Adam Parks (00:05)
Welcome to Applying AI, the show where we cut through the hype and get real about using artificial intelligence in regulated industries. I'm Adam Parks, joined by my season one co-host, Mike Walsh from EXL. Each month we'll bring real world stories and strategies from leaders applying AI to streamline operations and boost the customer experience, all while staying compliant. No jargon, no science fiction, just practical insights you can put to work today.

Season one of Applying AI is presented by Receivables Info and sponsored by EXL Let's dive in. So Mike, I know I've had the pleasure of knowing you for many, probably too many years at this point, but for anyone who has not been as lucky as me to get to know you, could you tell everyone a little about yourself?

Mike Walsh (00:47)
Yeah, sure. I got in the collection business in 1996, right out of college. Worked for traditional collection agencies, BPL. I worked at VoApps I managed sales at TrueAccord for a little bit. Got into Everchain where I did some sales over there as well. So I've done the gamut in this business from traditional collections in terms of selling, setting up programs, client success. And now on the AI side where we're here at EXL, we have some great technology and it can be used by creditors, debt collection agencies, as well as debt buyers as well. it's been a fun journey and it's been a long one. I'm a dinosaur in this business, everybody is, right? Once you get in, you can't get out. It's like the mob.

Adam Parks (01:35)
Well, it's a lot of technology that you've been through between the traditional sales, working on technology related projects, then working with the first, what's called digital first agency that was out there. Working for a tech enabled debt broker. And then obviously ending with EXL here where you're doing all kinds of different tools related to debt collection as it relates to artificial intelligence and being able to deploy these things. You know, for me, it was a little bit different. I came in as a debt buyer on a technology enabled organization. We were a debt buyer, we became a broker and done compliance related tools and other things that are related to the technology pieces. But I've only really dug into the artificial intelligence pieces over the last couple of years. And I think that's a great way, Mike, for us to transition into talking about the state of artificial intelligence in the debt collection industry because I just want to point out as the author of the TransUnion Debt Collection Industry Report for the last few years, three years ago, we had more than 60% of debt collection companies saying that they would never touch artificial intelligence. Last year, was 24%, which I thought was a dramatic shift. And now we're down to 7% in the 2025 survey results. That's a pretty big shift between saying I'm never going to use something to actively engaging in trying to deploy that type of technology. I mean, it's been really condensed. You've been with EXL for about that same time period now. What are you seeing in terms of the current state of the union related to AI in the debt collection space?

Mike Walsh (03:15)
I think those numbers are a hundred percent accurate. I really do. Cause I mean, and I think the ones holding out are probably commercial, although we have a product that can work with commercial. AI can help commercial agencies as well, but I get it there. It's not as, you know, ⁓ volume driven. It's not as repetitive. It's not as, it's not as clear a use case. So, But I think you're right. The switch started, I saw the switch at the beginning of 2024 happening. And then I think what also happened in 2024, 2025 is the products out there just got way better. Technology is way better. It's incredible. I've been at EXL for two and a half years. We don't use our old, when I started and we started talking, the demo I gave you is no longer.

Right. Like it's two years, you know, um, and the, new product is, you know, the agentic AI virtual agent composed compared to like what I think a lot of people call generative because they use an LLM is the only generative part of it, but is the templated base, um, AI you find out there is it's two different worlds, right? Um, it's a true conversation verse. You said this, I, this keyword was understood this utterance. I'm going to give you this template.

So just for that fact, know, debt collections, it's a, if the math works, this industry will find it. If the math and compliance works. This industry will find a way to use it. And, you know, nobody wants to be the first to jump in, but people have jumped in. It's been out and around.

Adam Parks (04:51)
Well, over that time period, think that we've more finitely defined the use cases of artificial intelligence. So now it's not just this theory, it's how can we actually apply the artificial intelligence to our day-to-day operations? And when we think about the use cases and how they've changed over time, the first companies that came to our space that used artificial intelligence, and I'm using air quotes when I say it, were really focused on scoring.

It was all about using machine learning to improve predictability and scoring into the future. And then it started to evolve. Then it started to evolve into content creation, content context, and it started moving down that path. But now we've moved into these more refined use cases. And as those use cases have been deployed by different organizations around the space, the comfort level has increased. Because when a creditor has had success with agency A, then when agency B, C, D and E want to do something similar, the fear factor is lessened and it's more likely that they're going to be able to go forward.

