← All episodes

Inside Banking AI: The Real Test of Transformation.

Mike Parsons HostRay O’Brien Advisory Board, FlowX.AI

Watch the full episode, or read the conversation below.

Key takeaways

  • Banks are all in on agentic AI, with reservation: many “let a thousand flowers bloom” projects had a business case that didn’t stack up and are being shut down, so executives now ask what they actually got back.
  • A personal chatbot can be 80% right; a bank cannot. Once agents work in sequence and make decisions about people, every decision has to be explainable, auditable and transparent, with ethics injected so forbidden drivers such as postcode are never used.
  • History is a warning: telephone, internet and app banking were each meant to replace the branch, yet every channel stayed and the cost base grew. The test for agentic AI is whether it retires legacy or only adds to the stack.
  • The winners start small on processes outside the core, build the data foundation and governance first, and keep AI sovereign: a narrow model, trained only on the bank’s own data, inside its own environment.

Never this fast, and plenty of froth

Mike

Welcome, I’m your host Mike Parsons, and again we are here at FlowX.AI headquarters where we are going to get the big picture. The big picture indeed, particularly around banking. It’s a very exciting moment because what we get to do is figure out what’s happening right now and what’s next. This is really a conversation to figure out what’s happening. Never before have we been in a moment where things move so quickly. And the great news for us is that we’re joined by an expert that has seen it all before. He has been there, he has done that. That’s right, we are joined by Ray O’Brien, advisor to FlowX.AI and somebody who has a wealth of knowledge and experience to share with us as we try and figure out how do we accelerate the enterprise? How do we bring the words banking and agentic closer together? Ray, welcome.

Ray

Thank you very much. Absolute pleasure being here.

Mike

It is good to be together again, Ray. We have met in all four corners of the world and so much has changed. So much has revealed even in the last few weeks. It feels like every week there’s new banking technologies, new large language models. In your thirty plus years working in banking and technology, have you ever seen it like this?

Ray

I think the acceleration this time round is higher. I have seen huge momentous moments. There’s many things I could describe through my career. The introduction of the internet, the dot-com bubble, the implement of client server technology, the implementation of virtual data centers, the creation of SaaS applications, many things that we’ve seen and happen. But this time around, it is bigger and it’s more changing to so many different aspects.

Mike

The velocity, it’s the velocity of it is and the sheer volume of it. You mentioned the bubble. I remember that moment very vividly. I’d relocated from Sydney to Amsterdam and I arrived. To do a startup, to lead a start up in April of two thousand. In March of two thousand, the NASDAQ managed to lose ten percent, fifteen percent of its value in one day. So my timing was a little off. I hope it’s better now. But you did mention the bubble. And there’s a lot of people trying to connect the dots on where we are now with AI, bubble, yes or no. What are your thoughts?

Ray

I think with any new technology, there is an element of reality and an element, and I apologize, of marketing. And right now, I’m afraid the marketing element is quite high. There is an underlying element of really good use cases and really valuable things that organizations will be able to use. But there’s also a hell of a lot of froth out there. And I’m afraid there will be a reduction of that frothiness in can I give you a date? No, but it will happen at some stage. If you look at what happened with the dot-com bubble, I mean economically what happened was there was a number of companies who were launching with no business case whatsoever. The one I always loved was pets dot com. Pets.com was a beautiful example in the dot-com bubble where they decided that everybody in the world were going to buy their pets online and that their business case was the percentage of the world that were going to buy pets through pets.com. Of course, what happens was human beings like to actually meet their pets and touch their pets and make friends with their pets before they buy that doggie. And Pets.com didn’t actually work that well in that way and it crashed and it burnt. But I remember being over in California meeting the people from Pets.com and me saying, are you sure? And they’re saying, you’re a Luddite, Ray. You just don’t understand. The whole world’s changing. And everything’s going to be online. Now they actually were slightly correct because if you look at something like Amazon and these other retail stores and how huge and ubiquitous they are now, they were correct. But to be fair, people still do not buy pets online.

