PegaWorld 2026: Reinventing Legacy: How Pennsylvania and Accenture Transformed Public Service with AI

Yes. Thank you, everybody for coming to our session on Reinventing Legacy. And like Elizabeth said, we accomplished this in 13 months, accelerated with AI. We'll dig into a lot of that. Um, and to get started, I want to introduce myself Carl Holmquist a managing director with Accenture. I'm the technical delivery lead on the ECM program. Good afternoon everyone. My name is Suba Sitaraman. I'm the implementation lead on the modernization initiative at the Commonwealth of Pennsylvania. Um, so yeah, quickly go through the agenda real quick. So, um, it was an unattainable system. It was a green screen legacy system, um, really processing a lot of volume on a regular basis. So creating a lot of manual work for the agency. The agency that we deploy this application for is the Department of Human Services for Pennsylvania. Um, then we're going to get into how they use their organization to make the go live successful, and the overall mission is successful as well.

And then we'll dig into how we leveraged AI on a few different ways to accelerate the transformation. And then we'll go through some of the success that the Pennsylvania has realized, and then some of the lessons learned that, you know, we're going to look to take to our future projects. Okay. So before we get into too deep about exactly the project, I wanted to kind of ground everybody with just how big Pennsylvania is. It's one of the largest states with 13 million citizens. 3.5 of those citizens are served by DHS. Right. So it's a really it's a really large number. As you can see, they're getting 12,000 appeals a month and 135,000 a year. So that was all paper before we did this project to compare it to like a country size of like what Pennsylvania is, is comparable to Greece at 10.3 million people.

And so DHS mission really would align to the Ministry of Health and the Ministry of Social Cohesion in Greece as well. So this application that we built really had to scale and operate at a country size level. Yeah. So the mission of the Bureau of Hearings and Appeals, in short, was to adjudicate appeals and administer hearings in a timely and judicious manner in compliance with state and federal regulations. DHS, the Department of Human Services, as Carl mentioned, caters to about 3.5 million Pennsylvanians, and it provides social and public assistance programs such as Medicaid, Snap subsidized child care, nutritional benefits, nursing home related services, long term living, behavioral health, mental health services, autism, you name it. Everything under the gamut of social and public assistance programs are covered by DHS. So the 12,000 appeals that he receives. They span across all these major program areas. The appeals are either received from individuals or providers, because a benefit or service is either denied to them or reduced or discontinued.

So the program officers, the appellants basically send their appeals to the program officers and through them to BHP or the Bureau of Hearings and Appeals. So the system that was in existence before we implemented Pega was a legacy system called Manoj Pappu. It was a Unix based system. And believe it or not, it was built in the early 70s, literally 72, 73, more than 50 years old. So there were three failed attempts at modernizing that system and eventually its success with the Pega Platform. So moving on to the next slide. Yeah. Unmaintainable and manual legacy systems, obviously 50 year old system. So when we started the project, the. The very first challenge we faced was the one person who had technical knowledge on that mapper system was retiring in 30 days. His name was Bob Robert Martin. Obviously, it's a 50 year old application, so most people who had worked had already retired and left.

So Bob was probably surviving. Technical expert on the state, but he was also retiring in 30 days. So we had this huge challenge to get a brain dump from him. Get the institutional knowledge from him. And compounded by the fact is that documentation of such an old system was very limited. Even the maintenance manuals were from 83, 84, something like that. So that was such a big challenge that we had to scramble to see how best and strategize. How do we get the information out of Bob? A brain dump from him in such a short span of time and get this project moving. So the Legacy System Mapper is primarily used, like I said, by Bcha. So they they would get the paper appeals. Everything was paper. They would. Appellants would send their paper appeals through the program offices. They'd come to Bcha. They would clean all the data in the mapper system.

And then the hearings were scheduled outside the system, all the correspondence letters, whether it was asking a appellant to come to the hearing or even adjudicating a benefit. Everything was paper, literally paper. After the fact, Bcha would go into the mapper system and key in the results or the outcome, because ultimately reports had to be extracted for federal and state agencies. So mapper was, you know, essentially used as a as a data store. But over the years, it was also patched up to meet federal and state guidelines. You want to talk about the other. Yeah. And so super talked about it a little bit. This paper process was very, very cumbersome. And if you can imagine, like if you're on Snap or Medicaid and your benefits are reduced or removed incorrectly, this is the only process you had to get that going back. And so they would use colored folders like blue, green, and red to kind of organize them and their filing room.

