PegaWorld 2025: Reinventing Work with Agentic AI

PegaWorld 2025: Reinventing Work with Agentic AI

Welcome everyone. My name is Nils Mueller. I lead our process automation practice on Europe and EMEA level for for Accenture, and I'd like to invite up Marcel Sohnius from Allianz and Vitaly Graf, a colleague of mine from Accenture, and we'll be speaking about legacy modernization using the power of GenAI.

So welcome to the session. MGM. MGM Grand is, I think their claim is the the home of entertainment or world class entertainment. So three German guys on stage, what could go wrong? Right. For world class entertainment. But we will try to to keep it loose and fresh. And if you have questions, um, there will be chance to to ask questions at the end.

I guess let's maybe go through some prepared questions and have an exchange here on stage, like a panel. And then at the end, um, feel free to ask your questions. I'm sure there will be will be many. And also, if the time then is, uh, not allowing to get to all the questions, come to our booth and we'll have a separate discussion then afterwards.

Okay. Let's start maybe with some setting the scene. Marcel. Um, maybe you can tell us about the challenge you faced or, I think, still face with your application portfolio and some of the constraints that you find yourselves in. Absolutely. Um, what a great crowd to talk to. So thanks for having me.

Um, I'm my role is officially head of it for AGC and AGC within the Allianz Group is the global line for taking care of the industrial insurance and commercial insurance. So basically my my responsibility is all the IT delivery we have towards uh, AGC and S and Allianz commercial for all the different IT uh, services that need to be provided to support the the value chain of AGC, that means we have a quite a quite a bit of a zoo of applications, roughly 300 different applications we're supporting.

And we're doing this on a global scale. And, um, from, from a technology perspective, um, we have quite a bit of legacy with regards to technology, with regards to, um, different setups, with regards to different locations. And um, obviously there's quite a bit of cost pressure, um, providing these applications as cost efficient as possible, and there's quite a bit of pressure on the regulatory side to do this in a, in a very, um, sustainable kind of manner.

And, um, looking at this, it's, it's quite imperative to make sure that we consolidate this application landscape to basically reduce the size of this zoo. Okay. That would have been my next question. What's your vision where you want to go. So consolidation reducing the heterogeneity making it simpler to maintain to operate.

And then also I think easier to drive in regulatory changes on a on a homogeneous platform. Is that sort of correct? Absolutely. It's basically making sure that, um, we, we are able to, to provide solutions to our business that create value, because in the end, we're basically not not doing this for an IT perspective from from a pure IT perspective, but rather to create business value for, uh, to support the value chain of our business.

So the question really is what if we can leverage AI and gen AI tools to basically consolidate our application landscape to make sure that we're, um, we're creating a consolidated application landscape that is basically reducing our IT footprint and creating cost efficiencies as much as we can. I think that's what many here in the room, if they're responsible for it, states having the same challenge and looking to do it.

Let's introduce our third panelist here, Vitaly. Maybe you introduce yourself quickly. Yeah. So I'm part of the, um, legacy transformation team. So I'm looking for our EMEA region and I'm also looking after enablement. I'm personally a Pega enthusiast for eight plus years. So very biased here. If you listen to me.

Maybe next question also to you, how did this pilot come together or what was your role in executing the pilot? Maybe you could speak about that. So, I mean, to circle back a little bit. We met with Marcel. I think it was six months ago. Eight months ago, a little PegaWorld in Europe. And I still remember how he walked up to our booth and he was a bit skeptical about our solution.

And he was like, oh, there's GitHub that could maybe do something like that. And then I showed him what our solution is. And I'm really glad that we got the opportunity to work together, because that was really the start of the last couple of months where we started with the legacy transformation for Marcel.

Um, yeah. Does that answer your question? It does. Okay. So how did you approach this collaboration of three? Right. It was a joint project between your organization Pega involved and then Accenture involved. How did you sort of see this collaboration? Yeah. Maybe we should give a bit of context, because what we chose, um, after this session at the small PegaWorld, was basically a use case where one of our legacy applications, which is used for underwriting.

