PegaWorld 2025: The Future of Banking in an Agentic World

PegaWorld 2025: The Future of Banking in an Agentic World

Welcome to the best breakout you'll go to at all PegaWorld. Oh no, maybe not the best breakouts are the ones that have clients talking, but what I'm going to do in the time we've got together now is give you an insight into where I think we're going to be going with the genetic AI now, after over 30 years in banking. I know it doesn't look like I'm that old, you know, maybe 20 years in banking, but after 30 years plus in banking, I really do think we're at a very important stage in transformation and change.

I was trying to think what to compare it to. Some people have said the best things banks ever invented was the ATM. I think that's rubbish. I mean, just getting cash out of a hole in the wall may be innovative at the time, but not that big a deal. Um, what about the apps? These days, we can do pretty much everything on an app. That's pretty cool. But I think instant payments. That's that's that's pretty good. Even even America and Canada and instant payments now. But I think really we're at a bigger a bigger stage, a bigger shift in the industry with the technology that's available. And so I really thought it might be a different industry. We could look at a different lens. We could look at change and what's happened and make an analogy with where we're going to.

50 years ago, two things happened. My brother was born. You're not really interested in that. But more interestingly, is Industrial Light and Magic was founded about four hours drive from here. And for those of you who aren't familiar, they started with models, a lot of model work and a lot of the model work. You can see the picture with the guy with the I think it's called an At-At walker, showing myself as a real geek there. Um, they do little micro movements. It was like stop motion technology. So very painstaking. A lot of effort went into it.

And then on the right hand side they did this effect called matte painting matte effects. So for those of you familiar with the Star Wars movie in the middle there you can see the bit where Obi Wan Kenobi was sneaking around the outside. But there's a painting. It was actually a painting, a really very skilled painting being done to do the effect around the outside, which not a lot of people realized. So they really sort of transformed the industry. And in our banking, business or world, I might think this was the stage where you had basic automations like simulated logins for robotic process automation. So it was getting there, but it wasn't really what I call super advanced.

The next big leap that was made with movies and television was around going full digital. The top left there for any Star Trek fans in the audience, was the first ever fully digital created sequence at the Wrath of Khan Star Trek movie. It's a sequence where the planet's transformed and it was fully done digitally. I imagine most of the room. Well, show of hands. Who in the room recognizes the bottom left? The metal figure from Terminator two? Probably. I'd say nearly everyone. Yep. That was very famous.

But perhaps you're not aware of the latest iteration and generation of what Industrial Light and Magic has done to the industry. They've done something that George Lucas always wanted to do, and they've got the perfect blend of technology and people in what they call stagecraft. So you can see a rendering on the top right, which is when it's not being used. It's a massive wraparound screen and a ceiling. And you can see a picture there from The Mandalorian TV show where they've got that screen very, very high definition set up with the backdrop and the model, although I don't think you can call it a model, it's like a building of that ship.

So what does this do? This, this this gets you to an incredibly authentic performance, an incredibly frictionless performance from the actors and actresses, because instead of being in front of a blue screen or green screen where they're imagining what's going on. They are literally in front of what's happening. And on the end effect is stunning. I couldn't get away with doing a presentation and making this analogy without showing Baby Yoda or Grogu. My daughter's favorite. She's 12. We watched all the movies together, but it really brings out when you when you listen to the actors talking about this environment, it's like they're on set for real.

And as we're in Vegas, you know, I have to mention the sphere. If you haven't been to it yet, has the biggest, largest high definition screen in the world. And actually Industrial Light and Magic have done the effects there for the the concerts by U2 and Dead and Company. So incredible transition, which I think we can think about in terms of the banking path to an agentic world where we've had very manual and really sort of heavily peopled managed processes. We've moved towards some automation and AI.

But you remember Rob Walker this morning saying, getting the best of both worlds. He was talking about getting the best of both worlds in terms of AI, statistical and generative. Right brain. Left brain. I think we're at a pivot point where the real, real change will be for the banks and other industries that can get the perfect blend of the technology with people. And I say that because if you think about any role you've done in a bank, any role, when I used to manage lending operations and collections, any role where it's medium to high complexity or risk, there's an element of there has to be a people escalation, there has to be a people review. And so even though we're going to get to a point of replacing people's work with Agentic AI, that's still going to be in that framework of what a person would do, what a human would do.

