Why Collection Data Silos Are the Biggest Barrier to AI Success | Ep. 5

Why are data silos in collections still preventing organizations from getting value from AI?

In Episode 5 of Applying AI, Adam Parks sits down with Daniel Yakimenko and Mike Walsh from EXL to discuss how behavioral data for debt collection is often trapped across disconnected systems, limiting real-time collection decisioning and performance.

Together, they explore how leading agencies are leveraging AI-driven collection strategies and digital engagement tracking to make smarter collection decisions. The conversation also highlights how unstructured data can reveal hidden consumer signals that improve engagement and recovery performance.

If you're looking for best practices for collections data management and practical ways to improve collection outcomes with AI, this conversation is for you.

Listen now and discover what your data might be trying to tell you.

Applying AI Podcast:
https://receivablesinfo.com/applying-ai/signals-across-data-silos-walsh-yakimenko

EXL Website:
https://www.exlservice.com/

Daniel Yakimenko LinkedIn:
https://www.linkedin.com/in/daniel-yakimenko-343b3b16/

Mike Walsh LinkedIn:
https://www.linkedin.com/in/mike-walsh-b88b271/

data silos in collections,
behavioral data for debt collection,
AI-driven collection strategy,
real-time collection decisioning,
unstructured data in receivables management,
collection data management,
digital engagement tracking,
debt collection technology,
collections analytics,
consumer engagement strategy

#DataSilosInCollections #DanielYakimenko #AICollectionStrategy

Transcript

SPEAKER_01 0:05

Hello, everybody. Adam Parks here with my co-host, Mr. Mike Walsh, joining us from EXL on the Applying AI podcast. Today I think we're going to have a really interesting discussion. Dan Yakamenko from EXL is also joining us today to talk to us about breaking down the data silos and activating data with all of this great technology. So with all the tech that we have, if we can't get to our data and we're dealing with siloed systems, we always felt like we were behind the eight ball and never going to be able to catch up. But now with some of the AI application tools that are built today, being able to bring that data together into a centralized location and then activate that data to improve your collections is something that's very real. Dan, very excited to have you on here today. Mike, thank you so much for always being here to co-host and ask the right questions. I really appreciate you guys. Dan, for anyone who has not had the opportunity to meet you before, could you give everyone a short backstory? Tell us about yourself and how you got to the seat that you're in today.

SPEAKER_00 1:10

Sure. Thanks for having me, firstly. And uh yeah, so I started as an um analyst back in the day, more than 20 years ago, was helping large banks, including Barclays, OTP Bank, which is the largest bank of Hungary, over their analytics and risk management and collections. After that, I was managing and leading collection agencies in Europe, mainly in the Eastern East Europe, but not only. So it was like Central and East Europe. It was deputy CEO of some largest European agencies that were sold later. And then I moved to the United States. And yeah, since then I've been helping large organizations, large banks mainly, with their collections analytics, only collections and risk management any longer, and uh was uh basic analytics strategy modeling, data management inclusive. And in 2017, we designed this solution called EXL Inventor, which I'm a HP co-owner and co-co-head of solution. Yeah, since 2018 it's in the market, trademarked name, and yeah, that's what we are currently doing in EXL. Deploying and helping companies across the board, across the whole collection cycle, uh get benefits of applying AI in their collection cycle with Pigmentar.

SPEAKER_01 2:27

I mean, that that sounds like you've had a lot of experience from a data and analytics perspective and understanding from the bank's point of view what they have to deal with. And if there's one thing I've learned in my years consulting with banks, is that their data tends not to be in the same place. And that fragmentation of documentation and data is something that they're frequently dealing with, similar to the debt collection space, where the agencies and some debt buyers are on more, let's call them legacy platforms that were not built for the current technology in the activation of that data. But you guys spend a lot of time in the space. I guess, Mike, starting with you, could you talk to me a little about what you're seeing in the market as you're working with clients from a fragmentation of data? I think it's everywhere, right? Like I we just we said banks and agencies and debt buyers, but it's you know, it's it's any company. We we even, I mean, maybe not fintechs as much, but still everybody has data they'd love to bring in to their collection strategy, and maybe they just can't access it for that need specifically, right? So I I think I I've always said this meme, you always talk about it is there's always a lot of fear around the data people have. And I always say, if you're collecting accounts, you have all the data you need to be using AI. And we've had clients where we pulled from seven different systems, literally, and just centralized it, used it together. I don't know how we did it, Dan did it, not me, but he figured it out. But but but I think that's a that's a key point. Like I love to talk to the market about is if you have it siloed, people can bring it together for you, right? Like they're there but that was a that was a big issue a few years ago. So if you go back three, I mean, even today, if you when you think about data management from silos, you think about having to rebuild or migrate systems of record, which is a 12 to 36 month project, right? That's a big undertaking. And I don't think that organizations were really there that long ago, right? The technology wasn't there to be able to bring it together and map it. So going over to you, Dan, how do you activate that? Like, how do you take that need in the marketplace, go into an organization, connect to all these different systems? I mean, that's a that's a big undertaking when you think about it from the old school perspective. But now that we're leveraging AI tools to do the work needed to feed the AI tools, which is scary in and of itself. What does that look like as a real project manager or leader going into that type of solution?