Mike Walsh (05:56)
Yeah. And it's weird. is the, like, I think creditors are leading this time. I think just because the nature of business, right? Like let's take the banks, for example. They see so many use cases, right? They have so much data. They have such effective, you know, possibilities, you know, like to use AI within debt collections and out of it that I think they've jumped first, which was not what I anticipated completely, but I think they're kind of leading. Like we've had more banking customers, for example, in the US than collection agencies, but that is changing quickly.

Adam Parks (06:34)
Well, that's interesting, Mike, because one of the things that I saw when we broke down the debt collection industry survey this year for TU was that creditors were very likely to be using artificial intelligence, but they were the most likely to be using a third party service versus trying to develop something internally. And I know when you and I were at RMAI, we talked briefly about Why that might be? Why might the creditors be so focused on these third party tools? And I think it's because they're realizing that their core business is lending, not collecting and not building art. They're not a technology company. They're a financial institution. so leveraging those third party services that have already demonstrated some success is the least common denominator to get there. And I had done a webinar with somebody from a bank back in the end of 2025 where they were talking with me about the challenges they had deploying any new technology, how many different departments they had to get resources from and all the things that had to happen in order for them to have a successful deployment. And while it's a lot shorter of a path for them to do that when they can plug something in versus having to try and develop that. And then me, you and Tim Collins did a session back in December of 2024 with the buy it or build it discussion. Do we buy artificial intelligence technology or are we going to build it? And if we're going to build it, what kind of responsibilities long-term does that ultimately entail?

Mike Walsh (08:04)
Maintenance, right? Like maintenance updates, you know, and I think Tim and all of us kind of agreed. And that, I think too, is because they had multiple use cases. I think some banks built some of it, right? But they were smart enough to say, can't build all this, you know, we can't. ⁓ There's only so much. And even the agencies or debt buyers I talked to, I'm like,

Adam Parks (08:06)
You're in. No, it's a big lift.

Mike Walsh (08:26)
and they're like, hey, we're thinking about building yourself. And I always say, try it while you build it. Because, it's a lot and you're going to have, who's going to do it. Are you going to hire new, who's going to be responsible for it? You know, there's implementing design, implementing, maintaining, and then getting the most out of the tool, right? Like data science or data analysts that can use this to optimize your current process collection processes and then change the ones that are outdated. You're not going to, you know, you put in AI, you're not going to wake up and make, you know, 30,000 phone calls that day. It just doesn't make sense, right? So things are going to change. Like the industry is changing and it's changing, I think for the better quickly, the technology is also changing.

And you can do, you can use more AI tools to do more aspects and make them more efficient, whether it's talking to, to customers or, you know, doing your books, you know, doing your accounting, doing your payment process. There's so many, you know, and you've covered the use case groups. And then when you dig into them, there's so much, you know, audits, everything, ⁓ RFPs.

Adam Parks (09:38)
Well, speaking of the use cases, I think it's important that we address the six use cases for AI and debt collection. And I think even in this conversation, I may even add a seventh that I've been considering for a period now. But the first one being data enrichment and segmentation. So actually being able to use artificial intelligence and machine learning to understand the data sets, to be able to reach into other data bring it in, it, improve our overall data hygiene and improve our ability to collect because we have better data than we had the day before. The next one being consumer communication and engagement, which I think is a major factor, right? In terms of being able to make the outbound calls and that could come in a couple of different flavors. Workflow automation and efficiency, compliance monitoring and quality control.

Predictive analytics and scoring models, which I think is where we all started and training and agent enablement. So talking in the agent's ear, demonstrating something like a Zenerate where you have the, the new collectors through a training program, calling in, going through these example calls and, and those kinds of things. going back to the beginning here, Mike, I thought it would be interesting for us to kind of talk our way through some of these in terms of what we've seen and kind of where it fits into the spectrum of the debt collection world and within how much of this is going to actually make it to the technology stacks of debt collection companies. But data enrichment and segmentation, I think, is something that we started with early on and is mission critical because it's a somewhat simplistic process that can benefit from some automation. But at the same time, it's something that we're doing a lot of manual work with today that maybe we don't need to.

Mike Walsh (11:27)
I agree with you. I also think we're looking at it like the way we're going to look at the data, just take a collection agency, for example, the way they're going to look at their data is going to change, right? With tools, because if you think about how many transactions a collection agency is just pulling a day, right? Like just that there's a tremendous amount of data and then how many payment plans they're running every day and how many break, how many continue, how many get readjusted. There's a need there to look at that more with AI, for example, and get those trends and get, we're pushing these people to a hundred dollars when they can only pay 75 and it's hurting us in the long run because they break and then they're going to go pay someone else. Right, like there's tools like there's a way we're going to have to look at this that I think is changing. And what AI can do for you is graph that out, make it very simple to understand. Your team can use that data because collection agencies have a ton of data. People always say, hey, my data is not ready yet. It's ready. This is my number one. You've been ready for this for 20 years. I mean, this industry, because if you're not, you're getting sued, you know, or you're out of business, right?

