How Vodafone, Rabobank and others are driving meaningful results with AI - Interview with Matt Healy of Pega

0:00So welcome to the next edition of the Punk CX podcast. With me today, I have Matt Healy, who is the Senior Director of Product Marketing for Intelligent Automation at Pega. He also leads Product Strategy and Marketing for Pega Platform. That involves helping drive, roadmap, and go-to-market strategy around AI, automation, and low code, all with an eye toward increasing business agility, operational efficiency, and developer effectiveness, no less. Hey, Matt, how are you doing? Welcome to the podcast. Doing well. Awesome intro. Thanks for having me, Adrian. I think I finally found my home on the PunkCX podcast.

0:35People have been calling me the Sid Vicious of AI for... I don't know. Who knew? For minutes, at least. You claimed it. So in that intro, anything you'd like to add? Anything we missed out? Anything that's substantive that you want to say kind of on top of the intro? No, I think you nailed it. You know, I'm responsible for the go-to-market around the Pega platform. And, you know, large enterprises, the world's largest brands, in fact, use the Pega platform to drive their mission-critical operations and customer service functions. So, you know, I'm really responsible for engaging with them, understanding their vision in those areas, and then really helping them, you know, transform those areas of the business faster, which I'm sure we'll talk about.

1:19And then, you know, figure out pragmatic, but super effective ways to adopt AI at scale. Perfect. Now, one of the things that we wanted to talk about, one of the things I wanted to talk about is that I've known Pega for a while and Pega's, the boat sort of shifts and lands on different, it develops new stuff and shifts and lands on different sort of like kind of themes. And I think that the way that you guys go about it, I really like that you're really focused on two things right now, legacy transformation, but also why it's important for organizations to have a real strategy with respect to AI agents in order to get meaningful results, other than just applying them everywhere and then crossing your fingers on hoping for the best.

2:03And so of those two things, I wanted to explore them a little bit with you, if I may. So if we can start with legacy transformation, I mean, it seems to me that the way that many folks often talk about legacy transformation, or sometimes they don't, you might think it's not a big issue or challenge. And then they'd be wrong, right? Yeah, I think that would be wrong. You know, it is funny, like the sort of presence of legacy debt doesn't get a lot of airtime in maybe like mainstream media forums. We're always talking about the future and AI and, you know, automation and the cloud and all this stuff.

2:46But, you know, if you talk to anyone at a large enterprise who's working either with or within the IT department, they know that they have hundreds, maybe thousands of systems which they need to get off of. They're just non-strategic. And, you know, they're dragged down as a result of these. We actually recently did a study that found on average large enterprises are spending$370 million per year on legacy debt. And, you know, that's spread across the more apparent costs like operations, infrastructure, licensing, maintenance. But then, you know, there's a multitude of hidden costs with legacy debt impacts to your efficiency or your customer experience or your employee experience.

3:37So it's a big problem still. And so I want to just dig into the sort of the legacy debt sort of term. I mean, because debt, that's quite a heavy term, you know, depending on your relationship with money sort of thing. It can sometimes have a lot of negative connotations. I mean, is it all bad? No, it's not all bad for sure. And I think that's one thing as you start to dig in with enterprises is there's no like one size fits all approach like to, hey, you have to get off of all of your legacy systems. A lot of these systems like, you know, mainframe is still, you know, widely used. I think like 70 % of Fortune 500 organizations still use the mainframe.

4:18And there's a lot of workloads that won't get off of mainframe in probably my lifetime. And that's fine because they work. It works for those use cases. And, you know, those are areas where maybe they don't need to drive as much innovation. Right. But, you know, there's still tons of workloads which would benefit from, you know, unlocking that data to drive AI. Right. from new levels of automation, from exposing them to customers for more self-service use cases, from improving the employee experience and giving people better platforms to work within. So you got to take a look at your inventory as an enterprise, figure out what's your overall strategy, what are sort of your principles for either of those transformations and then take it from there.

5:03Yeah. No, I just think it's important because I'm going to see these terms get like slapped on and you're like going, oh, that feels like a... It's like you're putting on this big rucksack and it's like wow it's this big heavy load you're like wow but is it i don't want to understand if it's that bad but but i mean i know you're talking about 70 of these big enterprises still rely on some of these legacy applications or mainframe applications and things i mean can you give us a bit of a flavor of i know that i've seen some demos and stuff a bit of a flavor of some of the sort of code and systems that we're kind of talking about kind of here because it's going to allow us to go and see, look into what's in the cupboards in the back of these storage rooms.