And the client, as they describe it as this like vivid scene where there's just like paper and folders everywhere. And then after four months, they would send them off to off site storage in case they needed to be referenced again, right? So like really, really old school kind of filing economics to make this thing work. Um, and the, the core question we had as we faced 30 days to get everything from Bob and then make this, you know, capture everything and make sure that we could go live in time to, to serve the citizens. Um, and so one of the ways that they accomplished this was through the organizational change and teamwork championed by Dorie McGuire, the DHS CIO who's here with us today. Um, they put in a place of change engine, basically a product manager. They were dealing with 19 different program offices, right.

And each one has slightly different requirements and needs. And so in order to not, you know, make, uh, you know, try to go after perfection and kind of continue to refine the requirements. This change engine really led to the success of our deployment. They aligned the different program offices, got them to get to a decision and get live so they could realize the benefits of a digital system in place of the manual one. Um, they also really involve the caseworkers in their managers. We did frequent playbacks and demos where they got to see the application. We did a lot of exploratory testing where they got to go into the system and kind of different try different, uh, you know, workflows and different processes to see if they could break it. And so those two things together really got everybody comfortable with the system. And then we got the approval to go live in just 13 months.

Yeah. And as Carl mentioned, the 19 program offices, that was they are the biggest stakeholders where we needed alignment. Every single program office, Obviously, each office is dealing with a certain set of services benefits that they cater to long term living office, office of the developmental program. So there were 19 such offices. Every office would get the paper appeal. And one of the missions of the hearings and appeals application was to standardize appeals across intake, across all the program offices. So the best that we could. So when we analyze the appeal and the cover sheet, the paper form for all these 19 program offices, we realized that about 60 to 70% could be unified, standardized, and things like demographics, right? And then there were, of course, 30 to 40%, which were service specific, office specific. But our biggest challenge really was getting an alignment amongst all these 19 program offices and bureaus.

Everybody had their own set of priorities, processes, strong voices, strong opinions. And this is where leadership, DHS leadership, HHS, IT leadership, and the program CHAMPVA that they had appointed helped us navigate through this challenge. Obviously, it wasn't easy by any means, and we had to do a frequent playbacks with them, show them the application frequently. As Karl said, we ended up giving them the application ahead of time in test environments so they can play with it. Exploratory testing, get their feedback. We ultimately had to build a comfort level, so they are very familiar with the process. They don't have any concerns. We get their feedback and that certainly was not an easy task. But we, with everyone's support, strong collaboration and leadership, reiterating the mission that we have to go live, we must go live. And it helped. So we finally were able to go live within a period of 13 months.

Yeah. So the AI approach to modernization, the biggest accelerator that we had with the AI, was using GenAI for data migration. So Bcha has a ten year data retention policy. They have to retain ten years of data and hearing data. And this was 1.2 million appeals and about 700,000 hearings. So all these cases have to be had to be migrated from the mapper system into Pega. It was no easy feat. And this is where GenAI helped us accelerate. So we built a lot of GenAI tools in order to interrogate the database that mapper had behind the scenes query the front end screens and, you know, get the. And obviously, we were limited with the technical knowledge that we had about mapper. So we used GenAI a lot in order to gather this information. and we fed them. We extracted this into CSV files, created Pega utilities, and these CSV files were used to intake the cases into the new Pega Platform.

And ultimately we were able to migrate these cases and the hearing case types. So not an easy task because the other challenge we ran into was that the data, right? Ten year old data was there was a lot of challenges. There was not much of a validation. So instead of a name field, you would have address instead of address, you'd have something else. So the the data, especially in the older cases and older appeals, was all displaced. So we couldn't obviously, you know, clean up the data because we couldn't do a, we couldn't tamper with the production data. But we did use the rules to make sure that the data is kind of the GenAI tools to, in a way, you know, move the data to the right table, to the right elements, to the right fields and columns. So GenAI helped us. The tools that we used helped helped us in streamlining the files, and we were able to finally move and migrate the data.

We had the production users check it out in multiple times. Obviously, like any other data migration, it goes through rounds of testing. So we had them validate in pre-production environments so that they are satisfied that it is working through the new workflow. The data is accurate and we were able to ultimately accomplish it with most of the, you know, with really good data for most part. Yeah. And in addition to data migration, we use AI on the ECM program in a lot of other ways. Definitely supported by the DHS leadership, they gave us access to Open API 4.0, 4.0 mini. We built a lot of custom tools because we're only on Cloud two on Pega 24.1. So we don't have all the great innovation that you've been seeing about yesterday and today. So we did a little bit more manual approach from the GenAI perspective, but we built tools that reverse engineered the legacy code and also the legacy training documentation, so we could forward engineer that into process flows, data models, epics, acceptance criteria, everything that you would need to, to do a project, right?

