Podcast: Rabobank, Santander Brazil see returns from Pega GenAI

Whitney McDonald 12:01:05
Whitney, hello and welcome to The Buzz a fin AI news podcast. My name is Whitney McDonald, and I’m the editor of fin AI news. Fin AI news has rebranded from bank automation news, marking the next step in our mission to lead the conversation on innovation in financial services technology. Joining me today, December 16, 2025 is Steve Morgan, global banking industry lead at Pegasystems. Steve is here to discuss how banks are using generative AI, including best uses of the technology, and also where Gen AI might not be the best fit. Thanks for joining us, Steve.Steve Morgan 12:01:36
Well, thanks Whitney, thanks for having me on. My name is Steve Morgan. I’m pega systems, global banking industry lead. And Pegasystems is a AI and workflow automation company that’s been around for 40 years. Great.

Whitney McDonald 12:01:49
Well, we’re going to talk here about the rush of getting into generative AI. It’s not necessarily a surprise that folks are interested in Gen AI and financial services, but they obviously need to, they need to weigh, you know, the benefits and the risks, speed to market and compliance, and sometimes that isn’t the case. Maybe you can kind of talk us through what you’ve seen in the space in terms of balancing that, that want for competition, but also the need to pay attention to compliance. Yeah, it’s

Steve Morgan 12:02:22
been, it’s been a really interesting journey the last couple of years as generative AI has become mainstream, and yeah, for sure, I can show you, share with you, some client examples and what we’re seeing in the market. I mean, what we did straight away was think about how, how and where is the best place to build into our product. And when we thought about that, we thought about in the context of clients. I work in the banking sector and in the most common focus for improvement and application of AI in general, AI and automation is sort of in broadly, sort of three big categories. One’s around customer engagement, one’s around customer services, and then operations more broadly. And in the last two years, we’ve seen clients look to understand the possibilities, the art of the possible experiment. Experiment internally, sometimes first and then do something external, client facing. And I think what’s become more important, important interesting debate is, let’s think about the business outcome first. Let’s think about the right kind of AI for the right use case, and where we need AI, where we where we don’t need AI, where we just need something rule driven or process driven to achieve the automation and the customer outcome.

Whitney McDonald 12:03:34
Yeah, I think that’s really important, identifying that need first, rather than working backwards and saying, You know what we want, AI, we want to be able to, you know, attach AI to the to the name or the brand, and then kind of determining where to put it later. Maybe we can talk here through what the risks of doing that backwards approach is what you might encounter.

Steve Morgan 12:03:56
So yeah, I think with any big program of change, or even small program of change, you need to think about, yes, as we said, the outcome first, and then how you’re going to approach the project and the change management. And if you do that, and you look end to end, to end of the whole process, you’ve got more chance of success. What we have found is in some organizations, there’s a real desire to experiment, test things out, try things, and if that’s not done in the context of the outcomes and the change management, project management, you can get issues, and it’s widely known, and in fact, there’s been almost like new words created about hallucinations, conflict, conflagrations, which is basically mistakes that generative AI can make, which is different, to say, statistical AI or machine learning, which is very disciplined, can learn and react and create predictions. So maybe forgive you, an example might bring it to life with a client, how they’ve approached it, which I think is very sensibly so rabobanks, a decent sized European client of ours, and they looked at a very important use case internally, which is helping customers manage financial, economic crime and processes around that. Clearly, it’s one of those like moments of truth for for a customer, how well the bank deals with that? So they wanted to see, firstly, how could they apply AI and automation to improve the performance their staff who are dealing with the customer, who are solving the issue, or getting to investigating it, getting to the bottom of it, and they used a combination of our technologies and others to to really improve the knowledge, the guidance, the coaching to their teams. They still had a person doing doing the doing the work in the large percentage of cases. But even if a person was involved, they had the ability to trigger automations. The generative AI let them do conversational like chatting with the policies, procedures, all the internal docs, and then they could trigger a workflow automation into our systems that could get the work done automatically. They did that internally, and they actually talked to our conference this June, just gone about going from millions of dollars of losses to 1000s of dollars of losses, which is brilliant, brilliant outcome for the customer and for the back. Bank, they then decided to take it to the next level and look at external facing, customer facing examples where the technology is fully exposed, and they, like many organizations, had a number of different chat bots or intelligent virtual assistants. And one good use case is plugging generative AI so it can become more conversational, more access better information so that you can get a better interaction with the customer, and also then a better, more clear action to complete for the customer. And they’ve been doing that with a couple of their their agents and and I think that’s an approach we’re seeing most most banks take. Is look at some internal use cases, you know, have a look at how what the outcome can be. And then look at some external customer facing use cases off the off the back of in some internal success.