5:43You're like, yeah, these things are still there. Yeah. It's funny. I was actually working with a state agency and they were telling me that one of their core systems for orchestrating welfare programs, it was actually developed under a grant which JFK signed off on. So still housed in the mainframe, you know, and as a result, it's like receiving welfare and signing up for those programs and getting your benefits. That's like a paper-based, document-based process in that region. So, you know, what we were talking about is what would it look like to begin to extract some of those core workflows, some of that business logic, bring it forward to the cloud and be able to deliver the constituents an improved experience where someone could sign up for welfare and check their benefits, see status on their phone.

6:39Right, okay. I see. It's like giving people access to some of those sort of things. So as you say, they can self-serve, but also can bring some of that technology, some of those processes, some of that data to life, I guess, or at least open it up so you can make more choices about it rather than being stuck in a cupboard somewhere in the back. Yeah. Yeah, exactly. And, you know, there's many different approaches for a different use case. That same state agency might analyze a legacy system and say, it's fine to stay here. Or, you know, there's also other approaches that are like, oh, should I, you know, retire this application and purchase a new sort of more off the shelf solution to manage this area of the business?

7:21So it comes down to, you know, having sort of those principles and analyzing your systems and then figuring out the right approach for the right type of workload. And so what's the art of the possible when we start applying sort of this new agentic AI technology to some of these legacy transformation challenges that many organizations are facing? Yeah. I mean, these large enterprises, they're no dummies. So while they may have tons of workloads still in on-premise legacy systems it's not like they haven't tried to get off them in the past or at least like had that conversation um and you know one one stat which sticks out to me is is 70 percent of monetizations fail yes and the reasons they fail are are really it comes down to one traditionally it's taken too long you know these are five-year projects to get off of a mainframe system or get off of, you know, a custom.NET system or whatever, you got to bring in a bunch of consultants to read through the code, generate a ton of documentation about what this application does.

8:36Because the reality is oftentimes you don't know what it does. If there is documentation, it was written, you know, 20 years ago, it's out of date now, and there's no comments in the code. It's just, so you have to understand the application and that's traditionally taken a long time. Then you've got to align people on the path forward. What are you going to bring forward? What are your areas to introduce new optimizations? What's your sort of rollout plan? And then you actually have to go through the process of creating new systems to replace the old systems. And that's been traditionally more custom development.

9:11So overall, you're talking 10 years, 50 million,$100 million to drive a legacy transformation program. And I think that's further complicated when you look at, you talk about the length of these programs. And if you think about the average tenure of like an exec, a leader, a professional in one of these enterprises, the average is probably less than five years. And so you end up with this employee kind of turnover and the priorities can like change, markets shift and all these different things. And so it's no wonder some of these big things kind of like, you know, have a hard time passing muster as it were.

9:46Yeah. Yeah, I'm a big sports fan. It's like, you know, a general manager in the NBA. They're not going to go make a big swing to like trade their star player because, you know, they'd rather ride it out for five years until their contract ends. And so some of these are traditionally higher risk, higher reward moves. And I think where agentic AI has really introduced a new opportunity is by really accelerating the end-to-end journey and being able to take off some of the manual work that's been traditionally run in terms of understanding these systems and beginning to build that path forward. Yeah.

10:24I mean, so I think that one of the, because one of the barrels for the bigger things, I mean, I saw this really cool demo when I was watching some of the keynotes at Pegaworld this year. And I mean, unfortunately, I couldn't be there this year, but I was able to watch it sort of live streaming. and one of the ones that stood out to me which speaks to this kind of point there was this old credit management system COBOL app that somebody was running they were filling it all in but doing a voiceover and it was all done as a screen capture with the real time sort of watching the screen and then doing the voiceover and they took that and tucked it into Pega Blueprints and it was able to spin up this is what you're seeing this is what it would look like kind of modern kind of platform and i was a bit like for me you have all these kind of whiz bang sort of other sort of announcements and i was a bit like that's the shit that's the thing right because some of these problems are very much holding organizations back i mean i know that there's a um i remember going to a course was it late last year it was at the copenhagen institute for foresight studies.