We basically was leveraging AI every chance we could. And then we would also pair that with Blueprint. When we're doing rapid prototyping, you can take that reverse engineering and forward into a Bpmn file or a DLL, and then load that into Blueprint, and you get a really good example of what that legacy workflow would be. But in Pega very, very rapidly. So yeah, the leadership definitely helped get us the right tools so we could do some of those, um, AI approaches. Realizing success. One of the earliest success that we realized after we went live was there was a federal mandate and the mandate needed. It was called the Consolidated Appropriations Act required all the states to redetermine the Medicaid and chip enrollees, and DHS is serving 3.5 million enrollees. So not an easy task by any means. So prior to this, if they had gone with the prior application, it would have taken probably years before they could redetermine all these enrollees.

And the new application helped them accelerate what was what would really have been years was able to be done in less than a year. So that was one of the first biggest benefit that was realized with the new application. And the next thing that comes to mind is actually it is not a benefit, but it was a mitigation that happened earlier this year, January 8th, 2026, to be precise. I remember the date I realized that the physical letters that had to go out to the appellants, either because they were asked to come for a hearing or a decision was made about their benefit, whatever. Right. All the letters he sends out about 102 different types of letters for almost a month. The physical letters hadn't gone out. The application had triggered the letters. It was printed. It was stuffed. But somewhere in the mail room, it just never went out.

So all of this happened during the post, the Thanksgiving, during the Christmas, December holidays and early into January. And that's when they realized that, oh, no one's showing up for hearings. What's happening? And people were lawyers, attorneys, people were calling and they realized that all of these mail had not even gone out for almost 30 plus days, working days. So there was a huge scramble because this was critical, right? There could be other implications. So this is where they were able to recover from this fairly quickly. We were able to retrigger because we knew the duration for which the letters didn't go out, retrigger it from the application. So they gave us some business rules saying, okay, let's say if a hearing was supposed to be on December 20th, we'll give a lead time of six weeks, given that it's all in the past and we implemented a temporary business logic.

Ran those logics and reprinted all those letters, and finally were able to mail them out through our bay, was able to get it mailed out. So appellants could get the new hearing dates and whatever letters they were supposed to have received ended up going through. This entire recovery process took about two weeks, literally. So what real. I cannot imagine how long it would have taken them with the old school paper files to find out which letter, which appeal, what had not gone out and tried to do that. The modern system with the Pega capabilities helped them achieve this in a. The recovery was really, really faster, so that was a big success for them, at least from that perspective, that it didn't turn out into a much more or a bigger disaster than what. What was the challenge there? Yeah, I think both of those scenarios, I think would have been almost impossible to overcome with a manual system and messing with all the folders, especially the recertification, they'd have to go to that long term storage I was talking about and figure out which lots to get in order to determine who needs to be recertified.

But in addition to those two really incidents that occurred and how the system helped us get through it, there were a lot of other really tangible benefits a 98 drop, 98% drop in paper filings. Right. So they're all going through the front door. They're submitting them electronically. The paper has really turned into a trickle, and that has resulted in 86% drop in office supplies, stamps, envelopes, everything that you need. Those colored folders, of course. So all of that, they've got a financial savings and they've got a, you know, less manual work to do savings. In addition to that, their scheduling time has gotten 33% faster. So again, imagine you've lost your benefits and every seconds counting when you don't have that money coming in or that benefit coming in. And so getting that a third of the like one third faster is just a tremendous amount of time saved for them.

And we've also got 67% faster appeal submission. Um, We're improving the service to the citizens in lots of different ways. We can check the judge's schedule so we can get them their hearing booked quicker, and we have digital approach and workflows to the entire process. Now, when again, just a completely manual approach before would have just taken a very long time. Okay, so lessons learned. I'll hit a couple of them. One of the first things that I wanted to talk about is just we went, we went live. Don't delay. Don't go for perfection and stop what's good. It's that initial go live. We hadn't gone live when we did the two incidents that was happening. They would have been dealt with in the manual system. Right? And so again, we had 19 different program offices all trying to agree on what was the right approach and what is the right thing to deploy.