SPEAKER_00 5:10

Sure, and you're totally right. Interestingly, there is a belief that the larger the organization is, the more advanced the technology is, and actually the easier and access more accessible their data is. It's completely the opposite, usually. The larger the company is, the more dated their solutions are, and the data is usually siloed completely on different servers, and some data is not even stored at times. I'll give you a real life example with one of the largest global banks having a huge representation here in the US. When I started helping them with their modeling, I realized quickly that their dialing data is not even accessible. Like you cannot use any promise-to-pay. How can you just think of it? You're building a collection model that predicts payment amount or pay-no-pay model, and you don't even know what the customer said during the conversation with the agent. You don't know if there was a promise to pay made by the customer, any broken promise, and all that. All you have is just payment information and bureau data and some project like other, you know, uh uh external data points which they buy, but internally they're storing the data on a standalone server and never used it. So it took me a it was a serious effort for me to help them get that data from that server, just physically get access to that server, map it to the pool of variables that is uh accessible as predictors for modeling. So, yeah, and like I'm telling you, it's a company that spends tens of millions of dollars monthly on their uh solutions and data management and licenses and whatnot. So, yeah, there is a huge problem today in the market. And I guess first thing is so you're right, AI helps with structuring that data and using it, making some actionable decisions based on that. The problem is you still need to get access to the data. That's number one. So you need to have good understanding of what data types of data you need for modeling. That's where experience kicks in. So you need to have some experience on what worked before, what didn't work in your modeling and decisioning. Which includes, I'll just name a few, I'm not gonna go deeper into it, which includes Tyler data, which includes payment information, which includes deliverability funnel. That's very important, that's what's always underestimated, but it's actually not to be. Um what I mean is like most of the companies today use digital or remote channels to collect SMS emailing. You need to know exactly what happens, whom you sent, at what time, what was the reaction? Click, open, delivered, inboxed, spammed, all that you need to know. And if you think of it, it's like a funnel. So you send so many. Out of those sent, so many will be inboxed versus spammed. So many will be open, so many of opened will be clicked, and then login, and then payment and/or something else, enrollment and the plan. So all that needs to be tracked. So that's the start. We need to give access to that data to AI. And that information is always available.

SPEAKER_01 8:04

I'm gonna interrupt real real quick here, like I always do, G. But like one of the things I think is a good first step is like what do you need and what what don't you need? What just creates noise and extra fields that maybe you think you need, because maybe traditionally we looked at that, but really isn't important once you're using AI because the AI is going to pick it up from something else. Like, I think that's a big key to what we deal with. Like, I think, oh, I need all these different fields. I I mean, I think that's where I think the market worries about their data or is surprised when me and you go talk to them and say, oh, we only need like 15 fields. And they're like, what? Right? Like that's I think the data's been their core for such a long time, right? Like these organizations are built on the core of that information. And so there it's been we have historically been in a situation where more is better because it gives us more perspectives and more point of views, but now we're at a different point in the math hierarchy, I guess, where now it's better to predict from a small amount of variables because you're gonna fall within a smaller standard deviation on those predictions, right? Like you're going to make more accurate predictions on less data because you can better see and understand correlations. Right. And and then you have like the bias, you want to remove all those data points, and then like and then to what Dan and I do is we get the behavior data, which is much more predictable, right? Like, so that becomes more valuable than some old, you know, 10-month-old bureau data that who knows if they even have that job or something like that. So uh that that's kind of I mean, I think starting with the outcome of what you're trying to do and then just getting the key data points for that, not everything you have, everything where it is. And I talked about we had a client with seven different systems. You know, that's the most ridiculous example we've run into, right, Dan, that I can remember. Uh but they had some really cool, like they had built some things that were pretty cool, like offer engines and things like that. So you could so it made sense, right? But typically, I think in collections we're talking about some sort of data for digital, some sort of data for a dialer, a CRM, which is like the host, and then payment data, right? And then everything else. Which I would like to think is in the CRM, right? Like I I would like to think that a lot of what you're requesting is in one system.

