legacy core environment in around two months rather than replacing the entire banking stack. Looking towards the future, how can partnership with technology providers help GCC banks evolve into fully AI-powered financial institutions? I think this is where the story gets really interesting. Because I don't believe the bank of the future will be built by only one technology provider, as I highlighted.

And I certainly don't think any bank should try to build all of it internally. So there are going to be specialists, let's say, in fraud, in identity, in payments, in customer engagement. So I think the competitive advantage starts shifting. It's much more about how quickly can I bring the best technology together.

And we are already seeing that ecosystem model in GCC as well. Let's say across our regional work, FinPool is supporting digital banking, Islamic inconventional banking, corporate channels, trade finance, treasury, and banking as a service propositions. One of the UAE models we are working with is particularly interesting because the financial institution is not only running its only banking activity, it's also enabling fintech partners through banking as a service. Now also imagine adding intelligence to that model.

A corporate client sends its financial documents, one agent structures the financials, another capability checks external information, a credit agent supports risk assessments, and a human credit officer reviews the recommendation. So the banking platform creates the facility, then the account continues to be monitored. Six months later, the customer's behavior starts to change actually. An early warning agent sees it, the relationship manager gets notified, or the opposite happens.

The customer is growing, transaction volumes are increasing, so the bank identifies a working capital or a trade finance opportunity before the customer asks. So now AI is no longer one project inside one innovation department. It's beginning to appear across credit risk operations, payments, customer servicing, and everything. And there is another reason why I think FIMPL is quite unusual in this space.

Because we are applying the same agentic thinking to our own software delivery lifecycle. We are working with agents across feature development, bank analysis, testing, support, environment monitoring, code review, of course, with the human approval built into the process. And I think that matters because we are not simply telling banks you should become AI native. We are trying to apply the same thinking to the way we build and operate the technology for ourselves.

So when we use the term agentic bank, we mean something much broader than an AI assistant. We mean a bank where intelligent agents, human teams, and specialized technology providers can work together around the modern banking platform. Senna Bahadur, senior sales manager at DCC for Fimple.