Mike

This is such a good point. It reminds me of all the promises around that first internet bubble. What was really missing and what was inaccurate was the timing. Because I remember being in the mid-2000s, the mid-90s, and I was like, actually, a lot of the things we thought would happen in 2001 just seemed to happen about five years later.

Ray

Yes.

Mike

It took some time.

Ray

It took time. It took time. If you look at the history of Amazon and these other companies, you know, starting as a book company, it took time. For those companies to grow and become, and I was thinking the distribution networks and everything else they built. My gosh, it’s like an enormous thing. It’s not just a technology thing they did, they created a complete seismic change in how people purchase one click, how they actually. Get items and their distribution channels and what they built. That takes time. That takes a lot of time.

AI gets its browser moment

Mike

So talking about time, I would love to know the moment when you saw AI, maybe a large language model, something agentic working. And it was your aha moment. Do you remember? Was there a particular moment where you went like, oh, hang on, this is going to be big?

Ray

I have a very sad fact about myself. I’m one of these people who started my career in 1989 but I did a degree in computers and maths and I specialized in a really boring topic back then it was called machine learning using using Prolog and Lisp and I built some of the first machine learning models and so when AI started becoming to the fore in recent years, I just sat there watching it going, oh, that’s nice. The world has discovered what artificial intelligence could be. And it is a slightly misused word, AI. Everybody somehow correlates AI to just LLMs, which for me, it’s like saying. Let me take you back to the internet. I am of that age group where I remember the internet when you used to actually have to do X.25 packets to each other to talk to each other. And you might be lucky and actually create a real basic email system. Then along came a thing called Spry Mosaic and a browser. And the world discovered the internet.

Mike

Yes. And just on that, though, you mentioned Mosaic. What we should, for many of our viewers, won’t remember Mosaic, but we should be clear: that’s before Netscape.

Ray

Oh, yes. Oh, yes.

Mike

And it put some sort of a UI on the web.

Ray

Exactly. It was a way for the. Humankind to all of a sudden start using this communication. And it was a browser. It was the first ever browser. Before that, you actually had to be a programmer. You actually had to be a geek to actually use the internet. It did exist. And we did build applications using the internet, but it was more for technologists as opposed. But then along came the browser. The same with LLMs. AI and the basic concepts of AI and the different types of models underneath AI have been existing for like 50, 60 years. But along comes LLM. Creates a browser interface and the world discovers AI, which is wonderful. For me, I love it. It’s great. It’s great to see so many people using AI and building on AI and coming up with thought processes on AI and coming up with. New and creating things using it. That is amazing. The underneath bits of it, actually, the maths hasn’t changed. It’s still the same maths. Yeah, it’s got better. They’ve got better algorithms. They’ve got better machines, better CPU, better disks. They can store more information. They can do sorting and heaping on a lot more data. But the actual mathematical concepts are the same.

From personal AI to institutional AI

Mike

Which is beautiful because if you master those foundational ideas, then you can really start to apply it into different areas, which brings me to the enterprise where do you feel? I would love to get a bit of a map of where we are right now with. Agentic AI large language models, when we apply it into the enterprise, I think it’s important to mention you’ve spent a lot of time working with or in large banks such as HSBC. So you have this unique perspective. Help us understand as AI is moving out of the nice little personal chat bot and it’s going into the workflow of the enterprise, it’s a move from personal individual AI into institutional AI. Dramatic shift. There needs to be a big bridge built here. But where are we in this moment? Can you help draw this map for us?

Ray

I think you’ve said it beautifully there. So. The vast majority of AI use cases in the world is that kind of personal usage where you bring up a chat bot, you talk to it, you ask it a question, it comes back with a response, you’re isolated, and you don’t mind if it gets it slightly wrong or that the answer might be slightly incorrect, you can change it, it’s only affecting you as a human being. You can change the the output, et cetera.