11:37And they had this model about sort of trying to predict what the future was. And they said, like, on one point, based on a triangle, on one point they had, you've got forces that are pulling you into the future, which might be the potential of new technology and so on and so forth. You've got forces that are pushing you into the future, which might be your desire to grow or expand into new markets or even just to harness some of this new technology. And most people kind of focus on those two sort of things. what they forget is the third point of the triangle, which is the anchors, which represents what is possible based on your current state of being, as it were.

12:16And I think the legacy piece very much speaks to that. And I thought the demo, the application of that old COBOL, this is COBOL that's possibly been written like 50 years ago. And there's a complete, drought and the number of COBOL programmers out there that can potentially understand some of this sort of stuff. So to be able to take an old thing and just narrate a user experience and then have it ingested into an application that then spins up a, here's what it would look like in a modern architecture that you can test and you can further develop and so on and so forth, I thought was crazy. But I also know, I'm going to shut up now, because I'm also going to know that And that's been a big, based on this legacy transformation thing, I think that's also a big area of focus for you is how can you kind of almost like widen the funnel in terms of the amount of inputs your blueprint could take as the entry gateway to your kind of platform?

13:18And I want to understand what sort of things have you been working on to kind of fling open the doors to give people more possibilities around that legacy transformation? Yeah, yeah. No, it's been a super exciting journey over the past almost two years. So about two years ago with all this generative AI stuff, I'm just going to give you the whole sort of, you know, origin story of this blueprint capability. So, you know, we're always doing studies on, hey, what's the long pole in enterprise transformation and really in Pega implementations? And we were finding that on average, teams were spending three months in what they were calling sprint zero.

13:59Okay. Sprint's supposed to be two weeks, but here we are, 12 weeks in. And, you know, really all that time is really requirements gathering. Right. All right. What do we need to solve for? How are we going to architect it based on the requirements? How are we going to turn that into a backlog, user stories? You're sending Word documents back and forth. You're doing murals and whiteboards and all this stuff. Just takes a long time. So we're like, hey, congenitative AI and really a new sort of approach to business and IT collaboration help here. So we introduced this capability called Blueprint.

14:35It's great. It uses AI to do research on best practices and help give you a starting point based on your use case. And then it allows business and IT teams to work in the same palette to capture their requirements faster and then turns that into the starting point of your application. So it's been great. We had Vodafone at PegaWorld talk about how they're using that to deploy new workflows to automate their network operations in 40 hours. Okay. From idea to a new automation in under a week, which is unheard of. But, you know, that really picked up for Pega Projects. But, you know, the enterprises we deal with, they were like, hey, hey, dude, like good stuff.

15:16But I'm not starting from scratch. I have a legacy system here. So we were like, oh, OK. So what can we do to, you know, really turn the lens and apply AI in helping understand what's already in place and then use that to inform your future state architecture and requirements? So over the past couple of months, we've begun to introduce a set of AI agents, each which are sort of purpose built to understand like a different medium of information. So we started with, you know, process diagrams, things like BPMN models and turning those into workflows in Pega. We expanded to data models and being able to take in like a database export and turn that into, you know, an integration map in Pega.

16:04And then we started to get into documentation because people were like, hey, I have a standard operating procedure or a user manual for an older system. Can I turn that into new workflows? And we were like, yes, AI is great at that. And then we started to get into, I have no documentation. I have no assets. But I have this system I need to get off of. And that's particularly kind of like, I think, kind of prevalent in this kind of like space where you have enterprises that go and acquire older firms, right? And they acquire them and integrate them. And then stuff just gets lost, right? When people cycle out and stuff gets lost and they're like, we're using this thing.

16:45and it's really important, but we don't know how it works. Yeah. Where's all the paperwork? Yeah, yeah. You do a merger, you know, you might like cut the workforce a little bit or whatever. And then there's just brain drain in terms of like the people around who understand what these systems do. So we were doing a complaints application replace. An enterprise, an insurer was trying to get off of Lotus Notes, which was driving their customer complaints. Blimey. Lotus Notes. Lotus Notes, yeah. Good stuff. I had never seen it before, to be honest. I don't know if I want to see it again. No, exactly.

17:23So, but we were like, hey, you know, you have no documentation, whatever, but could we sit down with one of your users who really understands the complaints workflow and, you know, just record 30 minutes of them actually using the application in the ways that they would walking through, you know, I would click this, then click this and here's why and whatever. So we recorded that session and we built an AI agent, which could take both the transcript. So that, that person narrating their day, and then also actually interpret the screens itself. Right. So we now have that productized and it's part of blueprint where you can import a video of a legacy system and it will extract the user journey, the user flow, the data elements, and then any user requirements as well to build out the framework for your application.

