In this episode of The Connector Podcast, hosts dive deep into the transition of AI in finance from hype to measurable value. They cite a MIT study showing a 93% failure rate of enterprise AI projects, framing the current "trough of disillusionment". To navigate this, they introduce BBD’s Time Filter Framework – three questions (necessity, automatable, delegable) that filter viable AI use‑cases. The conversation then contrasts generative AI (reactive content creation) with agentic AI, which plans, decides, and acts autonomously to execute end‑to‑end business processes. Two domains of agentic AI are explored. On the consumer side, a third‑party “customer agent” would use open‑banking APIs to optimise personal cash flow, debt, and investments, free from banks’ sales incentives. On the corporate side, they discuss a concrete case: Covecta’s partnership with Metro Bank, where agentic AI cut credit memo preparation from two days to 40 minutes and customer analysis from 90 to 15 minutes, and introduced a per‑task pricing model tied directly to outcomes. Supporting such agents requires robust orchestration to coordinate over 50 legacy systems. The hosts present BIAN (Banking Industry Architecture Network) as a universal translator with modular service domains, and the Business Object Model (BOM) as a standardized data dictionary that ensures clean, consistent data for AI and regulatory reporting. Regulatory pressures are highlighted: the EU’s Digital Operational Resilience Act (DORA) and AI Act (effective 2026) overlap on incident reporting, while the EU Accessibility Act (2025) adds mandatory accessibility for AI‑driven services. The episode also touches on the AI‑cybersecurity arms race, noting $73.4 billion projected AI spend in finance by 2025. Future profit opportunities are identified in wealth‑tech, where legacy client experiences lag digital expectations, and in social‑impact AI: OmniWave’s AI‑driven stable‑income engine for nonprofits and Finans’ AI‑personalised financial‑literacy chatbot for the Italian market. Finally, the hosts discuss the human professional’s evolving role—from transactional agent to trusted risk‑management and relationship expert—arguing that while agentic AI delivers efficiency, it cannot replace the fiduciary trust humans provide. Overall, the episode argues that real AI value in finance now lies in goal‑driven agents underpinned by standardized orchestration and governed by clear, overlapping regulations.
Chapters
Welcome back to The Deep Dive. For years, the whole conversation around AI and finance, it felt like it was just riding this massive wave of promise. You know, we were constantly hearing about the potential.
Right.
But today, well, that phase seems pretty much over. We're going deep into where the industry is actually finding, you know, verifiable measured value, moving past that early hype.
Yeah, and it's an absolutely essential shift. I mean, the landscape has changed dramatically. Lucky Nkosi over at BBD's Advanced Technology Center, he actually noted we've effectively officially passed that peak of inflated expectations. And you can kind of understand the pessimism. There's this MIT study. It found a staggering 93% of enterprise AI projects just fail outright.
93%. Wow. Yeah.
And that kind of failure rate, it doesn't just disappoint people. It drives the whole industry straight into what they call the trough of disillusionment.
93% failure. I mean, that's a pretty brutal statistic. And it tells you that just chasing the next shiny tech trend is, well, it's a good way to burn through cash. It sounds like these failures weren't just about the tech itself, but maybe more about integration or how they were actually implemented.
Precisely. Exactly. And that's why our mission for you today is really figuring out how to focus only on value-driven implementation. BBD actually offered a great, really practical tool for this. They call it the Time Filter Framework. And it's all about separating the signal from the noise, right? With just 3 simple, practical questions you ask for any potential AI use case. One, is this task something that absolutely Right, necessary. 2, can it actually be automated?
Okay.
And 3, can it be delegated?
Huh, simple.
Yeah. And if a project can't pass those 3 simple tests, you cut it.
That framework really forces you to focus then. It pushes you towards the tech that actually, you know, works. Which brings us to agentic AI. Now, for our listeners, I think the key thing is understanding this is really different from generative AI, which is, let's face it, what most people are interacting with day to day. So, what's the crucial difference there?