So what does this look like? And I've done nothing with ChatGPT or like many of the people have done with this presentation, but I did think I'd have a bit of fun with a free video editor online called Nvidia in Video. This took four minutes to generate.

I'm sure you've heard the buzz. AI is taking over banking, but is it really time to wave goodbye to human touch? Picture this you're sipping your morning coffee while an AI assistant juggles your finances. Sounds dreamy, right? But hold on. What happens when you need more than a balance? Check? Enter the perfect balance. Banks are using AI to handle routine tasks. Think instant transactions and fraud detection, freeing up humans for the personal touch. But don't worry, Arnold, the Terminator isn't handling your mortgage just yet. The trick is knowing where to draw the line, routine and repetitive. Leave it to the bots. Complex and emotional. That's human territory. It's all about creating a seamless off ramp from digital efficiency to human empathy. So next time you call your bank, rest assured you'll get the best of both worlds. AI precision and human warmth. Now that's what I call smart Agentic banking with Pega.

Okay, so apart from the cheesy bit where I asked it to say Pega at the end, that was took four minutes to generate. It was six sentences. I gave it and it took some liberties of introducing some different things into it. I like the bit where they talked about AI precision and human like interaction, human warmth. And I think, you know, for me, that balance is something I want you to bear in mind as we go through some use case examples of where Agentic could take us.

So you've heard a little bit about how Pega is thinking about this in terms of the power of agents and the predictability of workflows. I think to call something agentic, it has to do a few things. It can't be a simple basic automation like an RPA or something. It has to be something that's looking at making a decision. It has to be something that's triggering an activity, triggering an action in one or multiple cases. And I do think that banks and other regulated industries are going to want some predictability, some governance, some controls around this.

So let's have a look at some aspects of where this could go. So firstly, if you think about the whole area of sort of operations and customer service, I think there's an incredible opportunity here to improve the automations and accuracy levels around things that are, say, important to a customer, but maybe not necessarily as high value to the banking side, like a fee payment inquiry or like a loan payment change, things that are not so risky, but you move to an area where you've got some of the broader workflows that are medium to high complexity, like in the areas of lending where you're actually initiating a loan or know your customer documentation. I think you get to a different level there where you want some structured organization around it.

You think of these roles in the banks doing KYC or doing lending. There are risk appetites, there are policies, there are procedures. There's no bank going to get rid of that. There's no bank not going to have to do like I had to go in Australia once a month to the regulator and explain, are your manual credit assessments and your automated credit assessments performing in the right way? Are they giving a responsible lending outcome for the client. Can you show that? Can you prove that? And I don't believe a regulator will will buy an AI agent reviewing an AI agent. At some point, there's a cut off threshold where you look at that agent driven work. Having a human element.

And the last one I think is a big one around coaching help, advice. I mean, myself and my team are using Iris that you heard about this morning and other tools to help improve and speed up our work. But I think in the banking sense, imagine things like researching your client, looking into personas and getting information about those personas. And I'll show you an example of this in a minute. But also coaching the customer, coaching the client, giving them advice. I think that area is good. But again, you think about some of the regulated elements of banking. There has to be a point where they'll decide, Will I want to let an automated agent do this?

So let's just think firstly about the sort of broad areas that we're thinking about in terms of predictive, predictable AI and where this will impact banking. And then we'll look at some use cases. So the first one is around design. You've heard a lot about Blueprint. I won't belabor that, but it's definitely an area that can help conversation agents. Again, that's been showcased in some of the keynotes already this morning, and I think that genuinely can move a level ahead.