18:16Wow, that's cool. I mean, because it also kind of feels like that's great because also it feels like it also speaks to this idea that oftentimes people in organizations, like you said, they have a business operating kind of model or a business operating procedure or they have a process kind of documentation and they have all these different things. And then you're like going, yeah, but speak to people on the front line and they'll tell you a completely different story because they'd be like going, yeah, that doesn't quite work. But this workaround and all these different sort of things. And I guess that sort of helps capture the variance between what people think is actually happening and what is actually happening as well.

18:56Yeah, a thousand percent. And oftentimes in these legacy systems, there's like you have your code base or whatever. Eighty five percent of it is old, never used. Right. So it's important to like also sort of start with what's actually happening on the ground. And a lot of this stuff you can just ignore. And it's like true in the word debt that you can sort of get rid of. Yeah. And sometimes it's like, actually, sometimes it's just not obvious. You know, you think you've got debt and then you go like, well, there's the debt and then there's the real debt. And it's always like you need to shine a light on all of it to be able to kind of capture it because otherwise you're building something which is just not fit for purpose.

19:36Yeah. Yeah. And then so, you know, we've also started to work with our ecosystem on this. So we have a big partnership with AWS and a number of SIs, like, you know, the large guys, Capgemini, Accenture, whatever it may be. And they're all also driving, you know, legacy modernizations and transformation projects. And they have oftentimes platforms which can actually analyze the source code and produce documentation on what it does. so for you know maybe about 15 of these platforms we've actually built integrations with blueprint to ingest their source code document analysis and turn that into new applications in pega as well nice nice crumbs yeah well you've set up the tools and there's a lot of work to be done so i guess people just need to have at it really absolutely yeah and we have uh you know we have a lot of really good projects that are in flight and we're going to be showcasing this at reInvent too so if you're going to be there catch me well i'll be doing a talk about mainframe modernization and some of the client successes that we've had oh you get all the cool gigs mainframe modernization don't know maybe it's punk i don't know is it well i'll tell you what i kind of like it might be you know just because in in current times it might be i mean at the time of recording, I just wrote a piece around somebody doing, putting AI into on-prem software because some people are just going, I still want to be on-prem based on my estate and different sort of environments that I'm in.

21:20And you're thinking, well, why should they have to miss out actually? And I just thought, I like that. That's cool. Yeah, definitely. I think we've done a lot of work on like sovereign cloud same ideas so being able to deploy within various regions and have it all run there yeah so we talked a little bit about uh legacy transformation now i wanted to come back to the second point i started with which was why it's important for organizations to have a real strategy with respect to ai agents in order to get meaningful results i mean let's start with i mean there's a lot of noise out in the um in the um in the marketplace i mean what are you seeing in terms of what's going on and the kind of the rhetoric and all these different sort of things, what's working and what's kind of like kind of not?

22:08Yeah, there is a lot of noise, a lot of hype. You know, I feel semi-responsible for some of it, but it's all good. You know, the reality is I'm just a little bit of a skeptic by nature. Right. But I got to see something working in action in order to be able to trust it. And I think a lot of people are like that. And I don't know if there's been a more used or trovanon stat than the MIT study. You know, I keep away from that one. I prefer the IBM and the Accenture ones just because they've got a larger sample size. because the MIT one, I think, was too small of a sample size to be properly representative, but it's all in the same direction.

22:56A lot of people are trying. Many people are not being able to scale. Most people aren't being able to generate a substantial value out of this. And so the thing is, it's like, this is hard. Yeah. And there's a lot of, I think, reasons to that, which we could get into. But one of the main things that we're hearing, We work with a lot of like regulated industries and they're like, yeah, AI, we're doing, you know, AI projects across the business, but we've yet to do anything really in our core operation or our core customer service function. And the reasons is because of trust. I think we've seen some of these stories like Air Canada deploying a customer service agent that gives a bad refund or the McDonald's drive-thru that orders a billion burgers by accident because this thing wasn't appropriately governed or whatever.

23:50I saw the best one I saw, I think, was somebody who effectively conned a bot into selling them a car for a dollar. And they had to sell the car for a dollar. I love it. Life hack. But yeah, the reality is, right, like in a regulated industry, you can't deal with hallucinations. You can't, you know, you can't deal with variability. Things have to be done the same way every single time, following your rules, following your standards and, you know, driving repeatability and auditability across the board. Yeah. So that's really been our focus is, you know, introducing AI agents, which enterprises can trust, rely on and aren't, you know, subject to a lot of the hallucination.