Um, would basically um, and we would challenge the, the idea of how can we use AI tools to re- engineer the business logic out of this legacy application and then basically inject this into the Pega Blueprint solution and by this division. And that was the challenge I gave you guys. Um, basically, um, was how can we create, can we do this as quick as possible with an ROI that basically, um, proves within one year and um.

And maybe to, to sort of, um, build on that, maybe you can describe a little bit what our asset Gen Wizard does, because I think we've heard a lot about Pega Blueprint. And I think most of you will be by now aware of what it what it can do. But, um, we also have an asset that kind of directly feeds into that.

And I think you're the perfect person to speak about it. Sure. So the tool is called Wizard, and it's a tool we've been developing for many years. It's had different names, but Gen Wizards has already the name Gen I in it. So it's Gen AI powered, but powered with lots of other tools and it has lots of different functionalities.

One of the functionalities that's the core that we were using in this case is reverse engineering. Right. But we also have other things like forward engineering, um, generating documents, generating code infrastructure set up lots of different things. Right. So in this context, what we essentially do is we feed in the legacy code and we feed in the documentation.

So we received some handbooks, some training materials and stuff like that. And it processes that and then generates artifacts from that that we can then use through the API to then get it over into Blueprint. And that's where we then did a preview and playback and it becomes alive. It's kind of like a zombie that's becoming awake.

And you can see it in Blueprint live. So that's a cool experience. Okay. So that covers the reverse engineering and then the team once once we did that I mean it's I think an iterative process. Right. It's not like a one shot thing. You plug the tools and everything works like magic. Maybe you can speak a little bit.

What is the end to end process and what are the iterations? We we went through some of the challenges. So we follow the process where we started with the discovery part. So for discovery, um, as I mentioned, we uploaded the code and things like that. And then the tool has to start working. So we had three iterations in this case, um, it took us I think the discovery was about 3 to 4 weeks in this case.

Um, we had a bit of a setup time in this case, and it was a pilot. Right. So there was a lot of learnings for us as well, and a lot of improvements from this as well. Um, and that's where we had the first outputs. And then we still need a human in the loop. So it's not magic that we suddenly can just get everything done with a click of a button.

We need someone to review the results. We need someone to say, oh, this needs to be fine tuned. So in this case, for example, there was a particular scoring logic where Jen Wizard would just say, yeah, there's a calculation happening, but we needed to know the exact details of that calculation. And that's where we had to re-engineer Reprompt and do the next iteration to get better results.

So that's the kind of first part, that discovery phase. Again, over to you. Maybe. What were your biggest concerns when we were then after the initial discovery phase, we were going into implementation or the biggest challenges you sort of expected? It's a good question. First and foremost, the question really was, is I delivering what it promises?

And is this there's this teamwork of Jen and I, Wizards and and Pega Blueprint. Is it really creating the value that we envision it to do? So I was a bit skeptical and, um. Um, but in the end, it was very interesting to see how this evolved and how we were able to use this iterative process with a lot of tools, support and a lot of support by the Accenture team and by the Pega team to and then discuss this with our business to basically understand, okay, what is it really creating and how can we, um, how can we make sure that this is actually flying?

So I was a bit skeptical, um, especially on the question, is this a pure prototype or is it something that we can eventually put into production? And I think it turned out to be the latter. Right. So, absolutely. So this is why I'm quite I'm quite happy about the outcome. It's not done ready yet. We still need to add a couple of use cases to the application.

But it is something that is accepted by the business and perceived as as an improvement to the current situation on the legacy system. It has had just three months of of development time, which is super fast and it is really integrating into the Pega application landscape. We already have within my remit.

Let me double click on this improvement topic you mentioned, because I think initially the guidance to the team was bring it into Blueprint, but don't change a thing, right? Leave everything as is. We don't want to confuse the business. It should work exactly as before. But I think what we ended up with was something a little bit different.

Yeah, you can't just take something and put it into into Pega because there is automation you will get with Pega, right? Like for example, there's a certain way of working where your colleagues have to upload a PDF document, attach it, or something like that. Um, you don't do that in Pega. That's part of a workflow. PDF document is generated, is routed to to someone. So again, that's where the human comes in. Because you still need to think about how can we utilize Pega's out of the box features to make that process streamlined inside Pega? So I think you've seen many IT projects in pilots in your in your day, right?