However, I'll give you one example from two weeks ago where this has still got a long way to go. I was checking in, doing a self check in to an apartment in London just for one night. It was the cheapest thing I could find through Booking.com and there was instructions how to get in and they said the first door put this key code in. The second door, different key code. I got there, there's this big metal gate and I could see the door where I think I had to do the key code, but I couldn't get to it. I couldn't get through the metal gate. So obviously I go to the app and the website and I try and look for the phone number. Anyone want to guess how long it took me to find that phone number? I'm not kidding. 5 to 10 minutes searching for a phone number before I could get to someone to help me. So whilst the conversation agent conversation agents are phenomenally better using Llms. There has to be a really seamless, frictionless way to get to a person when you're doing something as simple as trying to check in.

The the next area is around, um, autonomous agents. This would be where you're letting an agent do the work that a person was doing, and there is no need for it to be have oversight. Now, there's an exception here because, um, if any of you want to hear a really good use case story after this, the Santander presentation on Re-engineering Legal Operations with GenAI, you'll see that they've got to a position where they've reduced the workload, but they've still got a certain amount of people doing QA and checking. And even with a fully autonomous agent, you're going to want an ability to check. You're going to want a system that can track it and show it. And then some people doing some analytics and looking at it.

Then you've got probably the more common case, I think, which will be ad hoc automation agents. So this will. They'll sit alongside people. They'll work seamlessly in an area where you need to escalate and you need to pass work around. And I think this will be the more common scenario we'll see in banking, especially in high to medium risk use cases, knowledge agents certainly will have a big role to play, which we've talked about already. And coaching agents similarly along the same lines.

So let's have a dig into two areas. Um, in banking, one very dear to my heart where I spent a lot of time lending and servicing as well. So where are we going to be maximizing the AI and agentic AI impact in banking for for lending? The first one, I think, is going to be around a critical area of revenue protection and growth. So in that whole marketing and initial acquisition phase, critical area for people to look at getting the right offer to the right person. A lot of countries do have responsible lending legislation. So we do have with our Customer Decision Hub an incredible engine there that can make sure the right things go to the right person at the right time, and it is appropriate. It's been checked.

The next really important area is when you get to the actual application stage, the loans come in, you're going through assessment and then fulfillment. And this is where if you're if your service levels aren't consistent, if your service levels aren't in the top quartile, you risk having revenue leakage. And data quality from third parties like brokers can become a real issue. I remember when I was running our lending operations in Australia, there we had up to 25% data quality input problems. Now data quality doesn't need a team anymore. I used to have 12 people that would check all of the data quality before it went to the assessors. It doesn't need that. That can now be done fully automated and fully with like an agent doing a checking role of all of that.

The next area is around the critical area of fulfillment, getting the actual product and the credit facility into the client's hands, as well as then servicing them afterwards, looking after them afterwards. This has always been, for what, the last ten, 15 years? Digital first banks in the Nordics and Southeast Asia have been leading the way with digital first, human touch second. Getting that right balance between what's automated and when a person gets involved. And I think agentic, I can certainly get involved here, but there'll be a point where you get to a point where, you know, a client needs advice or there's multiple products that could be applicable to them. And certainly in the commercial world where you want to actually talk through different credit facilities, different cash flow facilities. But there's an area here where it can be applied to pre delinquency and really improve pre delinquency monitoring before clients get into collections cases. Perfect use case for AI. Statistical AI machine learning. And again then pushing to a servicing agent who could help.

So let's have a look at a couple of areas within lending and digging a bit deeper. So lending origination and servicing. You've got a screenshot here of a build out in our blueprint tool. and the ease with which you can add generative AI into the process to actually check and validate information. This kind of use case would apply to KYC onboarding just as much as lending areas of servicing. Um, one of my colleagues was demoing to a large bank in America recently a loan reconciliation process where you have an extra loan facility coming in and you're comparing what's on the application to what's in the system they've got, highlighting any differences, looking at the credit risk policies, and then making a recommendation to the relationship manager who could then go and talk to the client or send an automated offer if you wanted to, to the client.

