Deep Dive: The Agentic AI Era in Banking

I’ll be blunt: who isn’t a little weary of the AI hype in banking? I watched Backbase's Engage 2025 event in London to gain insights into AI's role in the banking sector and what Backbase is developing. During the keynote address, CEO Jouk Pleiter asked the audience, "Who's a little bit tired of all the AI hype?" Numerous hands were raised in agreement. Despite the skepticism, he and others in the industry firmly believe that this wave of AI could be more significant than the internet itself. 

Why the gap? Partly because early efforts were siloed experiments, a fancy pilot here, a chatbot there that never touched the bank’s core operations. It’s been easy for executives to label projects as “AI” in presentations without truly scaling them. As Pleiter noted, we’ve seen the internet wave, then the mobile wave, and “we’re now in this AI-inside wave”, which is all about embedding AI into every aspect of banking. This means not just having AI as a side project in areas like fraud detection or marketing, but integrating it throughout marketing, sales, service, operations, compliance, and essentially every function. The next five years in banking will be defined by this infusion of AI into the fabric of how banks work. The promise is huge: imagine AI powering everything behind the scenes, turning banks into truly intelligent organizations. But to get there, banks must move beyond “AI lab” mode and bring AI into production at scale. How? That brings us to the new kid on the block: Agentic AI.

The difference between Generative AI and Agentic AI

By now, everyone and their grandmother has played with generative AI, you know, large language models that can draft an email or a haiku about interest rates. Even my own dad fires up ChatGPT to send me ridiculously well-written birthday messages. Generative AI (think GPT-4, DALL·E, etc.) is great at creating content. But here’s the problem: it’s mostly a brilliant talker, not a doer. It can generate a customer service reply or a financial report, but it won’t on its own go book a meeting with a client or approve a loan in your banking system. That’s where agentic AI comes in; it’s the difference between an AI that chats and an AI that acts.

In the Engage 2025 keynote, Pleiter described an “AI agent” as a virtual co-worker. Think of hiring a new employee: you define a role, train them, and let them loose on your backlog of tasks. An AI agent is the same idea, except it’s a tireless robot employee living in software. Instead of hiring more staff, you train an AI agent to do a specific job, and if the workload increases, it just scales up automatically, no coffee breaks or overtime required. These agents can become absolute specialists, I would say “superstars in what they do”, because they’re focused on one task and continuously learning at it. And unlike that one chatbot you had answering FAQs, an agent isn’t limited to spitting out text. It can be given access to tools and systems: schedule a meeting on an employee’s calendar, retrieve and analyze customer data, even initiate transactions with a human in the loop. In short, generative AI produces answers; agentic AI takes action.

Here’s a simple analogy: Generative AI is like a very smart advisor, it can tell you the best way to reorganize your filing system. Agentic AI is like hiring an expert assistant; it not only advises but goes ahead and reorganizes the files for you. Under the hood, agentic AI still uses generative models (the “brains” of the operation), but it couples them with context, memory, and connectivity into the bank’s internal systems. Backbase’s tech lead Deepak Pandey outlined how a banking-grade AI agent works: it can be triggered by events or API calls, consult an LLM for reasoning, look up internal knowledge bases, recall context from memory, and, crucially, invoke any relevant core banking APIs securely. In other words, the agent isn’t just chatting in isolation; it’s plugged into the bank’s data and processes, with guardrails.

This agentic approach is what turns AI from a clever toy into a transformative force. Early AI efforts were often point solutions, a fraud detection model here, a chatbot interface there that had limited synergy. By contrast, an agentic AI platform treats AI as an “operating system” for the bank. Instead of isolated pilots, the AI is embedded across channels, front-office, mid-office, back-office, the whole stack, and orchestrated in a unified way. The goal is “front-to-back orchestration” with AI agents, moving from scattered point solutions to an AI-native operating model. It’s like the difference between having a few smart gadgets versus a smart home that’s centrally managed. One is neat; the other is a game-changer.