24:44But also the other side of it is are maintainable because, you know, if you were to try and build the most reliable AI agent or consistent AI agent out there, but using a traditional AI agent development approach, you would build, you know, you'd go about it by building a prompt, which is the size of, you know, the Bible or war and peace. Right. You would have to write every single rule, give a million examples, and then publish that to the AI and say, follow this. And then what would you have? You'd have all of this text, which you'd have to maintain going forward. So what if your process changed?

25:23What if your rules changed? You'd have to go and control F and find every instance and blah, blah, blah. So our approach is about enabling predictability, but not by introducing 1 ,000-page prompts. It's really about defining structured processes, structured rules in the ways that business people approach them today, and then having the AI pull from that in terms of its behavior and be governed by those frameworks. And is that what, because I've also seen Alan and Don, that's Alan Traffler and Don Sherman, kind of talk about the right approach to implementing kind of AI. Is that what you mean by that?

26:00It's actually rather than just going, you give everything over to the, you know, you write the war in peace and you give it over to the, as a prompt and you give it, and there's so much kind of variance because you could submit that one question to the agent and then do five minutes later. And there's a likelihood that you might get something kind of different just based on kind of the way that these models are set up. But is that what they're talking about, the right approach to implementing AI? Yeah, exactly. I mean, the sort of reasoning aspects of these AI agents is awesome for creative work, really dynamic work.

26:35Like, yeah, have it generate stories for you and it'll be different every time based on the same prompt. And that's awesome. But if you if you want to get a customer service, you know, incident resolved end to end and you want to make sure it follows your process, you got to have a structured process on the back end, which is pulling from. So the way that we've approached this in our latest release is with what we call predictable AI agents. OK. Essentially, what they are is it limits the amount of reasoning which is baked into the agent. And what the agent acts as is almost like a conversational concierge to the workflows which you want to drive anyway.

27:20So your structured processes. So if a customer calls in, it won't make up a way to process it, but it will find the right process that exists in your system and walk them through that. Okay. And I think the important thing here for me is that it sometimes seems that some of the talk is around whether it's agentic AI or AI agents or agentic AI and things. It's always a bit like, oh, you can just throw away everything else that's happened and just can rely on this and they'll figure it out sort of thing. But actually, there's, and I think this is something that Alan and Don were sort of referring to, is that sometimes and oftentimes, there is value in the way that you do things.

28:06Because that can be part of how you differentiate yourself or just because it's how you approach things, how you do business, and that can make you a difference. And sometimes the value, there's IP in the process. And so therefore breaking it down and having it controllable and auditable and governable and all these different sort of things is starting from how do you want to do business rather than having something else tell you how you want to do business is a way of going about it, if that makes sense. Yeah. Yeah. A thousand percent. I mean, you know, your processes, your rules is what makes you a business, right?

28:43So leaving that up to just the inherent knowledge in a large language model or the hallucinations that come with it is just not a recipe for success. And there's another thing that I know that Pega folks have been talking about, which I want to ask about, which was about being very intentional about the sort of tools that you pick for the job or the problem at hand. because we all know it's like, oh no, generative AI or this or that. And I wanted to ask you to explain that because I think it's an important kind of point that doesn't necessarily get talked about as much as it should. Because we have to concede that AI offers all of these different sort of tools, but it's not like a one tool fits all type of thing, if you know what I mean.

29:36yeah yeah definitely so you know there's that age-old thing when you have a hammer everything's a nail or whatever um so if you if you start a project and you're like i need to let's go back to complaints i need to transform complaints and you open up your ai agent builder you're going to build the whole thing in ai as an ai agent even if like some of it may be more fit for a more traditional automation or statistical AI or whatever it may be. So the way that we sort of approach it is starting first with the customer journey itself. So what are you trying to drive? What are the major milestones of that customer journey?

30:17What is the work that needs to get done within it? What are the checkpoints? What are the regulations you need to pass throughout it? And then within that, you can fill in whatever makes sense. Do you need like a human to fill out some information? Do you want to call on an AI agent? Do you need to integrate with a legacy system through an RPA bot? So it all starts with having that sort of framework, the end-to-end customer experience you're trying to drive, and then within that, being able to fill in the right tools for the right job and have that whole thing orchestrated from end-to-end. Yeah.