So I think the ability to integrate and use generative AI in this space and effectively get an agentic process where you could kick off an automation would be really high and really useful, and it will be very dominant in the retail space where I would expect it to get to very high levels, more so than commercial. The reason for that is if you think about Today. Current practice. And you picked a retail lending product like credit cards. The straight through process rate for an application would be 80 to 95%, with 1,015% being checked and manually credit assessed. You imagine that being able to take that 8,090% straight through process and then add in the agentic capabilities, do a bit more. You'll probably be left with maybe 5% that might be actually looked at by someone. So I think there's a big, big opportunity here in the loan origination space to improve it and the end outcome for the bank. The business case for it will be that consistent service level, that reduction in revenue leakage and the protection, if you like, of the revenue stream.

Servicing a huge area where there's a range of options in servicing. So we'll just look at a few of them. First and foremost you're looking to retain the customer. You're looking to nurture them look after them. And again I think the capability to give immediate notifications and help here is huge. We've got clients globally that are very used to doing this now. So there's an economic crisis. You want to offer some economic assistance. There's a flood. You want to offer some relief linked to the flood, or there's another event like a fire and you want to offer some help. So I think that ability to, in the moment, get offers that can nurture, protect or suggest, for example, a review of your credit facility, suggest a review of your investment profile that can all be automated, that can all be done based on triggered events, that does not need a person getting involved, other than still maybe reviewing some of the information that would go out.

I don't believe that for certain important messages like this, you'll leave it totally to a genetic AI agent that will create the message. Send it out. I still think you want someone looking at that to sense check it and even check that the the location, the zip codes, the postcodes it's going to are the right ones. But a really good area to get fully a lot more automation.

And then when you look through the spectrum of activities you've got going across, you know, from account management through to complaints. Again, think about the context. Take complaints, for example, a complaint that's going to go to the ombudsman in your country, the regulator. You're going to let some magentic I look at it, yes. You're going to let it potentially put together a resolution. Yes. But are you going to release that to the Ombudsman with the result? I don't think so. And I don't think so for certain high value cases either. There will be a point where you want that seamless off ramp that the video said, that frictionless off ramp to a person to look at it.

And I think this applies to multiple other areas where you're looking at the transactions, the loan, the credit facilities, investment, wealth. You think about how in most countries there's a there's a regulation about who's allowed to give advice. So I think a certain level of advice can be automated and sent out to a customer. And we had this a few years ago with the wealth manager, where they sent out some automated, tailored advice. But the follow up was triggered from a relationship manager, a private banker for a phone call or for a meeting to try and get a meeting with the person. So it wasn't fully a full automated process, and I don't believe it would be unless the customer opted in.

Another we saw we saw this demo. You saw this demo this morning and the conversational agents. Now, the things I'll tell you about this is that when you design it in Blueprint and get the conversational agent to talk to Blueprint guess what? It can look at everything that's in that Blueprint it can look at everything that's in that design. But if you haven't got all the policies and procedures loaded, if you haven't got all of the parameters of what's allowed, what's possible with. In this example, I did a situation where a simple one I've lost my credit card. I need a new one, which these days would tend to be a digital card. But let's say I was asking like was being demoed this morning for additional credit facility. There are some situations that could be auto approved. Another one where it's like, I want 10,000. I'm only going to give you seven. In that situation, you have to make a decision of do you want to contact the customer to talk to them and lessen the blow and say, we can get you to 10,000, Steve, but in a couple of months for this reason or whatever, or do you want to just send it to them and risk losing their business? They go elsewhere.

So I think there's a revenue angle to this where the conversation agent can get so far. And then you've got two scenarios. Either the client themselves want to talk to someone, or the bank chooses to do an intercept because there's a high value situation where it's not just say that credit card, like this example, but Steve's also got a large investment portfolio or a large mortgage like I've got. So I think the ability to use natural language conversation huge here, the ability to use things like our Pega GenAI Coach great, but it's that decision making around how far are you going to reengineer things?

So a simple use case that should be easily implemented, and it is in some banks today in a way A is for automatically blocking and then reissuing a replacement card. A bank in the Nordic Swedbank five years ago had a situation where you could use your voice. You phone up, they identify Steve. It knows it's you. You ask it for. It was Nina. It was their tool. You ask it for a replacement card. It confirms you want it sent to your home address you've got on file. Yes. That's it. You could also say, I want to talk to someone and Nina would pass you on to a person. But that was all using voice recognition technology and then to a person. And that was five years ago being implemented.