Real banking use cases powered by Agentic AI

Enough theory, what does this look like in practice? Let’s talk about how agentic AI changes the game for everyday banking use cases. Consider a typical customer scenario: Emma is applying for a personal loan through her bank’s app. Normally, Emma would fill out pages of forms, upload a bunch of documents (pay stubs, IDs, proof of address…), and then wait. And wait. Often she’d get an email later saying something like, “Oops, we actually also need your bank statements,” or “That document wasn’t the right one, please upload another.” We’ve all been in that maddening loop of requests and resubmissions. It’s not just Emma who hates it; the bank’s employees aren’t thrilled either, spending hours checking those docs and ping-ponging emails.

With an agentic AI platform, that use case changes. The moment Emma begins her loan application, the bank’s process orchestration kicks in. As she fills in her details and reaches the document upload step, the AI takes action. Behind the scenes, an “Intelligent Document Agent” automatically reviews the pay slips she uploads to verify they’re correct and complete. Instead of Emma waiting days for someone to glance at her paperwork, she gets instant feedback, a quick “Got it, documents received!” or even a prompt if something is missing, right in the app. No more endless back-and-forth or radio silence.

Once Emma submits her application, the agent doesn’t just approve it. It hands off to a human loan officer with the main work already done. The bank employee gets a ping in their case manager dashboard that there’s a loan application ready for review and finds that the AI agent has pre-summarized Emma’s pay slips and highlighted any anomalies for attention. In other words, the AI did the first pass of underwriting, and the officer can focus on the edge cases or final judgment instead of reviewing 50 pages of PDFs. Emma, meanwhile, isn’t left hanging. In record time, she gets the happy notification: “Your loan is approved!” The entire process that used to take days or weeks is now maybe a few hours or even minutes end-to-end. Activities that used to take hours are reduced to minutes, resulting in happy customers (“faster time to yes”) and more efficient use of employee time.

This isn’t a one-off case; it’s a template for how agentic AI can streamline many other use cases. Consider mortgage origination, historically one of the most time-consuming banking use cases. A single mortgage application can generate a 500-page file with 50+ documents, and require 10-15 hours of manual effort from staff to review and verify everything. Now imagine plugging in specialized AI agents at each stage: one agent verifies the documents the moment they’re uploaded, another agent pulls in credit history and valuations instantly, and another handles all those repetitive data-entry steps. Suddenly, the bottlenecks start to disappear, tasks that consumed many hours are reduced to a few clicks and seconds of compute time. Robert Soetens noted that when you stack up the numbers (cycle times, first-pass success rates, etc.), the impact of these AI helpers is massive. We’re talking about significant improvements, not just 5% here or 10% there.

And it’s not only about speed, it’s about experience. When agents are at work, customers feel guided, not lost. The AI can nudge them in real-time (“Hey, your pay stub is a bit blurry, mind uploading a clearer photo?”) so they get things right the first time. That means far fewer frustrating redo requests and higher first-pass success. Meanwhile, your front-line employees (whether in a branch or call center) are no longer playing “document chase” or copy-paste clerks. As one speaker quipped, the focus shifts away from being a communications middleman or data-entry drone, and gives them back time to actually provide consultative service. The AI agents take over the repetitive work, letting the humans focus on the human stuff. It’s as if every employee suddenly has an intern (a supremely competent, never-sleeping intern) handling their busywork.

Conversation as an Interface

As we all know, banking interfaces have evolved from branches to websites to mobile apps. Agentic AI introduces a fourth and interesting shift concept banking as a conversation. Instead of tapping through screens, customers will increasingly interact through natural language, voice, and intent. This way, the customer experience becomes natural, like an everyday conversation. AI agents will not just answer questions; they will take action on behalf of the user, pulling data, running checks, initiating workflows, and completing tasks with near-instant time. Need to increase your credit card limit? Just ask the app in plain English; the AI agent can pull up your account, run an eligibility check, and process the change, all in one seamless interaction. Backbase even suggests we’re headed for a “third wave” of user interaction, beyond websites and mobile apps, into voice and dialogue as the primary interface. And because an agent can tap into all your data and tools, it’s more like talking to a really competent bank employee than today’s dumb chatbots. Or consider a personal finance coach agent: living in your banking app, it observes your transactions (with permission) and proactively gives you tailored advice. It’s not just spitting out generic budget tips. It knows you, it can pull in your data, maybe even talk to other agents (like a market data agent or a credit score agent), and then say, “Hey, you can save $500 by refinancing your car loan, want to explore that?” Truly personalized service at scale, delivered by an army of digital specialists.