SPEAKER_00 10:42

Ideally, it needs to be a single customer view. So CRM, by definition, it needs to be a single customer view. When everything merges there and stays there, source of truth. And everything is interconnected if it's a SQL relation database, it needs to be merged, and all the data points are all stored in one of the tables of the schemas, which are all interconnected by the account ID or customer ID or transaction ID. So you can always merge it and get anything out of this data. So, yes, you're right. And the funniest part of the example Mike made about the seven data sources, it's a professional collector, not just some lender or healthcare provider. So that's the funniest part. That even professional collection service providers can face those issues. And it would be funny if it was not true because it's not the only example, I'm sure. And I've seen worse, and it's not it's not a professional collector. But I believe that uh see my mic is making an example of we do not need that. So in our case, we only need 15 fields with payment or but Excel paymentor was designed so that it generates all the required information on its own so that we only need to know who we're collecting from. We just need the dialing list as an input with the first and last name, phone number and email, and the rest will the solution will generate, collect, merge into the right view, and share it with the client in a very structured manner so that the client doesn't need to worry about it any longer. How to store, where to store. We'll just create the whole schema structure for them and upload it to their system so that all they need to do is just depend the daily files from us. Or if it's an API-based integration, it will just be a few.

SPEAKER_01 12:19

I think that's a good point, Dan. Like some of the things we run into, which is I think interesting, is we generate data that companies cannot store. Right? Like they don't have fields for, right? Like so if I get behavioral data and it's usable, right, and it's trackable, a lot of times they have to create a field to store it, or we just store it for them, right? And then they have to access it. So I think going forward, flexibility to add fields to your CRM to use data that AI providers, all sorts of solution providers, you know, no matter not not just on the front-end customer, but back end. I think scoring models are gonna change. I think skip tracing models are gonna change. Like, I mean, just look at some of the cool stuff. I always forget the name of the vendor who's you know pulling license plates all over the all over the world. Yeah, yeah, those guys are amazing. I was blown away by that product. Um, and you think about like, you know, if you're an auto provider, you want to take that in, right? You want to take that data in. And you know, so I think that's gonna be where CRMs go. Dan, do you agree with that? Like more flexibility.

SPEAKER_00 13:32

Absolutely. There are still some CRMs. I only know one example so far, but they're massive still, where the their clients who use their product cannot access their own data. So technical that we can do. I know that I know that one.

SPEAKER_02 13:48

I'm familiar with that problem.

unknown 13:50

Yeah.

SPEAKER_02 13:50

So it's always a way in and out.

SPEAKER_00 13:53

Yeah. If you ask for one insight or advice for companies that have this problem, change the CRM. So you need to have access to your own data. That's where it starts. If you have access to your own data, you think you don't need it, you actually do. Because the next advice I would give anyone is think of your decisioning process, of your current collections process. Do are all the decisions made based on the data that is generated automatically, or there are some rules for like common logic-based decisions or experience-based decisions. Because our brain is very faulty. It cannot hold all the details, it cannot take into account all the peculiarities of each and every account, situation, stage in the journey, or detail or payment or whatnot. Too much data. It needs to be done based on the data automatically by the decisioning, right? By the analytical platform or model, ideally model. And if you do not have access to the data, you can never be able to do that. So that's where it starts. And then all the decisions, especially the big ones, have to be run by the models and some uh data-driven platform solutions or whatnot. It cannot be done based on some decisioning which is based on experience. So that's, I mean, where we see highest lift when we start upgrading, and even before we start doing AI and bringing AI into collection space, even by helping our clients simply move away from the like uh punch-based, collector experience-based.