So I think the ability now is the difference is the agents could more easily than that, than that voice. I go and look at policies, procedures, credit risks. You know, what's been going on with your transactions. Make a recommendation based on that. Take an action or take multiple actions, which could be I'll reissue the digital card, but I won't up your limit. I'll pass you to a person to talk about that. So I think a really a really good use case. An easy use case if you like, which should be quickly adopted. Let's turn and turn a bit and talk about commercial relationship management and commercial banking. So on the screen it's going to be a bit small. You can see it when we issue the slides onto Pega.com on the screen. What I've done is I've shown an example of where I've gone and asked our internal sales automation tool to help me with the sales process and negotiation process, how I would go through with customer service, and to then go and look at a at a proposal, sorry, a sales opportunity, evaluate that sales opportunity and say, what should I do next?

So you think about this in the banking world, where in the commercial banking sector they've been squeezes on the relationship management side trying to do it more automated shrink. The teams SME market is predominantly not relationship managed anymore. So if you have time pressed commercial bankers, what better than to give them easier summaries insights, Suggestions of what to do. This sort of, for me, replaces a little bit where you might have had a manager or coach. You still want to have a manager to go to, but this to me replaces it and equally applies to, say, commercial lending, where we used to have teams that were specialized in different industries. You still need a bit of that, but imagine the amount of information you can give them on those industries now straight away.

Another important area payment exceptions to a bank and a very important area to banks because both on the retail side and commercial are being squeezed out by new players. This traditionally has been an area where it's been up to 14% of a bank's profit from this area, but under pressure, and for the commercial payment exceptions that are done still predominantly through Swift globally, Pega are actually processes about 90% of the global volumes of Swift payment exceptions. But you see, there's something that some of my colleagues at the conference a couple of years ago, quite nicely called gobbledygook, written there on the on the left hand side of the screen pop out.

So this bit here, how easy now it is to translate that into simple English and not just English in terms of decoding some of the code, but actually very easily understood. So in the past, you'd have someone in a middle back office function who was a specialist in payment exceptions. Now you have a simple summary. You could have anyone do that. You could have someone cross skilled from retail payments into commercial payments. A few years ago. No chance. Because they're looking at all that Swift stuff, interpreting it, understanding it, and to take it to the next level of having an agent suggestion that could be fully authentic, could be fully automated. What about a suggestion of what to do next? Because sometimes in a complicated, complicated case with multiple counterparty banks investigations to do, it can be sort of a bit overwhelming of what should I do next? But a suggestion based on your policies, based on your procedures, to say what you do next. Brilliant.

So I think for us as a company, and certainly when I take an industry lens on this, I do believe that in regulated industries like banking, we need to think of any workflow being trusted, the AI being predictable, the frictionless as possible scenario between a person and the automation, either assisting them or moving between the two really only happens if you can get some good orchestration, some good governance, some good auditability. And I think I think back to when I had situations with, you know, high risk, um, delinquent default customers asking for certain facilities. It had to be reviewed by someone and then sometimes signed off by a committee if it was a high value at the bank in payments, if it was over $200,000, a human had to look at it when it was a reconciliation.

So that will still happen, but you will have an agent helping you. And what this means as an experience is you get access to all of your workflows, all of your intelligence, all of your procedures, and you importantly try and get the best out of both worlds. So taking us back to that example I gave you, The Mandalorian. You know, instead of someone not knowing what's going on behind the scenes, standing in front of a blue green screen, they've got the best technology helping them. They've got transparency of what's happening. It's really clear what's going on. It's not opaque. They don't have to go and ask a colleague, can you explain that Swift language? They don't have to go and ask a senior credit assessor. Is this okay from a credit risk point of view? But but at some point I believe they will need to. And I believe you will still have regulators who will be asking banks to come and explain how it's working, how it's not working. Prove it. And I don't believe in some of the talk around AI. AI agents checking AI agents, checking AI agents. Not when you get to an area where you have a simple credit risk appetite and a simple rule and policy.

Thank you very much.