You've been with Allianz for quite many years. What made this one feel different in terms of speed and outcomes and maybe also transparency to you? Speed for sure. Um, getting into to this result within three months and probably four weeks of of this was basically setting up the infrastructure. So um, so the, the, the net amount was, was really short.

Um, so speed was very, very important and very impressive. Um, the iteration or the iterative approach, um, how we could basically do the re-engineering, get into a workshop with the business, and present a Blueprint draft of of the B2B application. Um, being able to showcase this in a, in a final workshop and getting the the business approval of this, this was really, really impressive and, um, certainly different to other IT projects I've seen.

Okay. Um, was there a moment during the journey when somehow it clicked for you when you said, okay, because you mentioned you were skeptical in the beginning, right. So what was this moment when you said, okay, I now really start to trust and believe in this solution? Was it when you were presenting to business and getting the positive feedback from them or some other moment?

Pretty much. Well, my my IT technical team was was quite fond of the solutions. So that was a plus. And I believe them definitely. So and then we had the the workshops with our business presenting the interim solution. And the last bit was when we presented the result of the the pilot and asking business, are you happy with this?

And they you could feel and sense the energy. Yes, we are happy with this. And this is even better than our legacy application. If I can add to it, I remember some of your colleagues were like, oh, you're offering us so many options. It's like, I'm in a candy store. And I could also do this and this and this with Pega, which I couldn't do before.

Exactly. So a couple of features come out of the box, like reporting, like document creation, what have you. And this is very easy to to augment the legacy application we have. Before we take a look ahead to what's next. Maybe one more question to you, Italy. You were very close to the technical implementation, right?

Is there any specific challenge or anything that that you encountered where you say, next time I'll do a project like this, this is This is something I would watch out for from the beginning. Some lesson learned basically. Yeah, I think the the, the part where we lost most of the time was the setup time.

And that's something that we will definitely improve. Um, the other part is we have pre-developed patterns. And that's part of our IP. So for Java we have a pattern. That's why we could very quickly get things started. Um, but I think there's areas where we can further improve, for example, on the user story generation, which is one of the artifacts we generate to be more tailored towards Pega requirements, because that would then shorter the application build time even more.

Okay, thanks. Then maybe let's switch gears a little bit and start to look forward. So, Marcel, you're responsible for a large portfolio. I think I heard 350 applications. This is just one of them. So what's your sort of thoughts on how to potentially now scale this for a larger portfolio? Maybe not every application is suited to be migrated into this direction.

Some will be, I guess. So what's your idea of how this could scale? Well, number one, um, this this kind of approach can be utilized for a couple of other applications I'm responsible for and basically drive consolidation of the application landscape in a quick and efficient manner. So, um, we're currently investigating on other applications that that would basically fall into the same pattern that we can leverage and reuse here.

A plus for sure is that we already have some sort of Pega infrastructure and Pega application landscape in my remit. So basically it's it's more like adding some more Pega applications to the infrastructure we already have. Um, as opposed to having to set up this on a, on a, from, from scratch so that that really helps.

Um, if key enabler for for this for sure is the capability of the tool to basically re-engineer whatever legacy technology we have. Java maybe was an easy one, but we have some pros and some Lotus Domino stuff for. The challenge, which. Which is certainly a different kind of challenge. But um, but I would be positive that we find some sort of a solution for this.

So looking into this and making sure that we use AI and create the the speed and the ROI for, for this endeavor, I think is key. Okay. And I also think the selection of this particular application was an important process step as well, because not every application will fit into a Pega target application.

Right. And maybe I can add to this I had in the Innovation Hub or what it's called in Clyde conversation earlier. where I made the point that there are some applications which you potentially can lift and and adapt to Pega 100%. Right. It just goes to Pega. End of story. There may be other applications where you want to choose a kind of a hybrid approach, right, where you maybe want to retain the data layer, maybe some application API layer and a traditional architecture, cloud based architecture, whatever, and then leverage Pega on top as a process automation and UI layer agility layer.

That could also be an alternative. There's no one size fits all solution. I think you have to look at each application individually and find out what's the right target setup for that application. Definitely. And maybe to add certainly the end game in the legacy decommissioning business would be to make sure that we are able to migrate the legacy data out of the legacy applications so that we can really shut them down and make sure that the zoo doesn't get bigger, but a bit more restrained.