SPEAKER_01 15:23

I mean data. And I think there's a balance there, Dan. Like, sure, experienced people who know the business, right? Like, who've done this, great collection managers, great call center, like VPs. There's still a space for them to tweak this strategy based on experience, right? Like, especially in differences in in portfolios. Like if you're an agency and you got student loan and telecom, that's two different worlds, right? They cannot a co-borro on a cell phone. It's like apples and Mac trucks. It's just, it's not, we're not even talking about fruit. Well said. It's a fruit stand and a Mac truck. Yeah. Right, like, and a co-borrow on a student loan account is where you're gonna make your money, right? Like, so they they're two different. That's where I think that collector experience, the collector on the phone is still valuable, right? Because student loan account, you're educating usually a student on the value of paying your debt. Where they don't know, right? Like they're not worried about a car, they're worried about their next meal, right? Or going out to a concert. You're like, my daughter's begging me to go, right? Like, so that kind of collecting skill is, but that you know, that's hard to put in the AI. That's still gonna be there. But the data that tells you ask for this, ask for this, should be structured and should have rules, as Dan said. You're just gonna be everybody's more productive, and then let the human with that experience make an exception. You want it standard, and then the exceptions to come. And so we've talked about behavioral data being kind of that differential, right? Like, how did they behave based or what reaction did you get from your action and using that as predictive pieces? Are there specific signals within that subset that are like that's the one? Or is it a series of just kind of understanding the blend? That's a good one. It's a complex question. I know it is, that's why I threw it out there.

SPEAKER_00 17:24

That's a great one, though. I mean, that's important. So so to answer that, we need to split the collections process into pieces. So, what is collection strategy in general? It's a communication strategy and treatment strategy, meaning how to communicate with the customer, and then what type of payment option to propose to the customer. Now, let's delve deeper into each of each of the two. So let's start with the, I don't know, let's start with the simpler one, treatment strategy, because it's a kind of single purpose. It again consists of several things: type of payment option. It can be settlement, it can be race period-based payment plan, it can be just a standard payment plan, and then it's the amount. So, how much is the customer capable of paying? Can afford pay. So, and the combination of these will answer the question. Right then, then we're not now we need to start predicting those things based on the data. Everything needs to be driven by the data. And then, depending on the predictors or the power, predictive power of those data points that you have, you will define each and every one of these two. And I just don't want to spend too much time on how it's done. We can have a separate conversation on that. Now you get real nerdy. Yeah. But the other part is the communication strategy. All right, now we need to define the tone, the time to communicate in today, the intensity, the frequency, the combination of channels, like phone versus SMS versus email, or both, or just voice and SMS. All that is again data driven. So for each of these decisions, which are all, like I said, need to be data driven, you need to define the corresponding list of predictors that have sufficient predictive power to make the right prediction, which is stable and accurate. Because any modeling prediction has two dimensions in terms of quality, right? One is accuracy, the other one is stability in time or segment. If it is not stable across segments, then you need to have a segmented approach. For each segment, you need to have its own model. AI, by the way, and ML machine learning makes it simpler. You don't need to create segments, they will be created by the model itself. But still, accuracy and ability in time stay. So you need to keep an eye on that. So, and now you need to delve deeper into the data and see, okay, do you have how accurate is my prediction based on the data that I have today? And then you look at Genie score or whatever, and you see that, okay, not that accurate actually. Or peraccurate. How is it stable? It's stable enough. Okay, good. Then no need to do much. Now you can keep it as is because why fixing something if it's working? But if you see that the prediction level is poor and you're not making it stable, or you're not making it accurate enough to use it in real business, then it's time to think, okay, what other data points can I use? What else can I store or use for predictions? Ideally, this is where experience kicks in. So some of the teammates need to tell you, okay, we also have this, that, but that data is not stored or not provided to the list of predictors. So now you need to start enabling that. Um or if you've exposed everything you generate in-house. Because sometimes some companies they use vendors that execute their email or SMS. But those vendors don't provide all the details, the whole deliverability final back to their clients. They just do it for them. It's convenient, they have a very convenient front end on the browser, it's very easy to manage the process, but no data back. So that's a first no no-no. You have to get the data, it's your data. Payment, or even though it does everything end-to-end with all the omni-channel approach, when the customer starts discussion, one channel switches to another, and the conversation moves on without any rollback. We share everything with our clients because it's their data. Every day at a transactional level, our clients get all the data, all the clicks, events, open events, and so on. Additionally to that, we also provide always this quick side view, right? Which is like a BI tool where they without any coding, without any SQL querying, they can get whole visibility of what's going on with their portfolio by checking those methods. Yeah. That's the end goal. So if your company cannot get it today using the vendors you use, that then it's time to change the vendors. You have to use another vendor who will do it and share the data with you. I think too though.