Right. So most people know GenAI, generative AI. It's fundamentally reactive. It's great for generating content, maybe drafting emails or acting as a kind of traditional chatbot. It's everywhere, but its impact on the core business metrics, the bottom line, it's often been, frankly, frustratingly limited. This is actually what McKinsey called the GenAI paradox. You see high adoption, amazing capabilities, but very low verifiable ROI that's tied directly to major business outcomes.
Got it. So GenAI is a great writer, maybe, but not really a great doer.
Exactly. AI is the doer. That's the key difference. It basically transforms GenAI from just a reactive tool into a proactive, goal-driven, well, like a virtual collaborator. An agentic system actually plans, it decides, and it acts autonomously.
To automate processes end to end.
Right. Complex business processes end to end with ideally minimal human intervention.
Okay, this is where the landscape gets really interesting. You mentioned 2 key areas for agentic AI. The Consumer side, you called it the customer agent utopia, and then the corporate operational side. Let's start with that utopian idea, a customer agent working just for me.
Yeah, the whole premise is pretty revolutionary because its mandate, its only job, is optimizing your financial health, full stop. It's never influenced by the bank's sales targets or their profit goals.
Okay.
So this agent, this third-party agent, it would use open banking APIs.
Right, to pull data from different places.
Exactly. To compare prices, fees, rates, you name it, across multiple banks. And its goal is just optimizing your cash flow, your debt, your investments.
I can definitely see why customers would love that. But hang on, if that agent's job is basically to steer me away from the bank's most profitable products, wouldn't the banks just fight that tooth and nail, try to block it?
Well, that's the friction point. Absolutely. That's the tension. But the need for this kind of thing It's undeniable. I mean, look at the data. In Belgium, less than 50% of households actually feel financially healthy.
Oh, less than half.
Right. So banks, they kind of have to adapt to a world where that customer inertia, you know, the tendency to just stick with what you have, which banks frankly rely on, that inertia gets eliminated by technology that acts entirely for the customer. If the banks don't offer this kind of optimization themselves, someone else, a third party, definitely will.
That makes a lot of sense. Okay, let's pivot then to the corporate side where the ROI seems much more immediate. You mentioned solving that infamous drop-off problem in lending where waiting times are just absurdly long.
Oh yeah, sometimes up to 180 days for a commercial loan.
Yeah.
Customers just get fed up and walk away.
180 days.
Exactly. And this is where Egentic AI is delivering real, like, structural efficiency gains. Take the example of Covecta. They partnered with Metro Bank to completely overhaul their commercial credit operations.
Okay, what were the results?
The results were pretty dramatic. Credit memos, these are critical documents. They went from taking 2 days to prepare down to just 40 minutes.
2 days to 40 minutes.
Yeah. And customer analysis, that used to take 90 minutes. Now it's slashed to 15 minutes.
Wow.
That massive reduction in friction, it means customers actually stick around. And the staff, they can spend their time on high-value judgment calls instead of just manual data gathering.
And what's really fascinating there is how the pricing model changes too. It's not the traditional way we think about buying software.
Not at all. Traditional SaaS, software as a service, it's usually priced per seat, right? So the vendor's revenue grows as the client hires more people. But these agentic AI platforms like Covecta, they're priced per completed task.
Per task, like per memo generated.
Exactly. Per credit memo, per reconciliation, per complex analysis performed. This directly ties the cost to measurable business outcomes. It incentivizes the current teams to deliver huge increases in output without necessarily growing the staff count.
That shift from headcount cost to outcome value, that's powerful. But okay, to support these really complex autonomous agents, you obviously need some serious architectural foundations underneath. We know banks spend a ton on the shiny front ends and the big core systems, but we're seeing data suggesting processes are actually getting more complex, not less. You mentioned needing over 50 systems for a core process now.
Yeah, over 50 systems. That's a 19% in just 5 years. It's kind of staggering.
So, how do we stop these smart agents just becoming like isolated islands, unable to talk to anything else?
Yeah, you've hit on the missing link. It's orchestration. Orchestration acts as the command layer, the sort of, uh, air traffic control that unifies all those 50+ systems, the data inside them, and the workflows between them.
Okay, the connective tissue?