Future implications for customer experience and operations

What does a bank look like when you multiply these AI agents across every function? In a word, transformative. We’re talking about a future where your bank runs on a human+AI collaboration model at every level. Jouk Pleiter painted a picture of a “holistic conductor,” a platform orchestrating thousands of specialized agents in concert, like an AI-powered symphony running the bank. In this future, every task that is labor-intensive today could be augmented by an agent. The agents handle the heavy lifting, and humans provide oversight, creativity, and empathy. Rather than replace people, the AI agents become their coworkers. Your top relationship managers and support reps might each have a handful of AI assistants: researching customer data, drafting personalized recommendations, scheduling follow-ups basically supercharging each employee’s productivity. In fact, Backbase anticipates that a skilled employee working with a dedicated agent (an “agent assistant”) could at least double their productivity, and one supervising an “agent factory” (a whole team of AI helpers) could see 10x productivity gains. Ten times! That’s not just doing things a bit faster; that’s a different universe of operational efficiency.

For customer experience, the implications are just as exciting. Banks have talked about personalization forever, but agentic AI makes hyper-personalization feasible. Instead of segmenting customers into broad buckets, you can serve a “segment of one”, truly tailoring offers and advice to each individual in real time. AI agents will crunch through data to understand each customer’s needs, behaviors, and goals, and then automatically generate the next best action, whether it’s a product recommendation or a helpful insight for that specific person. The result? Customers feel like the bank really knows them (because, in a way, it does) and is looking out for their financial well-being.

On the UI side, as mentioned earlier, we might finally break free from the traditional menus and forms. Conversational interfaces, chatting or speaking your request, could become a primary way people interact with banks. It’s a full-circle moment: banking started with conversations (with your branch manager or advisor), and with AI, it might return to conversational, but delivered through a digital medium with superhuman intelligence on tap. Imagine saying, “I’m planning a trip to Japan, what’s the best way to pay there?” and your banking app’s agent not only tells you about foreign transaction fees, but also offers to set up a low-cost travel card and alerts you to any travel advisories or insurance options. All in a quick exchange, no manual research needed.

Meanwhile, behind the scenes in the operations departments, agentic AI could finally conquer the “dark and difficult places” that have resisted automation. Think of all the tedious, error-prone processes that still exist in banks: from reconciling payments, to reviewing flagged transactions for compliance, to updating legacy system records. Many of those could be handed off to specialized agents. The future bank’s ops team might be a fraction of the size, but each person is managing a fleet of digital workers. And crucially, humans will still be in the loop where it matters, they’ll just be decision-makers rather than data-entry clerks. The AI will prep the case, do the number crunching, maybe even recommend an action, and the human gives the yea or nay. This not only boosts efficiency but also makes the work more meaningful for the humans involved. They get to focus on exceptions, relationship-building, and strategy, not mind-numbing paperwork.

Of course, moving to this model raises new challenges: banks must ensure these AI agents are reliable, transparent, and fair. Trust is essential in financial services. The good news is that agentic AI doesn’t mean uncontrolled AI. We’ll likely see robust governance and “human guardrails” as part of any deployment. The Backbase team emphasized that in their architecture, everything the agent does can be monitored and governed, with an AI gateway to control model usage and filters on what data goes in or out. Banks will be able to tweak how much autonomy an agent has, require approvals for certain actions, and audit decisions after the fact. In other words, the future is augmented banking, not autonomous banking, AI helping humans, not replacing them entirely.

All told, the customer experience will become more seamless and proactive, and internal operations will become leaner and smarter. Banks that embrace agentic AI could operate at a significantly lower cost-to-serve while increasing customer satisfaction. It’s a classic have-your-cake-and-eat-it-too scenario: better service and better efficiency. But it won’t be evenly distributed, it will favor the banks that move early and decisively.

Disclaimer:

Fintech Wrap Up aggregates publicly available information for informational purposes only. Portions of the content may be reproduced verbatim from the original source, and full credit is provided with a "Source: [Name]" attribution. All copyrights and trademarks remain the property of their respective owners. Fintech Wrap Up does not guarantee the accuracy, completeness, or reliability of the aggregated content; these are the responsibility of the original source providers. Links to the original sources may not always be included. For questions or concerns, please contact us at sam.boboev@fintechwrapup.com.