SPEAKER_01 21:31

Can't disagree with anything that you just said. I think that's right on point. One of the back to the qu like the quit like what is the key data point for collections? I think it's engagement, right? Like behavior, so the engagement from those behaviors. But when you were talking about putting that data back, it's not like these folks can just, you know, boy.

SPEAKER_02 21:56

You got a little one on the rock? Let her in, let her in. She's welcome. She is more than welcome.

SPEAKER_01 22:03

She's gonna burst them in like the Kool-Aid man. Two doors. I'm sorry, what were we talking about? So we're talking about behavior and providing it back, right? Okay, yeah. So when we think about the data that you're providing back, they're not just creating a field, they need to create a module that's tracking this because you're providing back so much detailed behavioral data. And if that's where the gold is, then they need to be able to house that and try and find ways to use that for their phone strategy or their other non-digital channels. But that feels like a whole learning curve in and of itself. Is that something that you guys are are helping them prepare for, or you're kind of walking them through how to do it and giving them the blueprints so that they can build it within their own world? Right. So like imagine so the the tool's a shared dashboard, for example. And and you know, everybody's got one. Tableau, Power BI, quick site is what we use on AWS.

SPEAKER_00 23:06

I think it's uh generalization by big margin, not just that everyone. I don't think everyone even knows what tableau is. Okay.

SPEAKER_02 23:15

Some people have used it.

SPEAKER_01 23:17

Not many he met a guy once that knew what Tableau was. Let me uh clarify data analysts use this all the time, right? Like data science people love it. You know, they are they they're they're pushing you for these tools. I think what we do is provide the tool, and then part of the you're right, Adam. A key thing is we walk, like after implementation, you go live, you're gonna you're gonna see some status, right? So then we're gonna walk you through a dashboard so that A, you know how to use it and get value out of it, right? You don't just give it to you and say, good luck. Because there's so much stuff. I don't even know how to use our dashboard. To be 100% honest, I said I'm the least technical guy working in an AI company ever, which I still hold true. But we walk them through, and there is I I love the data because there's so much. We've seen data even in our demos, look at this and be like, how much do you charge? Well, how can I get this, right? And I don't think you need to take all the data back because I don't think CRMs are caught up on that yet, Adam, to your point. Like, as long as that data is shared and accessible, you know, you're just gonna log in and you have it, it's basically our clients. I don't know, you know, they have 20, 30 people on there looking at that data. We have on early stages provided dialer lists from it because behavior, like to me, if Dan goes on a website and he can't log in to pay make his payment, and I go on and I click the link, right? Like we both got emails, we both click a link. He couldn't log on and I could. And I sat there for 20 minutes, but I didn't make a payment. Those are two different responses should go back, right? Like Dan says, here's how you log in. Simple, right? Like so many, but we don't like you know that is a huge thing. Get him to log in. You're talking about basic e-commerce, right? Like you're talking about leveraging basic e-commerce strategy that's been proven since the late 90s, right? Nobody wanted to wait it two days for delivery, and now it's right, but now it's the one of the largest companies in the world. So I think that basic e-commerce strategy works with current consumer behaviors, works with current consumer mentality. It works. Like clearly, it works. It's the same reason that we all buy stupid shit on Facebook and Instagram. But right, right.

SPEAKER_00 25:42

I didn't raise this point because somehow, I don't know how it happened, probably the matter of priorities for lenders and other industries. Somehow, marketing and front end, like like sales and marketing, they were more advanced sooner than back end functions like collections.

SPEAKER_01 26:01

Somehow, because there's no regulation, they're not dealing with like they don't have an FDCPA, they don't have seven and seven.

SPEAKER_00 26:08

I think you can still float somehow around that FTCPA. It's not that restrictive. You can still make it work. We are doing it today, right? It's the same FDCPA. RegF was added, but still.