Exactly. Without really robust orchestration, even the smartest agentic AI is basically paralyzed. It can't turn its decisions into actions across all those different departments and Orchestration makes sure outputs become actual outcomes. And crucially, it maintains operational control. That's a huge concern for firms. As they rely more and more on automation, they worry about losing control.
Okay, so orchestration manages the action. But if those 50 systems are all speaking different internal languages, like corporate dialects, the agents still can't communicate cleanly, right? That must be where BIAN comes in— the Banking Industry Architecture Network. You called it banking's quiet revolution.
BIAN is basically the universal It's like a universal translator for banking systems. Think of it like, like Lego bricks for banking tech. It provides a global standard, a shared language, and these modular building blocks, they call them service domains.
Service domains. Okay.
And they help systems talk to each other consistently, whether it's internal systems or talking to external partners. It's the framework that lets banks finally move away from those tightly coupled monolithic systems.
The old mainframes.
Right. Towards a truly composable sort of plug-and-play architecture.
I like that Lego analogy. Is this actually happening? Are banks using it?
adoption. HSBC, for example, they're using BEAN to basically future-proof their entire global tech architecture. And PNC, they used it to map their whole application landscape. The goal was to drastically cut complexity and importantly reduce vendor lock-in.
So the technical win translates directly into a business win there. Less complexity, more flexibility.
Absolutely.
How does BEAN specifically help clean up that data mess for the AI though? Because AI needs clean data.
Good question. That's the role of the business object Model, or BOM, within BN. So if BN is the standardized architecture, the blueprint, the BOM is the standardized data language used within that blueprint.
Got it. Like a dictionary.
Yeah, exactly. It ensures that every system— lending, deposits, compliance, whatever— interprets information the exact same way. Customer means the same thing everywhere. Transaction means the same thing. That consistency is just non-negotiable. It speeds up regulatory reporting massively. And crucially for AI, it provides the clean, reliable data needed for accurate decisions.
Okay, moving from standards to safety. As all this innovation scales globally, you need those twin engines, as you put it, efficiency from AI, resilience from cybersecurity. And the spending reflects that, right? Global AI spending in finance is set to hit, how, what was it, over $70 billion?
Yeah, $73.4 billion by 2025. It's huge.
And that growth feels necessary because we're kind of operating in an AI versus AI battleground now, aren't we?
Absolutely are. AI is now a critical defense tool. It's scanning millions of transactions, login behaviors, all in real time, trying to spot anomalies. But that same technology is being weaponized by the bad guys. We're seeing hyper-realistic phishing campaigns generated by AI, adaptive malware that learns as it goes.
So, the defense has to evolve just as fast as the threat does?
Constantly. It's an arms race.
And that kind of rapidly evolving, high-stakes landscape really demands a unified regulatory approach. Which brings us to Europe's big moves— DORA and the AI Act.
Right. DORA, the Digital Operational Resilience Act, that's kicking in January 2025. And then the AI Act follows, fully applicable August 2026.
Okay, so how do these 2 overlap? Because it sounds like they could cover similar ground.
They definitely do overlap, and that's crucial for listeners, especially in finance, to understand. Under DORA, AI systems are generally considered ICT assets or ICT services.
Okay.
And then institutions get classified. They're either providers if they develop their own AI or deployers if they're using systems from external vendors.
Provider versus deployer. Got it.
And both regulations mandate incident reporting. DORA requires reports for major ICT incidents. Think system outages. The AI Act requires reports for incidents involving high-risk AI systems. So firms really need to map their AI applications carefully to navigate these dual compliance requirements. It could get messy otherwise.
Right, of getting overwhelmed. And speaking of mandatory requirements, we should probably touch on accessibility too. The EU Accessibility Act, that's effective June 2025, right? Making accessibility a legal standard. That feels like a huge collision point with all this automation.
It's a critical point, and honestly, it's often overlooked in the big rush for speed and efficiency. Think about AI-driven onboarding, you know, using biometrics, automated background checks, It's fast.
Yeah, seamless.
Seamless.
Yeah.