Adam Parks (12:30)
Yeah, it was ready yesterday.

Mike Walsh (12:41)
Like we are heavily regulated and that data has to be clean. f you, know, Most agencies I worked at, if the data was wrong, we sent that account back. Like if it didn't have all the fields.

Adam Parks (12:50)
Well, the data hygiene, I think is an important piece here. And so when it comes to artificial intelligence, the quality of the data will either amplify the solution or the problem. And if your data is bad, it's going to amplify the problem. And that is an issue that we need to address as an industry. I think the ability to leverage automation, go ahead.

Mike Walsh (13:11)
I think our account data is crystal clear, clean, right? I think the data, maybe our reporting data needs work just to optimize it. think some of the back end stuff like our audit data that I don't think we've optimized that yet, but there's tools out there that do it, right? Like, so I think that's coming along. I think that is been an early use case that's proven.

QA audits were there, but I think our data as an industry is very good. And I've, you know, I've done implementations with seven data sources. It's there somewhere. It's just where you're pulling it from. And once you pull it, you're fine. It's just setting it up to pull it. And you know, once you do it, the headaches over and you get through it. And it's generally not as hard as people think.

Adam Parks (13:57)
Human readable and machine readable are two different solutions. And so I think just because it's in seven different locations doesn't mean that we can't bring it together to be understood by the machine and carry context, but I think that's the differentiator in our minds is like, well, I can't see it the way that I want to see it. So it's not ready, but not thinking about the fact that the machine does not look at it through the same lens that I look at it through as a human.

Mike Walsh (14:00)
Correct. That's a very good point. Right. machine pulls it. Yeah, machine pulls it into one place and then spits it out for you. So it's easy, right? Like, you're right. That is that is a very good point. I'm stealing that by the way.

Adam Parks (14:31)
Well, I know we're going to be having some more conversations about this as well as we go into some of the other episodes that we have planned for this particular series. But use case number two, consumer communication and engagement. I feel like this is the world that you live in. Like you live and die in particular space. Talk to me about how that has changed and what you think the industry is looking for in its next step.

Mike Walsh (14:45)
Thank So in two and a half years at EXL, this has changed for the better. And two way SMS, chat bots, virtual agent, it's all improved. Virtual agents made a huge leap and is much better. Now, I think now we're looking as not as an industry, is this going to work? Is it real? We're looking at, show me a case study. How do you perform? And that's the difference from when the day I started to now. And I think what's great about that is More people are open, more people are listening, more people are exploring it. I always tell people, if you don't like it, don't use it, but take the demo. Listen to this world because we as consumers are getting more used to talking to AI.

The technology has, as of May, as of February last year, has totally changed. And it's changed for the way better. The use cases are more and more. The collections are proven now. It's a cool time to be in this industry because you can do a lot with the same amount of people. You can cover, I mean, to me, it's about scalability of who you can contact.

Like phone calls are now not only for 250 and above, right? Like they're for 50 and above or 25 and above. Like it's going to change how our contact rates is going to change the scalability. And, and that's exciting because, um, there's so much debt out there right now. You know it more than anybody. It's just incredible. The TransUnion report is frightening. Yeah. I mean,

Adam Parks (16:11)
the volumes are just going to continue to rise.

Mike Walsh (16:29)
It's just crazy. So everybody's been so busy. Hiring's been a challenge. I think this is gonna help. The technology's gonna help a lot, lot.

Adam Parks (16:38)
Hiring has been an impossibility for most organizations. So being able to deploy technology that allows you to still maintain the high level of customer service for the consumer without putting them into the IVR death loop. Because no matter how much time and energy you spend building out your IVR, you're still going to have individuals that end up in that.

Mike Walsh (16:57)
I'll give you an RMAI example I did in the hallway right in front of salt, right? Like, he goes, this guy said, I'll never use it. We believe in customer service. And I said, listen to it. I've known this guy for years and he, his, was an incredibly smart test. He's like, yeah, let me go get my wallet. That's what he said to the machine. Like right away after it's a fake name, we identify, he's like, can you make a payment? He's like, yeah, I can, but let me give my wallet. And he's like, This is, he covers it and goes, this is usually where, you know, I'm out and it breaks. But it said, sure, take your time. You know, like that is, so with all the noise, it's not quiet there. All background. It's a mess, right? And that's the sort of little things that have changed where, hold on while I get my wallet. You know, that type of thing. And it templated, it's like, there's no template for that, you know? So.