Mike

The famous like 80% good enough.

Ray

80% is good enough for most humans, it is.

Mike

Yeah.

Ray

Now you transfer that into the corporate environment. 80% is not good enough. And actually, it’s not an individual anymore. It actually has to work in sequence. With lots of other bots and lots of other processes. You have to bring all the data together. You have to make sure you have one source of the truth. You then have to actually make sure that these agents are not hallucinating, that they’re transparent, that they’ve got governance around them, that the inputs and outputs are actually. Defined and that people can audit that. This is huge. Also, you need some kind of framework around those things because it’s no longer just one person using it, it’s actually a whole process using it. And it has to be repeatable.

Mike

It’s really interesting. So you’re mapping out. What the reality of the mortgage process looks like. And what I’m keen to understand is: what do you see when you talk to large banks when they’re thinking about bringing AI closer, not just to summarizing a document, but perhaps making a decision? Where are we now, and what’s the tension inside of that?

Ray

Perfect question. So, if you’re adopting an agentic agent or some AI to maybe scan some documents and merge things together to optimize a piece of a process internally, yes, that can be done, but in a governed, coordinated manner. Again, you need to have many of them, you can’t just have singularities, they’ve got to be repeatable. But as soon as you get into making a decision about a human being’s life, like if I told you that you’re getting a mortgage, you’re very happy. If I say no, you’re not getting a mortgage, you immediately like to know why and who made that decision and what was the basis of that decision and what data did you use to make that decision and were you biased in that decision and is it ethical and I’d really like to understand how that decision is made. These problems have faced banks for decades. It’s not new. So if you look at any kind of credit risk modeling, any kind of capital calculations, this model governance and this kind of implementation has been around for decades, where it has to be explainable. It has to be auditable. It has to be transparent. You can’t just say a black box gave me an answer. It’s not acceptable.

Mike

And heaven forbid that when we ask it the same question again, it comes back with a different answer.

Ray

Even worse. And you cannot implement. And I love it when. People from a non-financial background go, but Ray, why can’t we just implement the model? And I’m going, because I need to explain it. I need it to actually be completely transparent. I need it to have a governance framework. I also need to inject ethics into it. So I’ll give you another example. Let’s take the credit lending space, lending money. If you put in postcodes into a credit model, you would immediately pick certain places that you don’t want to lend money to. But what happens in reality is you create ghettos. That is not good for humankind. So, in most countries in the world, it’s actually illegal to use postcode as a driver for making a credit decision. And there’s loads of things: the color of your skin. You’re in a lot of places what sex you are, etc. Etc. These are all ethical decisions. Mathematics has none of these ethics. If you leave it purely to the maths, they will just use whatever driver they think is the most appropriate to get an answer. You have to inject ethics. So it’s not just being transparent and having governance, it’s also about actually injecting forcing certain paths not to be followed.

MAY BE USEDpermitted driverpermitted driverpermitted driverMAY NOT BE USEDpostcodeskin colorsexblocked:ethics and lawcredit modelmaths, no ethicsdecisiona decision you can explaintransparent, auditable,governedafter a no, the applicant asks:why?who decided?on what data?was it biased?is it ethical?
Ethics has to be injected: some drivers are closed off by design, and every decision must answer the questions that follow a no.

All in, with reservation

Mike

So you’re painting this really interesting picture where we’ve got the complexity of that mortgage flow, key decisions and gates through that process. And there is some serious work to be done there. Now on the other side of this is the promise of acceleration and efficiency, because there’s a lot of manual ineffective work being done, or just very laborious, time-based work being done. There’s sort of like a dichotomy here that’s appearing. When you talk to banking executives in this moment, where are they at in embracing this? Would you say everyone’s all in? Would you say it’s a fifty fifty? What’s the temperature right now? What are the opportunities and threats here take us into the pulse, if you will, of banking becoming more agentic?