Whitney McDonald 12:06:52
Yeah, that’s definitely something that we have been seeing and following as well. Using those internal use cases first kind of being able to practice and see how it works within, within the company, before deploying anything on the Gen AI side externally. Do you think coming into 2026 we’re going to see more of those forward facing, client facing, Gen AI applications?

Steve Morgan 12:07:16
Yeah, definitely. We’re already seeing a few. I mean, two weeks ago, I was at one of the large retail banks in the UK. They’ve got a tool which does, does take some if you’ve got any extra deposits in your account, it will sweep it and put them in a like higher interest saving or invest them, if you want to. And they were talking about that tool and saying it’s quite basic. It’s quite basic automation at the moment, but they want the ability to trigger like a next best action for the customer, and even trigger potentially either the customer automatically decides, you know what, there’s enough there. I want to invest it in the market, or put it in a term deposit or something else, or I’d actually like to talk to someone and get some wealth advice, some investment advice, and advice on my pension. So it’s taking a simple say, automation and making a bit more advanced. And then we had a client this year, actually, Santander Brazil, and they did. They talked as well at our conference in June, who took a really interesting case, a very important area as well for a bank of legal operations. And they wanted to not only improve the internal operations of it, but for the customer. You can imagine, with legal cases, you know, banks have to provide information to the client, to the judge, to the lawyers, plus they have to use a bunch of lawyers sometimes review documents, and no one likes paying money for lawyers. So they looked at, how could they improve that process, both internally and the client facing aspect. And when they introduced generative AI to it, they took what I think is a very sensible approach around going, we’ll point it just internally, at documents we know are the proper procedures, the proper policies, the proper escalation paths, and we’ll see how good it can be. When they first did it, it was 67% accurate, which is not good enough. Their best staff member doing it was 95% accurate. So they trained the model with experts in legal and legal operations area. They trained over a period of three months, and they got it months, and they got it to 98% accurate, so better than the best person doing that role. And some of that role was taught boring stuff, you know, like 200 page documents, extract the information, analyzing it. So they enabled some boring work to be made redundant. They redeployed people into some more interesting stuff, which is like, you know, I have to talk to you, Whitney, the lawyer, or I have to talk to the customer and give them certain documents and stuff like that. And the best result of all wasn’t removing the redundant work. Wasn’t removing the the steps that were unnecessary or boring. The best result was for the client that the service levels went to like 96 to 100% and the service levels used to be sort of 50, 60% you know, like, so I mean that great outcome for the customer and also for the bank, because they they’re freeing up staff to do much more interesting work.

Whitney McDonald 12:09:55
Yeah, that’s a great example of how it’s, you know, improving internal operations and what you you know, have to do manually or not, but also improving that a customer experience. It kind of shows both sides of shows both sides of the coin. So thanks for that example. Now I want to talk through this idea of prompt and pray that you’ve mentioned before. Obviously, with generative AI, you need to have the right prompt you talk through the training and why that’s important, maybe kind of tell us what prompt and pray means to you and how financial institutions can avoid that, you know, I hope this works.