Adam Parks (17:33)
No. Yeah. No, there's too many potential possibilities and responses for you to template to that level. There are some things that you can start those engagements, but there's limitations. Yeah.

Mike Walsh (17:57)
Works great in text. Yeah, works great in text, but you have to go true agentics in voice, you just do.

Adam Parks (18:04)
Which I think is a great tie into the next use case being workflow, automation, and efficiency. Because now we're starting to think about the workflows between it. If we've got the consumer contact pieces, we've got different pieces of our business, and we've appended data so that we're actually fueling the machine with good gasoline, now being able to move things throughout our processes should be easier, more accurate, and more efficient over time. And it's only going to get better as it continues to learn how those particular consumers and portfolios are behaving.

Mike Walsh (18:39)
I think this is the funnest part. Like this is my favorite part of the job is like, you can go technical and geek out, but like how it applies to it's a, like all AI is when it's to the consumer is a engagement tool. And then now we're talking about how collection professionals, people have been doing this for a long time, can quickly get information that they can use to improve the process or test

Mike Walsh (19:04)
different processes or test different. Like this is where you can say, okay, they've looked a lot and they haven't, they've been on our portal looking and they haven't made a payment. They keep trying. What's, what's the problem? Are they waiting for pit? You know, and that leads to questions, right? Like, and you're very good at this. out. Like, are they just waiting for payday and they just want to make sure the offer is good? Are they not finding the offer they like? This is that whole e-commerce you discussed that, ⁓

Mike Walsh (19:31)
Where was that? DCS or CRS? Yeah. Yeah. Right. And now you're like, okay, how do we make this easier? How do we make, how do we make these offers, you know, workable? it, Do they just need a plan for now until they get a job or pay off, you know, a kid's broken leg or something like that with them? There's all these challenges that I think

Adam Parks (19:32)
All the time. Yeah. All of the conferences I talk about e-commerce and its application. We're an e-commerce business in 10 years, whether or not we want to admit it.

Mike Walsh (19:57)
these tools will give and the, and the agencies themselves or creditors, collection experts, debt buyers, you name it. Those people who are in the trenches will take that information and make it even more efficient. We have people coming back with use cases. Hey, we want to do this. It's, it's redundant. It's a waste of time. Can you do this? And then we just build out the customer journey. it's so that's, I think the funnest part of this technology is, your people are gonna love it and love the tool to enhance their own people. I always say, You're driving more calls to your inbound calls. And I've been in this business since 96. Inbound calls gold, human to human is still gold. That's not a bad thing. Driving more of it and doing less wasteful outbound is what you're trying to accomplish. And making those workflows just sing is what I think we're all gonna be doing in the next couple of years.

Adam Parks (20:51)
Understanding the context of the consumer engagement and then driving it down the appropriate workflow, especially when it comes to highly sensitive specialty accounts, bankrupt, deceased, cease and desist, all of those things and understanding that context and being able to move it down the correct workflow, I think is a value add across the board. It's just another opportunity for us to take the lift off of the limited humans that we're able to hire

Mike Walsh (21:02)
Correct.

Adam Parks (21:20)
and still be able to provide the highest levels of customer service for the consumer, which I think is everybody's goal and ties really well into our next use case being compliance monitoring and quality control. This is something we've talked about for years. There's been, I think this is one of the earlier use cases that we saw well adopted across the debt collection industry because it's not engaging with the consumer. It's allowing us to amplify our compliance coverage across a wider subset of opportunity. Meaning, five years ago, we listened to 5% of the calls with live people and we addressed what we heard within that 5%, but that sample size was so small. Now, the same limited resource that we have to listen to calls, the same limited volume of live people that we have are now able to focus on those exceptions that are identified by the monitoring tools and we're able to hyper focus our efforts on those that actually require human intervention, training, etc. And I think that's why that one has become so popular so quickly. But I think this is something that many, many, many organizations are using across the debt collection space at this time.