Ray

So I think they are all in, but with reservation. And what do I mean by that? There has been a lot of organizations who did the classic let a thousand flowers bloom and they created lots of small little projects and they spent a lot of tokens and a lot of startups, stand ups and things and the business case actually didn’t stack up. It actually didn’t really do anything of major benefit to the organization. So a lot of those projects are actually being shut down.

Mike

And just to jump in right here, are they casting a shadow, do you think, across this march towards agent?

Ray

They are because the senior executives are looking at this and how much have we spent and how much have we burnt? And what is actually the benefit? Then if you actually really truly are looking for beneficial AI, actually really changing processes and really optimizing either the cost or your go-to-market speed or your profitability, et cetera, et cetera, you have to embed these and have to implement them in a coordinated, governed way, which most of these standalone projects weren’t doing. They were just like kind of desktop. Yeah. Projects. And it’s okay if you want to do a new type of report and come up with a nice way of looking at information.

Mike

A dashboard.

Ray

A dashboard. Brilliant. But if you want to fundamentally change the operational processes or accelerate them or reduce the amount of human need in these things, you have to embed it massively. To do that for an organization, you have to start at the really boring bits, which is you need a data foundation, you need a fabric, you need a complete governance model on top of that, you need one source of the truth, you need these agents all to be coordinated, and they all need to actually operate seamlessly together. And if they’re making critical decisions, they have to be completely and utterly transparent and auditable.

Mike

So this is fantastic. What do you see in this crossroads that banking executives are at? Hey, we did lots of experiments. We threw a lot of stuff up against the wall. Lots of flowers that we tried to.

Ray

Some of them worked.

Mike

Yes. Yeah. Some worked. ROI, a little questionable. What do we get back? Is this worth it? All those kinds of fundamental questions. You’ve got that over here. And on the other side, you’ve got the hard work of building rigorous, deterministic, like bulletproof, agentic solutions. What are the types of executives? That are getting that? What are the types of executives that understand that? Describe those people. How do they think? What are the best practices? Shine a light for us on those that are winning with AI.

Ray

Digital transformation. It’s not a new concept, it’s not a new word. But digital transformation means you have to start with. All the flows of your organization. Think of data as like the lifeblood of your body. If you have veins and arteries which are blocked or you have things which are not connected together, and then you start building something on top of that using beautiful AI models and agentic agents and lots of other things and LLMs, it’s not going to work. It’s going to give you wrong answers. It’s going to actually not function correctly. And the body is going to starve or it’s going to get cut off.

Mike

Ray, what’s a characteristic or what’s something you notice about an executives? A tell, if you will, that they’re not going to build a house of cards, but actually they’re going to put agentic at the heart of the business. What are some of the things you see in execs when you meet them? And you’re like, they get it.

Ray

They’ve gone through it before. This is not the first time. So remember the workflow? You remember how workflow was going to revolutionize all processes and how workflow was going to completely automate all things?

Mike

The workflow step up?

Ray

The workflow step up. It was the same thing. Why didn’t it work? Why didn’t we get these enormous to be fair, on workflow? Lots of things did change and lots of things were beneficial, but it didn’t give. The amount that was claimed that it was going to give because they realized that actually they had more fundamental problems. They hadn’t managed to bring all their data together and actually do it in a way where you could actually automate it. And they get that. And so when you talk to some CIOs, some COOs, some chief data officers, they’ll get it, especially if they have a bit of experience. And they’ll say, I love the idea of implementing agentic agents. I love the idea of implementing [unclear]. I love the idea of this technology will. Actually, automate some of these things. But I also realize if I don’t create the foundation, I’ll have a Taj Mahal in the sand.

Every new channel, none retired

Mike

Okay, let’s dive in onto this. Let’s go back in the last few decades of banking. Do you recall the introduction of a new technology that was vastly underestimated or was just hugely painful? What comes to mind when you go back to those that have done this before? What might be some of the battle scars that they’re wearing?