SPEAKER_01 26:18

But regfanged the Reg F changed the format for us, right? Like Regf opened the door to say, like, it is okay to text an email, because if you look at the growth over that time period in terms of deployment of that type of technology, it grew significantly post-reg f like exponentially just skyrocketed. I think, Adam, back to the original thing is though, but I think e-commerce made Reg F. Because Agreed. Agreed. There is no Reg F clarification. If people weren't like, why don't you just text me this nonsense and stop calling me? Right? Like the consumers won't. It's very true, right? Like it is, yeah. Uh and I remember talking to Kelly Net Persevens about this. She's like, Of course, how is this a good form of communication calling people all day? Like, nobody knows it's real. Like an email with information where to call back, you can you can go to Google, look it up. It's a way better format. It's also easier to catch a spoofed email than it is a spoofed phone call. Of course, and a spoofed, you know, link and a spoof, right? And like they're gonna spell, you know, a bank's name wrong, who's supposedly the original creditor, right? Like all this. So I think this is like let's face it, like Target and Walmart are now shipping bikes to my house in an hour, right? Like, yeah, because of the Amazon made them, right? Like, there is no more non-e-commerce. Banks are e-commerce, right? Like, creditors are all e-commerce. Well, I mean, you've you've got fintechs with no physical locations. That's an online business, and the consumer historically has wanted to be serviced through the same channel in which they originated the debt. So if they originated online, there's a high probability they want to deal with it online. If they if they filled out a credit art a credit card application in the bank branch, they're probably going back to the branch to make their payments and engage with questions. And I mean, look, maybe it's just because I live in Florida, but I see it at the bank all the time. Like people going in there to make their payment and to have a conversation when totally unnecessary, but that's the engagement that they want because that's how they originated. And I think that continues going forward.

SPEAKER_00 28:25

And I think it's what we call Amazoning of debt collections. That's how our tool developed, right? Exactly under the same strategy. So the idea was to provide the same very experience. These the customers are used to across other functions, but were not provided in that collection space. If you look, but this is land if if you look at lenders, it's not that bad. At least they use some digital channels. There is some proactive payment options proposed. But if you look at other industries like healthcare, oh my God, they they still like account a cryptic statement on their patients after the service with a high bill, like sometimes hundreds of thousands of dollars, without proposing any proactive payment option, like payment plan at least. They just send the letter, hoping it will be delivered, and then they expect the patient to go back to the reception and discuss it with somebody who will answer the call. And it may take some time to wait in the line to get to that person on the phone and negotiate it verbally, like and now the healthcare provider spends time with that person discussing something which could have been done actively and sent digitally, maybe several times in a row, uh without wasting the time and money on that human call. But now we're going a little bit to a different area though, which is more of a customer experience.

SPEAKER_01 29:39

Well, let me I want to go back to the customer experience because I think that's an important part of this conversation as well, because the real-timeness of data is something I want to make sure that we cover. But since we were talking about the signals, I wanted to ask you, it always feels like the collector notes is where there's a lot of gold, but unstructured data is the enemy of efficiency. What has been your experience? Have you found gold in the collector notes? Have you been able to organize them or extract value from them? And once you do on the backlog, how do you make that happen going forward?

SPEAKER_00 30:14

I have a real life example on this one with a very large lender. So they were struggling with the RPC rate, right? So they could not make it more than like 3% or something. At that time it was low, 10 years ago. Here in the US, I'm talking. And um so they wanted to find out which data provider, external data provider, would help them best. And luckily they started using, luckily for them, they started using Excel to help them with this decisioning. So before even advising them, and they started already getting some data points from different data vendors, we delved deeper into their existing data structure and data availability, and we realized that like most of the cases where they could not reach the customer, because originally the customer's phone number was validated during their origination, right? But then somehow people were changing their phone numbers, email IDs. And interestingly, they were telling all that to the agents during the conversations with the collection agents or even during the maintenance phase at the branch. All that information was stored in an unformatted manner, yes, but stored in the logs. But nobody ever looked into it. So instead of buying an extra data point from some vendor and pay money for it, of course, we increased the RPC uh 3x. We increased it. So by just using the logs. Yeah, yeah. So uh so before even looking into the external data providers, look into your log. So unstructured data is the key. Two things though. You need to firstly uh record everything as much as possible. That's another advice I would give to the industry. Record, record, record everything, and secondly, store the recording, but it's expensive in terms in terms of weight because audio recordings are heavy. So it's always better to convert it into text. So STT tools like speech to text, there are lots of them in the market, or you can use uh vendors who will convert it for you and store the text because text is light on the data, so you can store it on your servers. And then there are so many ways to get insights from that data. So here you can hire data analysts or data scientists who will do it for you internally, or you can use another vendor who will make use of that data, like EXL, for instance, but there are other companies as well, just who will take everything you need, sometimes what you don't even expect to see in that data, and store it in a structured manner for you, so that you can use it in your conventional models. Or you can use some um you know types of models that work with unstructured data that is also available today. That's where AI actually shines because AI can work with unstructured data and get use of it.