Mike Walsh (22:37)
Yeah. And I think the change there too is now the coaching is live. The auditing is live, right? Like it's not looking at your tapes anymore. You're doing this while someone's on the phone talking to a customer. So the little pop-up says Mini Miranda, you know, all those disclaimers, all the check, the check, like that easy check that you had people doing the checklist, that's all done on the call. Right. Like, and then you're right. Those exceptions kick out those bizarre cases

Mike Walsh (23:05)
where the person's flowers, if it's a utility account, were run over and they're not paying until you come fix their flowers. Like I've heard that call. Like those are the ones that are now kicking to the human being. But the coverage, I think it's peace of mind too for not only creditors or clients, but you as a agency owner, you can identify sentiment now. Like if someone's loud and getting angry

Mike Walsh (23:29)
and it's your collector or a call in, you can tell. Like you could hop that up onto your manager saying, and have them walk over there with the portable line and say, take over the call. Like there's so much. So again, these are tools,

Adam Parks (23:45)
Proactive management, it's proactive management of the collectors on the phone, of the live people that you're gonna have. I think you can divide off some of your incoming or some of your phone traffic to be managed by a bot, but those that are still being managed by a human, those exceptions, et cetera, I think you still wanna have some monitoring on that and be able to dig deeper there. As our account volumes increase, we have to find new ways to achieve the same level of quality with a larger volume of calls and no additional revenue. So we have to find this balance between those two worlds, which, you know, and I'm going to tie in one of the other use cases here because I think it's similar, but not the same, which is the training and agent enablement. It's that ringing in the ear of that agent while they're still on the phone versus that post-call training that comes from the compliance monitoring and quality control. This is more about engaging them in real time to help feed them the information that they need to be successful on the call. And at the same time, do some of these training things where you can call in and have a conversation with the quote unquote, you know, consumer, but now the machine is able to throw some different tricks and ideas at you to help really prepare those collectors to be talking to live consumers on

Mike Walsh (25:11)
I totally agree. And like, are we typing during this podcast? Like that job is a, not an easy job being a collector, right? So like now you don't have to type. You can just focus on the customer, helping them solve their problem. And that's what these tools can do. It's amazing. And then yeah, you have one minute to wrap up just making sure the AI summary is perfect. Cleanup adds, maybe add a next step. The next step should be there, but maybe you don't agree with it and changing it.

Mike Walsh (25:39)
So I think the expertise will still be needed. I also love the idea of two, like, you know, with all these state laws changing constantly, can, you know, the AI training, you update it and here's here's Colorado's new law, like boom. And just, you could train that in like five minutes. Like you don't have to bring a whole class in. There's so much you can do. Some people can. Yeah. Correct.

Adam Parks (26:00)
You don't have to burn as much time off the floor. You're able to actively engage with these folks, you know, from their chair, save yourself a whole bunch of, mean, every time that you're moving everybody in and out of a meeting, I think we've all calculated the true cost of meetings in our organizations. I, God, I saw something on Instagram. My flight back from RMAI, was like, you know, what's your job? Email, meeting. That's my whole job. Email, meeting. Which I thought was, you know, it's funny, but also horrible at the same time.

Mike Walsh (26:24)
There are great needs to underrated. Yeah, they're phenomenal. But yeah, I mean, you save time. You keep those people working in and productive. And you also keep them updated for all these, the CFPBs now. I think you said it or somebody had already. It's now every state you have to deal with.

Adam Parks (26:45)
Yeah, it's amplified, right? Like we've multiplied the problems that used to be centralized with the federal government. Now we have all these states going together, which, you know, thinking about the last use case here as it relates specifically to the debt collection industry was predictive analytics and scoring models. Now in the survey results for 2025, I could see clearly that debt buyers were the most likely to be applying artificial intelligence on this particular use case. Which I thought was interesting because I believe that they see it as an opportunity to create a competitive advantage through their scoring methodologies and their behavioral analytics. So I think that's interesting, but definitely something that we're gonna see more of. I think it plays into prioritizing accounts and workflows, but it is a different use case altogether.

Mike Walsh (27:32)
Yeah. And it's protecting their investment, right? Like, and I believe creditors have been doing this for a while. Like I can tell you, like, EXL has a whole modeling data analytics. We've done this with tons of creditors, across different verticals and they've been doing it for a while. They've been using, I think it's coming to debt buyers also have been using it. I think for, for a while, I think. Agencies are starting to say

Adam Parks (27:35)
I don't know.

Mike Walsh (27:59)
let's model this out and see what we should be at. Even the scoring models out there, they're getting better in terms of what this should be. But now I think the biggest difference is right.

Adam Parks (28:11)
Well, I think the predictive nature has improved over time. We've seen the ability of the tool set improve its use case over time. And it's definitely something that I think continues to learn because we're feeding it more structured data and it gets a better understanding of where to go with it. And then there's a whole list of use cases beyond the debt collection industry.