Actually capitalize on the on the cost saving, right. That was a in the briefing that you gave me one of your key concerns, right. It's not only about modernizing and having new, let's say, cases run on the new new platform, which is the nice for the business from the user experience side and so on.

But you actually want to capitalize on the cost savings for which you have to really decommission the old system, which sometimes in regulated environments, you need to be able to still search the archive for older cases and so on. So this is something complexity that comes on top that we obviously have to keep in mind.

Okay. I think we're about to kind of open it up for questions from the audience. But maybe before we do last round of kind of key takeaways from both of you, maybe Vitaly, you can can start. Um, key takeaways I would say like don't underestimate it. Um, especially like you mentioned, there's a lot of things like data migration as well. It's not just getting the processes out. Um. Think about your ways of working, because it's not just technology, it's people that are involved here. And I think we will have huge savings in terms of time for your people. If we look at how much quicker they'll be able to work in Pega. And I think the next question is, how can you now utilize their time smartly to maybe be more creative and innovative?

So yeah, there's a lot of things to consider. Thanks, Marcel. Any final thoughts from you? It's a mixture of speed, technology and collaboration. And maybe plus one is business acceptance. So making sure that all these parameters are met is not super easy. But it's doable. And I think that's that was what this pilot and this approach has really shown.

Okay. And if I may, um, being more in the moderator role, but I'll allow myself an opinion as well here. So I think what what is we shouldn't be so naive to think we can just pluck two tools as great as they may be, and then get an outcome that we can start more or less to use, right? It's not. It's not like that for sure.

Not. So you still also need people that understand the legacy technology that are in this case, it was a Java application that can go in and read the Java code, understand what the application does, because as we all know, and Ellen showed it in the morning I think in a really cool way. GenAI get stuff wrong at times, right?

And you need to be able to assess what the GenAI tool tells you, if it's really reality and helpful, or if it's maybe misguided and you need to tweak your approach. And for that end, you still need also the people that understand the legacy technology, that understand the business process and that can orchestrate an end to end, potentially complex project.

Right. I mean, this was a, I think, a bit smaller application, the complexity maybe not yet so high, but if we're going to larger and more complex application, that becomes a big factor as well. How do you do this? Also this cutover you have the old application running. At some point in time you need to cut over in a clean way.

And we talked about data migration and all of that. So that drives complexity as well and shouldn't be underestimated. And yes, here we are focusing very much on the Generative AI integration of these two platforms, which is amazing, frankly. But it's not the whole story. Right? And don't think it is.

There's much more to it. Okay. With that, I'll open the floor for questions. I heard there's some app with which questions could be submitted, but feel free just to walk up to the mic or shout your question into the room, and we'll try as best as we can to to answer a comment. From Accenture I've just got a question around the initial setup, so I know we talked about Gen Wizard, so how easy was it from Allianz point of view?

Allianz point of view to understand data security. Data privacy. Was it an Accenture Cloud. Was it in the client cloud? And what are the key learnings for anybody who is coming in new in the space to implement this going forward? Thank you. Obviously we we made some arrangements on sharing the the Allianz.

IP with with Accenture who are, from my understanding, using the the GenAI wizard on the Accenture cloud. But this is doable. And we're talking about code. We're not talking about actual customer data here. Right. So this this was was certainly a feasible. And I would not change this approach. And it helped.

We had an NDA in place already. So it was quite quicker. I mean you probably know this right. There's multiple models available. If a client wants to have this tool on premise for some reason, it can be done. We can use it in a hosted cloud based version. There's many options to do that. OK looking around.

More questions? I'll go ahead. Yeah. Sorry for turning my back here on this side. So how did you guys you mentioned that some parts of the business process were changing. You gave that PDF example. How did you handle educating the business or on sort of updating the way they execute the application?

Okay. I think I'll pass that to probably to you. Yeah, sure. Um, this is still a to do. Um, but I think it's important that the business was involved or the key stakeholders of the business were involved in the whole process from a very early stage. So they would actually know how to how to educate their, their users to basically do so.