Ray

I’m going to pick retail banking. And I’m going to pick the revolutionization of retail banking three or four times in my career. So do you remember a thing called telephone banking? Telephone banking was going to get rid of all branches. And it was going to be everybody was just going to use the phone to talk to the bank. And it was amazing. So we created whole new systems and processes to cope with telephone banking. It didn’t. Then along came internet banking. Now internet banking was going to revolutionize the retail bank. It was going to change everything. Everybody was going to do everything on the internet. You didn’t need branches anymore.

Mike

Didn’t need the telephone either.

Ray

Didn’t need the telephone. It didn’t. Okay. Then along came the app. Application, but with the app, Ray, you see, it’s going to revolutionize retail banking. We’ll have this app. Everything will be done through the app, and we don’t need the branches anymore. Now, what’s the common theme? People do use the telephone, they do use the internet, they do use the app. But the branches are still there. What actually happened, they actually increased the cost because we actually increased the amount of channels that we had to then service. And all of a sudden, you couldn’t be a bank without branches, without telephone banking, without internet, without apps. And, oh, hang on a second, how much are we spending now on this? And yes, they’re all have traffic and they all are channels, but none of them actually ever got rid of the previous. So, one of the lessons learned is: as we adopt this beautiful agentic AI or any type of AI, is it really going to replace some of these other things, or is it adding on to the stack? And will it actually end up as yet an even more bigger cost to the organization? And I still have these things in the background that I have to run. And I’ve just actually added to my cost base. And that was not a great thing to happen. So it’s. It sounds weird right now, but it’s what has happened previously in these beautiful technology revolutions where the promise is it’s going to change everything. And it does. But does it get rid of the legacy?

Mike

It as you were talking there, I just couldn’t help but feel somewhat sorry for that poor old CTO or that CIO who is just like fighting a battle on so many fronts from the branch to agentic and everything in between.

Ray

That poor chap still has his mainframes running down the back. You know, the cloud was meant to replace all that. Do you remember? Yes, it hasn’t. They still have all the bricks and mortar data centers as well as the cloud, as well as this. So, a lot of these new technologies somehow seem to solve some parts of a problem. They are very good at certain elements, but they don’t fully replace a lot of this legacy. And sometimes, when you add it all up, the total is actually more at the end of it than actually at the beginning. And I’m hoping that agentic will, or AI in general, will replace some of this legacy and that the actual total TCO, total cost of ownership is actually less at the end than at the start.

cost of running every channelreplace something, or add to the stack?branchbranchesbranchtelephone+ telephone bankingbranchtelephoneinternet+ internet bankingbranchtelephoneinternetapp+ the appbranchtelephoneinternetappagentic AI+ agentic AI
Each wave was meant to retire the one before it. None did, so the real test for agentic AI is whether it takes something off the stack.

Pilot to production: start small

Mike

There is a moment where we can truly figure out the total cost of ownership and determine whether we truly can use that. And that is in this very special moment in a bank where our stressed and overworked CTOs and CIOs.

Ray

God bless them all.

Mike

Oh, the poor things. They’re about to go from pilot to production. Some might say this is a tipping point. It is a moment of truth. There’s perhaps some more explicit words to use these moments as well. How on earth can we help those? Men and women who are trying to move their agentic pilot into production using the lessons of history, how might we help them to do this successfully?