But actually going live was one of the really key lesson learned. Do not delay for for perfection. And then the next part of that is your go live is really just the beginning. Since we've gone live, we're on a monthly enhancement cadence. We've had every month. We've been enhancing the system. And that's when you really because those 19 program offices are really kind of trying to theorize what they think they want in production. But once they start using it on a day to day basis, then they can really articulate what are different ways to improve the system and enhance it moving forward. So yeah, make sure you go, go live, don't strive for too much perfection. And then that is really just the beginning because once you have it there, then you can continue to enhance and use new features. Yeah. And basically in conclusion, right, this experience taught us large scale transformation and modernization of the most legacy of systems.

I mean, this was a 50 year old system is achievable with the right approach, the right platform and the right set of AI tools. So with Pega assisted and AI assisted analysis, we were able to achieve in 13 months, including this whole data migration of ten years of data. So all of this just disproved that, you know, with despite three failed attempts, we were able to succeed with the Pega approach. And more importantly, it was not just replacing a legacy system. Bcha was able to simplify their processes. There was a lot of redundant steps they were able to eliminate as they started working with the new system, or working through the exploratory testing and giving us feedback. And the Blueprint that Carl was mentioning helped them visualize how the workflow would look. Oh, then they realized that we don't have to do this in the system, so it helped in easing out unnecessary steps.

Definitely wastage was a was a big reduction, paper wastage. Everything was paper. So that kind of was a big benefit from this. And last but not least, right. There's a human element to this. Think of it over the holidays people needing food stamps benefits were denied for let's say they were adjudicated in their favor. All the Snap benefits, food stamps, related benefits, those kind of adjudication that were to be on for them, and that was possible to be achieved. Operational efficiency, especially with social and public assistance programs, they take a different light altogether, and this application helped in improving the operational efficiency, helped to adjudicate faster, get the citizens, you know, the right decisions that they that was appropriate. And it finally led to a successful implementation. That I think if there are any questions, let us know. Thank you. Can you guys hear me?

Yeah. All right. My name is Rico Hightower. I'm with Synergy Computing. I have two questions for you. I think I know the answer to the first one, but I'll still ask, as you guys moved over the case types for the 19 program offices that you support. Was it a hard cutover from the legacy system to the new Pega system? It's the first question. And then the second question is with the data migration that you guys were doing, did you guys have a plan in place to, I guess, protect the the PII, the personal identification, identifying information for users? Or is there any lessons learned that you guys have to deal with during that migration? That was it. Yeah. It was a hard cut over. No, no, no phased approach or anything like that where people were working in both systems kind of doing swivel chair and assuming you want to take the migration question.

Yeah. Your question was whether the PII was protected. Yes. In the we used initially the test data, but ultimately mapper didn't have all those environments. So initially we mocked up some, you know, data. We asked actually the end users to mock up and give us some kind of a data. We gave them the template, they gave us the data, but ultimately we had to do it in pre-production environment only because they did not have all this, you know, lower environments in that mapper system. But we did. It was very secure. Only people who had obviously, because of the PII, Phi, all of that, it was really restricted access. Only people with need to know a very limited set of folks, even in even in Accenture side, very few people had access. So we would just get the feedback from them, what worked, what did not work, and we would have to adjust accordingly.

So it was certainly a challenge given that we didn't have lower environments to test out multiple times. It really was done in the pre-production. Yeah, they only we only put the real data in pre-production and production. That does kind of limit your testing of all the different scenarios in the lower environments. You'll find a little bit more in Pre-prod. But yeah, that's it. But it is to protect, especially with these kind of sensitive services. I think it's really important. Other questions. Hi. Good afternoon. My name is Arnaldo Garcia. I'm with Accenture Federal Services out of Washington, D.C. and my question to the group here was about onboarding. So after you built the new system, the legacy system, what was your onboarding like for your users. Onboarding for the users? So prior to the, obviously a couple of months before the go live, we, we got a list of all the users, both from program offices and from Bcha.

And we created a template for them to, to be able to tell us what their role is and so that we can assign the right access groups. We created Pega utilities, automated utilities, so that we can quickly onboard. In fact, those utilities are still being used because every, every day we get some new people who need access. And then it's not just 1 or 2. It's like a really a, because there is attrition and there are new people joining in or people move departments. So we do have these utilities built so we could automatically create those, you know, access in Pega again. The other aspect is the authentication happens on the state side. So we had to work with that system as well. And it's, um, it's a long run process to get that, but we were able to streamline on the Pega end and then be able to develop the process to be able to work with the state's IAM system.

Thank you. We were going to train the trainer kind of capacity as far as like the change management aspect of onboarding all the new users. So we were responsible for creating the initial training material, ramping up the trainers that did the actual training to all the, all the users as well. So that approach worked, worked nicely. All right. Well, thank you all for your time. Yeah, I appreciate you guys attending and listening to the story.