Steve Morgan 12:10:31
Mentality, yeah, we have a I adhere to this personally, but the company, we have a firm view on this, which is grounded, wouldn’t surprise you in our workflow automation or process policies, background, any important piece of work at a bank has policies, procedures, escalation, paths, you know, audit, risk oversight, right? And then, in many cases, regulatory oversight as well. Just because we’ve got new AI and automation tools, none of that stuff goes away. None of that stuff goes away. So if you want to replace a person with an AI agent, they still have to follow so. And policies, procedures, escalation paths, oversight, checking, you know, QA, so you know the prompt piece, prompt and pray piece is referring to people thinking they could just use a generative ai, ai agent, not grounded in policies, procedures, escalation paths, not grounded in a repeatable process with clear stages and steps, and, you know, almost allowing it to have some element of creativity. Well, are you really going to want creativity when someone’s making a decision on a small business loan or a home loan with with the bank already has a set risk appetite. They have a set risk profile of customers. They want to lend to, amounts they want to lend, exposures they want so there’s an, there’s a, there’s a chance of doing a misusing generative AI, which can be good for creativity, good for summarization, good for pulling information out. But unless you point it and tell it to have no creativity and just look at a certain set of documents, you can you can have mistakes coming up. You could have errors, or they call them hallucinations. I hate that word. It’s errors. It’s mistakes. And you there’s no tolerance for mistakes or limited tolerance in today’s processes, even if they don’t have some AI and automation. So Why would, why would you treat those procedures any differently by applying AI and automation?

Whitney McDonald 12:12:23
Jeremy, yeah, yeah. I think that’s great. And I think that it’s it’s a great reminder that you need to have those policies in place. You need to have, you know, some sort of guardrail that’s still going to allow you to be compliant, still operate how a bank operates.

Steve Morgan 12:12:39
Just, yeah, yeah. The other thing I thought, You mean, it’s important to think about, is this is auditability, Governability, transparency. I mean, we’ve, we’ve had built into our AI for at least seven or eight years now what’s called a transparency switch on our decisioning AI, which is for next best actions, next best decisions. But the most important thing that we’ve done, I think, is look at being being able to be transparent and clear on what’s the AI doing. So, whether you’re using predictive machine learning AI, whether using generative AI, whether you’ve got an AI agent, it needs to be explainable to the level, like a bank has to do today when they go to the when they have to go and see their regulator once a month, with the chief risk officer and like head of credit assessment or lending, and say, how are your algorithms working? How are your decisions working? Are you seeing any uptick in collections when you’ve changed your automated lending procedures? Have your manual procedures stayed in line? How are you checking that your people are doing the right thing? Same thing applies to an AI AI agent or AI or automation. How do you know it’s operating the same way? How do you know what’s happened when you’ve made a change? How have you made sure it’s gone consistently across all channels. How has it gone consistently across your staff? What’s the customer experience? So I think there’s a huge element there which people underestimate, that any regulated industry, but especially banks, have to, have to and should comply with so that they can make sure that the outcome is the right one for the customer ultimately,

Whitney McDonald 12:14:05
yeah, I’m glad you brought up auditability, because you need to be able to follow that path backwards of how certain decisions were made, or, you know, what, what was presented to a client. You have to be able to, you know, cross those t’s and dot those, i’s along the way. That’s 100% necessary. Now, you mentioned a couple of examples with some of your clients. Maybe we could go a little bit further there and talk through some use cases, both on what Gen AI should be used for, what are some good applications, and then maybe what it’s not ready for, what it should not be used for? Where’s the risk not worth it?