Ray

It’s a great question. You have to start small. You have to pick simple processes and items and workflows that you can automate and that you can do in an isolated case and prove it. And with that, create the framework and the governance around that. Then you start expanding and building on that. You don’t pick. Your core banking system and replacing that as your first step. If you do, God bless you and good luck. But I would rather you didn’t because of the risk. I would rather you pick some of the more tangential items that are slightly outside of the mainstream and start there and then build on them and do so it’s a stepped approach. Secondly, you’ve got to have the foundations. Put in. You’ve got to have the frameworks, you’ve got to have the governance. And that’s why a lot of people in the tech community are so frustrated with the facts. Ray, why does it take so long? For God’s sake, we could do it in a week. You can do the development in a week. That doesn’t mean that I can change a bank in a week. And honestly, guys. You wouldn’t want a bank to change in a week. This is where all your money is. You want it to be safe. You want to make sure it’s tested. You want to make sure it’s not making mistakes. Changing a bank in a week, that’s not a good thing.

Mike

No, I mean, we don’t want one of the zeros being removed from our savings account if you know what I mean.

Ray

Or some cyber hackers getting in, or all these other kinds of algorithms.

Sovereign AI, layer by layer

Mike

Absolutely. Absolutely. So, the winners are those that are building those foundations, starting small, chipping away with a long-term vision. I’m very interested to know where you think there’s a very hot topic within banking is avoiding the token gouging. Yes. Avoiding putting your data with a frontier model.

Ray

Yes.

Mike

And bringing things into the organization. So if you want to put AI at the heart of the bank, what we’re really talking about is this concept of sovereign AI. Now, this seems to go. Some way to the foundation that you’re talking about. What are you seeing on this hot topic in banks of sovereign AI, local open models with their orchestration, the data is not leaving the premise what are you seeing on this topic of sovereign AI? And what’s your sense of where to win with this idea?

Ray

It’s a beautiful question. So let’s look at sovereignty for a moment. And let’s take it on the literal explanation of sovereignty. So let’s take a country, and that country wants to keep its data sovereign and wants to keep its banking sovereign. What does it really mean? It’s the hardware to start with. So where is that data center? Is it in the cloud? Where in the cloud is that? Is it actually a data center in this country? Where is the business continuity plan for that data center? If that data center goes down, who’s responsible for it? Who’s running that data center? Are they financially solvent? If the parent company that’s running that data center goes down, will there be a problem with that data center in this country?

Mike

And you’re literally only bringing up questions around. The location of the hardware. And if you think about the stack, this same sort of questioning will happen right from the bottom, right to the top.

Ray

I’m going to go through really quickly.

Mike

Yeah.

Ray

The next layer, the software. Oh, so now all of a sudden we’re introducing software on top of this hardware. Okay, who owns that software? Where was it built? Is there a kill switch in that software? Was there that software built transparently? Do we understand exactly how that software is being used? Who trained that software? What data sets did they use to train that software? Etc., etc., etc. Can we actually have the source code of that software? Can we actually operate that software without the parent company? Next elements the data. Where is the data residing? Who can access that data? How much can that data be shared? Even within an organization, data is not shared fully. So never mind sharing it externally. And then you get to the pinnacle, the application. And then we go to a thing called LLMs. And LLMs is a beautiful thing. We all know about the big global LLM models. And they are wonderful. They’re brilliant. But most corporate institutions cannot really use them for corporate use. Why? Because you’re sharing your data publicly with a lot of these things. You therefore have to come up with some private LLMs. And then you have to come up with can I actually implement an LLM within my own environment where I’m not sharing anything outside it? And immediately you siphon. That means that LLM is not going to be as functional as a global LLM. I know. That’s the issue. And that’s the drive at the moment is to trying to create those private elements that are as good as possible within an environment that’s completely encapsulated and completely sovereign.

Mike

Well, this is fun because we both know it’s the model and a whole lot more as well: the harness, the governance, et cetera, the ontology. But on this, if we were working on this with one of our harried CTO friends from the banking enterprise, I think one of the cases I would try to make is you can build your own model. Explicitly only on your data, on your patterns. So it doesn’t need to have summer holiday recommendations and restaurant guides. It can just focus on your business.

Ray

Exactly.