It's it's still an open to do is say? I'm just asking a quick follow up. Is there an opportunity to potentially have to sort of pull in the delivery of training materials and stuff like that as part of that sort of the outputs of what you guys are delivering? Maybe that's a question to you. So maybe one step back.

Um, to give you an example. The business initially had given us a requirement to connect the tool with outlook because we involve them through involving them. We indirectly educated them about the tool and how you can utilize automation. So then we put the requirement back to them and asked them, are you really sure you still need the outlook integration because you don't.

Right. So that's kind of learning on the job. But I think the question was can we find a way to make the learning or the change education of the users part of the, let's say, delivered application of the delivered solution itself. Is there some opportunity to do that, do you think? I'm not sure I understand the question.

If we can provide training as part of that, or I mean. As part of sort of the deliverable, you deliver this new application and Pega now is Blueprint of all architecture design, all of that comes out of that floating share. Is there a thought of maybe saying, okay, is there material out of this that we can also convert into some sort of.

Like a context sensitive user manual and so on, right. Yeah, we could, we could. So one of the outputs from Jen Research is what we call a wiki. So it's essentially your documentation of this new application. So that could then easily be converted into educational materials and things like that. Yeah.

Okay. Further questions. I'm going to add the GenAI. Right. So use the Blueprint. That's the GenAI part of the Pega functionality to generate the process that you have. Yes. So there's just two parts. We have GenAI in Gen Wizard and we have GenAI in Blueprint. Yeah, that's what I was waiting for. Yeah.

So what's the solution that you've built. So you have the blueprint, then you build, you build the actual solution. So what are the capabilities you use in the end automating process. Second part of the question is were there any parts you used to. Because I know a lot of times people mix and match both the terms. And I want to just call out, I want to differentiate the two. So you have both in your final solution. And how did you use either of them? There was no authentic AI in this solution. It wasn't needed for that. It was just purely GenAI. So what are the AI capabilities in your solution apart from the blueprint itself?

So it's analyzing. It's summarizing. Um, it's generating new documents and artifacts. So like most of the Gen Wizard solution is GenAI driven. I think your question is referring what GenAI functionality are we using in the Pega runtime in during runtime? Not not during create time. Currently we're not not enhancing or the business process with JNI functionality.

Um, obviously this this would be a very easy add on. And this depends on on basically the business appetite for, for doing so. Um, currently we have not implemented this. As I said, main purpose for for this exercise was basically to get rid of the legacy application and migrate it to to a Pega based infrastructure.

Everything else comes on top, I'd say. So you didn't have your own LMS use anywhere? No no, no. Just moving from one. Exactly. So yeah, it's it's a multi-step approach. Obviously, first you want to decommission the legacy application and then then go forward. I mean, this is also something we discussed in our prep, right?

That now once you are on the new platform, it becomes tempting, I would say, to start thinking into this direction of how can we improve? How can we maybe add in agentic steps in a workflow and increase the level of automation that's now much easier than obviously than it could be, could have been done in the legacy.

But as you heard, we are not there yet, right? That's maybe for the future. Step by step. Please go ahead. Well, thank you very much. There were two components talked about here. One is the wizard and the GenAI from Blueprint. Right. The wizard. Um, what gaps did you guys see in the Blueprint capability of ingesting documents and generating these assets versus using Gen Wizard?

Can you talk a little bit more about that? So at that time, Blueprint didn't have such capabilities. And as you might know, it evolves so quickly. Like every week, there's a new feature coming out. Um, the key here is analysis of code. We could upload the documentation we have into Blueprint, but it wouldn't be enough because there's a lot more inside the code.

For example, we need to understand the data model. You can't build a data model only from a training document or something like that. So there's a bit more to that. Gotcha. And is Jen Wizard a SaaS solution as well that you're marketing to your clients? Yeah, we have various approaches. Um, it's one of them.

Yeah. Um, it sounds a lot like when you guys are talking about Jen AI. You did it more of, like, a full application approach. So evaluating existing application and doing how do I get 100% of this application in 100% into Pega. In some cases that's not feasible, right. We're going to identify like what we can do when we can do it, how we can approach that.