Unless you're a client with, say, a visual impairment trying to use a retina scanner, or someone with motor limitations trying to navigate a complex automated process. If these supposedly efficient processes aren't usable by everyone, the institution risks non-compliance, big regulatory fines, and serious reputational damage.
So, you can't just measure success by speed anymore. Equity of access has to be built in from the start.
Exactly, across the entire digital journey.
Okay, let's shift gears slightly. Where are the next big frontiers for profit in fintech? We know that initial consumer gold rush, trying to get everyone onto a slick banking app, that's kind of maturing now. You point to wealth tech as the next major opportunity.
Yeah, absolutely. This is the huge, traditionally underserved wealth management sector, and the amount of money we're talking about is just staggering. Yeah, we expect something like £5.5 trillion to pass between generations in the UK and EU by 2030.
Wow, trillion with a 2.
Right. And the fintech pioneers who got really good at consumer banking, they're now turning their attention here because, frankly, the current client experience in wealth management is often, well, archaic. Functionally archaic.
It's that 3-click Tesla problem you mentioned. A wealthy client can configure and order a 6-figure car online in minutes, but just adjusting their investment portfolio takes, what, 3 weeks of back-and-forth emails and scanned PDFs?
Exactly! That jarring disconnect is a huge catalyst for change. Especially when you see stats like 67% of high-net-worth clients under 40 are actively considering leaving their current wealth manager specifically because of a poor digital experience.
The urgency is definitely there. But AI isn't just about chasing profit and wealth, right? It's also being used for social good— financial inclusion.
Yes, and that's really encouraging to see. Take OmniWave, for example. They use AI-driven analysis decoding real-time market sentiment, they claim with 93% accuracy, specifically to generate stable income for nonprofit organizations.
For nonprofits? How does that work?
They use proprietary tech to find stable income opportunities, basically de-risking investments for these organizations. They actually cited a 93.5% net annual return in 2024 for one case study.
93% return for a nonprofit?
That's—
that fundamentally changes the game for social impact investing.
It really does. And then think about the foundational problem, financial literacy.
Yeah, huge issue globally.
It's huge. There's an Italian startup called Finans. They're using AI as a social enabler, is how they put it, to democratize financial education. They're tackling that OECD finding that, you know, only a minority of Italians actually have the necessary financial knowledge.
So how does the AI help there?
Finans creates these personalized learning paths using algorithms. They use micro-content, bite-sized lessons, and they have a specialized AI chatbot built on over 170 different datasets to guide users through their financial journey. It's about leveling the playing field, giving more people the knowledge for economic participation.
What a comprehensive deep dive. I mean, we've covered a lot from the hype crash to agentic value, architecture, regulation. So to kind of bring it all home for you listening, right? The core finding really seems to be that the real value of AI in finance today is in these goal-driven agents that actually do things, but they have to be underpinned by robust standardized architecture like orchestration and BIAN, right? And the whole thing needs to be governed by clear, strict regulation.
And that framework, that whole picture, it raises the most important question, I think, for the practitioners listening. What does this huge structural shift actually mean for the human professional? If you look at the insurance industry, maybe it's a proxy. AI is already starting to automate underwriting, claims processing, even product recommendations. The traditional role of an agent as just a product seller, that's being directly challenged, maybe even replaced. eliminated in some areas.
So, the human professional's role has to transform completely then. It's moving away from just handling transactions or making sales towards becoming more like an integrated risk management expert and, maybe just as importantly, a trust creator.
Exactly. While agentic AI offers incredible efficiency, speed, data analysis, it can't replicate the human element needed in a fiduciary relationship. Not yet, anyway.
That trust factor.
Absolutely. The human professional delivers that irreplaceable value by Combining your expertise with, well, trust, empathy, creativity, especially in those really complex, ambiguous situations where AI just can't handle the nuances.
Yeah.
So the challenge for you listening isn't just about adopting AI. It's about deeply understanding its capabilities and its limitations so you can translate that efficiency for your clients, but also provide the critical human judgment, the empathy, the strategic thinking. That's the truly scarce resource looking forward. That's where the real value will be.