Mike

So the cost of training for that model depreciates a lot. It can be purpose-built. You can do so where you do not forego any control. You can run this. In the enterprise, control it in the enterprise to do only the job that you needed to do, enterprise.

Ray

And it’s a bespoke LLM. And it’s bespoke on particular types of use cases. It’s not a general LLM. It’s a very bespoke, narrow LLM.

Mike

And that truly, is that not for a bank? The aha moment? Like, we don’t need general intelligence to do. Cooking guides, travel guides, et cetera, et cetera.

Ray

Very useful, but not really useful in corporate banking culture. You can do it yourself as a human being.

Mike

And is that that’s also really why general intelligence and frontier labs can give us something that’s 80% good. Yes. And they cannot do what that local, open, purely proprietary model sitting inside the bank?

Ray

100%. And think of it as a two-step approach. The large LLM companies can create a model that has learned to walk and talk and read and write. It’s got the basics. Great. It’s got the foundations. Now we’re going to take that, we’re going to implement it in our own environment and train it with just our own data, completely private, for specific use cases. That’s the second step. And that’s what a lot of organizations are now moving towards. The basics, great. Let’s take a model that’s been that’s learned the basics, but now let’s actually take that and put it into our own environment, completely encapsulated, that actually does the finishing training for a specific use case. You’d look at it and say, but really, this model can only be used for this and that. And I go, yeah. It’s been trained to do exactly that. It’s never going to tell you what the weather’s going to be tomorrow.

Mike

No, it’s certainly not going to do that. And when you’ve built your own proprietary model like that, it’s certainly not going to creep out the back door and hack the Australian government and the US government, which we know has happened.

Ray

It’s the number one threat vector at the moment. Right now in cybersecurity is the new AI threat vector, which is quite fascinating. So cyber security specialists around the world are really worried about how many different back channels are coming in through these AI models that are being used in organizations, not just banks, all corporates and governments. And it’s huge, right? And the hackers have figured this out and they’re building things to actually allow them to try and get in the back door.

hardwareWhere is the data center? Who runs it?What happens if they go down?softwareWho built it, and is there a kill switch?Can we run it without the vendor?dataWhere does it reside? Who can access it?How far can it be shared?LLMA private, narrow model, trained only onour own data, inside our own environment
Sovereignty is asked layer by layer, from the hardware up. Only at the top comes the model, and for a bank it has to be private and narrow.

Great promise, hard work and the FlowX.AI opportunity

Mike

And it feels like what we’re doing in this conversation a little bit is we’re oscillating and leaning into two different sides. You might call it the pros and the cons. Yeah, you might say it’s you know, local and frontier models. It’s really interesting how this conversation seems to have a pendulum to it. What I see is enormous promise, but it’s almost like the button with hard work. So, if you want to deploy an agentic workflow, make decisions in the bank. With AI, it comes with great promise.

Ray

Yes.

Mike

But it also comes with, hey, roll up those sleeves and get to work. And by the way, all those other channels are still open.

Ray

Yes.

Mike

Right?

Ray

Yes.

Mike

The phone, the branch, et cetera, et cetera. They’re all still open. Now you’ve got a new one with great promise as well.

Ray

But you’ve got to build those foundations. You’ve got to create those governance models. You’ve got to create all these elements before you can put the icing on the cake. And a lot of technologists are frustrated with that. They’re going, but we can implement them so quickly. Why can’t we just do it? You could. But you would break the organization. There would be many of these things just running in isolation and there’ll be no coordination across it. There’d be no governance. There’d be no mapped out processes. That’s hard work. That is hard work. You can’t just expect AI to just do it. And you wouldn’t want it because it’d be a black box. You’ve got to actually define exactly what you’re trying to do.

Mike

So for our friends in banking, I think we come with two parts of news. We have great promise, but with the hard work.

Ray

Yes.