Steve Morgan 12:14:38
Yeah, sure, it’s a really good question. So and we like to stay in touch with the market through both our clients, but also our partners that are really important that we work with. And in fact, it was just three weeks ago I had like, the heads of banking from our key partners, you know, like Accenture, ey, cognizant, Capgemini, et cetera, get together, and I actually asked them this question. I shared with them a list of, like, sort of our main pipeline areas and areas of focus with banks. And I said, which areas are you guys and girls all seeing as being the key areas for the application of agentic AI and Gen AI. And there was unanimous agreements that there was, you know, a couple of main areas where it’s really applicable and is being looked at and will be used more next year. And I think the first one is definitely customer service and operations teams, which most operations teams now have a an element that touches the customer, an element that doesn’t touch the customer, non customer facing. They have a mix generally. So those areas of customer service, which people tend to think of as contact centers and area and branches, and then operations, which tends to be everything that is everywhere else, but still customer facing and non customer facing, those are some of the best use case areas for the application of Gen ai, ai and automation more generally. And part of the reason for that Whitney is because they are large cost pools for the banks. But the real other important thing is there the service levels sort of give you a very good goal. Of how happy the customer is, whether they’re going to be thinking about renewing whatever they’ve got product wise with you, like a home loan or extending and using other products. So they’re a good influencer to the revenue stream as well. The other area I’d highlight would be just broadly, the one of customer engagement, where, wherever you’re interacting with a customer, whether it’s digital and digital only, or whether it’s with a combination digital and getting, you know, really seamlessly or frictionlessly to to a person to help. I think that, again, is a critical area where you can use generative AI with no create with limited creativity. So you could almost set the creativity to zero, but get it to help with procedures, policies, options, and then at a certain point, either the customer might request, you know what, I want to talk to, someone live, or the bank may go, You know what? This is a point where we should intervene to give them some advice, discuss some options. Yeah, so I think there’s, broadly speaking, those, those the main areas, the areas where you don’t want to use, we don’t use use any creativity linked to generative AI, but you could still use elements of generative AI for like summarization type things could be areas like payments, payment exceptions, payment disputes, payment fraud, where there can be quite complex cases, quite wide ranging, especially on the commercial side. So there, it’s quite handy to have something like generative AI that can help you summarize it, summarize the situation, summarize the case. But again, it’s again looking internally. It’s looking at your fact base, your knowledge base. It’s not what some people think of as Gen AI as a large language model that’s public and being used, you know, for everyone on their mobile phone, for example. So I think the other areas where you’d stay away from would be similar to, like the payment fraud dispute example, areas where it’s it’s very clear, and you know, very clearly set what the process is, the policies are, and there’s a high amount of interaction. So unless you want some conversational element, Gen AI is probably less applicable in some of those examples.

Whitney McDonald 12:18:06
I’d like to close out with with this final thought of, let’s say you’re an institution that’s looking into Gen AI, approaching it, you know, kind of starting to think about what their what their pathway or journey to Gen AI should be, what would be step one before taking on a project. You know, you can, you can think about, what will the returns be? Or, you know, how will this improve our operations? But really, what’s step one? If you take it down to the bare bones,

Steve Morgan 12:18:36
I think step one is, what’s the art of the possible? You know, if you’re there, thinking about your area, whatever processes, whatever area you’re working at in the bank, what is the art of the possible. And then when you think about the process and what’s possible, you then, obviously, by implication, look at how to transform, re engineer the process, but also what supporting technologies can do. So So you look at the non technology and technology aspect of an end to end process, and yeah, think where do you want to get to and what’s possible? And I there’s a number of clients doing this in different ways, having different ways, having things like hackathons, where it’s a mixture of idea generation, plus using technology to come up with a, you know, a prototype solution quickly, in a matter of hours a day. I think those sorts of things are very effective, because then you’ve got some ideas gone through some process, you’ve looked at Tech and non tech processes, then you can put together a really strong business case, get support and put together a good project team, you know, after that. So I think that some people call it ideation. I don’t particularly like that word. Things are made up word, but yeah, like sort of

Speaker 1 12:19:37
creating ideas, ideation and hallucination, yeah, made up words.

Steve Morgan 12:19:43
It’s basically coming up with the ideas and where you want to focus, and then that leads to everything else. I do think Gen AI and AI and automation right now has such a big impact potential on most industries, especially banks. I do think the change management pieces are going to come more and more important because it’s affecting it will affect lots of people’s jobs, processes, the way things are done with the customer, so the change management becomes critical with

Whitney McDonald 12:20:08
all of it. Well. Thank you so much for joining us on The Buzz. It was a pleasure to have you and talk about all things. Gen AI, looking forward to having another conversation about this in the future.

Steve Morgan 12:20:17
Thanks very much. Whitney, it’s been great being on thanks.

Whitney McDonald 12:20:22
You’ve been listening to the buzz a fin AI news podcast. Please follow us on x and LinkedIn, and as a reminder, you can rate this podcast on your platform of choice. Please be sure to visit us at finaI news.com. For more finaI News. Thanks for listening. You.

Transcribed by https://otter.ai