How do you do that with this approach? Like, can I continue to use Pega GenAI to help us continue to enhance the application as we go? Or is it literally just fire and forget? Next time you start it, you start over again. Do you guys have any context there? So what's nice about the solution is that we have an open end point output. So this particular situation here was with the target to Pega. But we could easily put this into SAP or Salesforce whatever is more suitable. But of course again the human in the loop is important because you need to decide where do you have your processes that you want to put into Pega and where do you have other things?

Maybe, maybe I can chime in. The logic would be we we used this approach to basically create the Pega application that now we can basically make productive. And from there, we're not recreating this with a Blueprint approach, but rather we're incrementally improving the. The Pega workflows adding GenAI capabilities to to the workflow.

To the solution we created. So it's not like fire and forget and always. Recreate the application with using Blueprint. So if I understand what you're saying essentially is what you've done or what you would propose doing is identify the critical things that you want to do first, and then are you still using GenAI for those added abilities in the future, or are you more.

To create the application? You mean like using taking that documentation you got from your Gen wizard using the Pega GenAI in order to generate your new case types? As you as you start adopting those things that you basically said at the beginning, like, okay, I'm not sure that we have enough value there to do it at this point.

Maybe V.tal you can jump in here. But the approach was basically use the Blueprint output as, as for, for bootstrapping the application and then do some, some Pega customization work. Based on this output. And this GenAI support in the general Pega developer tools as well, right? Blueprint is something that gets you started, and then eventually, once you're happy with the application Blueprint you move into the developer tools.

But those are also enriched by GenAI capabilities as far as I know. Right? And in this particular case, our ask from myself was lift and shift. He didn't ask us to look into forward engineering. This is where Blueprint could add a lot, right? Because once we extract it into Blueprint now we can discuss how can we improve it and make processes better.

Appreciate it. Thank you guys. Thanks. Hi, gentlemen from Sunlife, Canada. Um, you talked about cycle time reduction in development, cutting through red tapes and such. I'm curious to know about this journey as it applies to the testing cycles. One of the things that we most sometimes conveniently bypass is how long it took for someone to test the application, right?

So yes, we built this with this great GenAI we plug this into Blueprint. How did you achieve tangible reductions in testing cycle time? Did you at all use GenAI for testing? So this pilot was a bit small for that. There was a little bit of quality testing with a few users. Um, we have witch and wizard test cases and test scripts that we can generate.

However, we didn't need and apply that and there is still room for innovation. You touch on a very good point because there's a lot of efforts on the testing, especially if we have a more complex application. Thank you. But for sure, the the advantages that the GenAI wizard creates, those test cases that then can be used to to automate your testing.

Yeah. Thank you. I think we have time for two more questions, so please go ahead. If you have any questions. I think we can't hear you too well. It would be good if you. Could. Come to the mic or stand up and speak up a little bit. Or shout. Thanks for making the way up. All right. I think this is better. Much better. Okay. Just curious. Like during lift and shift, was there any reporting that, uh, was getting out of the legacy application and if, uh, Jen, uh, the, the Wizard and Blueprinting if that kind of helped create those same reports or kind of like reduce the effort in any manner. What reports are you referring to?

Basically progress reports of the project execution or reports from the application itself. Yeah. The process. Okay. Does the application sort of even produce any kind of reports? Is that part of the functional scope? That is part of the functional scope. Um, business is requiring certain documents for, for a handover between underwriters and different underwriting teams.

So, um, this was automated and basically, um, part of the application we created. I'm not sure if you're referring to a reports towards regulators, um, which in this application, just the. Operational reports that the business need were they anyway accelerated the the build of those reports. So Blueprint that was one of one of the key moments where business said, oh, that's really nice because, um, the Pega Platform per se creates a bunch of, of different reports that you can use for basically doing statuses of your processes, of your cases and open cases and whatnot.

So, um, this this for sure was one one of the very obvious advantages of, of using and jumping to the Pega platform, because the legacy application would have been a nightmare to basically create all this wonderful reporting. So with Pega insights and then the GenAI capability to chat with your data, because we move the data and the data model into Pega, it's very easy.

You just ask the question and there's your report, right. Okay. Thank you. Thanks. Do we have one more question? I don't think we do. But feel free to come find us in our booth in the Innovation Hub, or just approach us in the hall if you feel like it. Thanks for joining the session and have a great rest of the day.