Mike

And I really like the way in which you’ve said. Hey, we’ve seen with retail those four waves that we talked about. We’ve been here before, we know it starts with foundations starting small, experimenting, and learning. What’s quite interesting is: you know, a lot of the news is taken up with the general intelligence frontier labs, but this space. Enterprise agentic space is somewhat an underestimated field. And I think there is enormous potential to accelerate the enterprise to improve a number of these things that we’ve talked about, how workflows move through the organization, how data moves through the organization, how software can maybe improve itself over time. Yes, so much promise. But specifically, now I want you to cast your eye towards FlowX.AI and this opportunity. What do you see in the possibilities? You’ve seen all these. Patterns, ways, and trends before, Ray. I’m very curious in your mind: what is the size of the pie here? What is the opportunity for FlowX.AI?

Ray

It’s enormous. The thing that attracted me to FlowX.AI, why am I here today? It’s because they didn’t just build a bunch of agentic agents. A lot of people can do that. What they did is they created a whole framework around it and they actually created a way that it could be implemented and adopted in a transparent, repeatable way. In a way that large corporates, not just banks, any large governed organization could implement and use and adopt pretty quickly. That’s what I like about FlowX.AI. The talent is enormous because what you’re fundamentally doing is you’re re-engineering these organizations using these things, and the business cases are huge. Let’s take consultancy companies. God bless all the consultancies out there. For 30 years, what they did is they would go into a COO. I was a COO. They would go into a COO and say, hey, if you outsource your operations to us, we’ll save you 30% of the cost. How did they do it? Did they A reengineer the whole process and implement technology? Or did they B actually move most of that workforce to somewhere cheaper in the world? I’m afraid the answer was B. And actually what we’re now doing is actually saying A might actually work, and we might actually be able to automate through technology some of these processes. That opportunity alone is enormous. And then there’s all the other elements of decisioning and things that you can do. I was just talking about business process operations, but there’s many other things and features within an organization. So I think it’s an incredible opportunity for FlowX.AI. And they’re doing it in a way. Where an organization can adopt the product and feel comfortable and safe. And they’ve got proven examples of where they’re being used in highly regulated organizations.

Mike

Could I go as far as saying, Ray, and you can push back on here. I think what you’re saying is we were weighing up there’s possibilities, but there’s hard work to get there. Could we make the argument that having the platform like FlowX.AI, you’ve actually reduced the amount? Not completely.

Ray

Yes.

Mike

Taken a bit.

Ray

Not completely. It’s still hard work.

Mike

Right. There still will be something, but a lot less.

Ray

Yes. You’ve reduced the amount of hard work needed because they’ve gone through the thought process and they’ve come out with actually: here are the steps that you need to take, and this is how we can do it with you. And this is actually the process. And they guide you through that. That is what a lot of these organizations need help on. And if you can accelerate that adoption, of course. Every CIO, CTO, COO, CDO, whatever CXO you want to mention is going to be very happy because all of a sudden they’ll see real beneficial, tangible benefits on some of these AI projects. I’ll go back to what I said at the start. A lot of AI projects in large corporates was a thousand flowers bloom. And a lot of those flowers, I’m afraid, didn’t really have a proper business case. If you can all of a sudden actually implement these things at scale and actually really have a business case with a return where the TCO actually comes down or your sales revenue goes up, et cetera, et cetera. You’re on a winner.

Mike

What a great way to wrap up our conversation, Ray. Thank you so much for joining us today.

Ray

Thank you very much. A pleasure talking to you.

Mike

And thank you to you, our listeners and our viewers. What a wonderful journey we’ve been on today. We found out there is so much possibility on hand for banks, for mission-critical applications in those enterprises, but there is hard work. And we’re not going to shy away from that, but you can get a nudge, a little bit of an advantage with FlowX.AI. So, if you’d like to learn more about FlowX.AI, check out the description below. Give us a like while you’re there. And we look forward to having you next time as we find out where we are now and what’s next in banking and AI.

Next

Put AI at the heart of your operations.