SPEAKER_01 32:57

Yeah, we we we have a tool that can pull it from like all these inbound sources, your emails from the customer, calls, everything, texts. And you get if it think of it just like simply, right? Like if there's something, if there's 15% of responses are on one subject, and you're not tracking that subject, you're missing out, right? Like we did one with the utility and we saw you know like uh unstructured responses, like unusual responses in like 10 or 12% of the population. But we could pulled six common responses out of that, made a category that they can now track. So that's again where I think, and then you tie that to someone with experience in the collection space, and that in that unstructured data is now gold for them to either contact that uh customer, maybe check special offers to those type of customers, or stop working those customers. You know, like whatever it is that data is telling you is like, oh, they're not gonna, you know, a factory closed, they all lost their job. Or they're all on furlough until you know the government is back, you know, like don't don't call them for a month because they know like it's a government show because you're not helping. Well, that but that goes that goes back to the customer experience conversation, which I I want to make sure that we get back to here because as you break down the barriers between silos, you build bridges. And as you build those bridges, now you're changing that user experience into a more real-time situation. It's less about passing big files back and forth and more about connecting APIs or data sources so that your entire data-driven ecosystem is able to accelerate based on that data, right? Like that's the fuel for the engine. We want to make sure that it keeps going. What have your experiences been like? Is organizations have reached that real-time capability? What kind of differential do you see from an organization that's been living in a green screen world who all of a sudden has real-time data flowing throughout the organization? And where across the organization do you see impact? I'm thinking of a certain franchise.

SPEAKER_00 35:26

Like I I think there's so usually what we see with the first deployment of the decent data structure and some feedback on the data as well as some digital outreach, they usually go hand in hand because you once you get the data, you immediately can deploy decisions based on that data. So once it happens, the lowest lift in collected amount we see was 20%. Usually it's higher, but 20 is the least we've seen. Yes. That's that's what we usually see, but that's just the beginning because then we start applying AI for the outreach, and now we have two drivers of success. One is the collected amount which increases, the other one is the cost to collect because it reduces using AI. So, but that's already kind of beyond just data point, right? I just also wanted to expand the previous topic, which we kind of covered, but some of our listeners may think, okay, great. I am recording everything now. How do I make use of it? But what if I don't record anything? What if I simply don't have that luxury? So what do I do then?

SPEAKER_01 36:31

You have recordings and you can't mine them for data, right?

SPEAKER_00 36:34

Like, or let's say those who do not even record anything. There are lots of companies who do not record it, store for a month and delete because they don't want to store it as expensive, right? So for them, actually, then there's still a way out, interestingly, because there are, well, let's take payment or because that's the tool I know best, as an example. We recently got a patent on reinforcement learning. It's a very advanced uh machine learning technique that experiments with with the with the process. So in in our case, it finds the right combination of channel and time to communicate, frequency, and it learns from the from each customer on the fly. And it also reacts to the customer's change of preference. So if one customer is responding to email and then realizes, oh wait, I can respond to text and I can talk like to like a real human, and that preference changes to the SMS, right? The tool will immediately see that because it tracks the customers, it experiments with different channels and time and sees where the customers are mainly engaged and starts concentrating there. And that's what reinforcement gaining is about. Experimenting. So basically learning through constant experimentation. So if you do not have any data, start using tools or vendors that provide things like that that will quickly start experimenting on your population and immediately make sense of your data and understand the preferences of your customers and take it from there. So you don't need to have years of data stored to be able to start using AI or anything advanced. You can actually uh in less than three months, you can accumulate enough data to be able to make data-driven decisions. If you use the right vendors or the right tools or modeling techniques, if you have enough talent in-house.

SPEAKER_01 38:15

Yeah, sorry, I kind of no, it's that's a that's exactly what I was trying to, but that's what I'm trying to understand. Is so like it we're improving that data flow. We get to this real-time data structure because I feel like the inclusion of the data is one piece and the connection of everything is another, and you probably can't measure those two things separately based on the way that I'm I'm hearing it. But I mean, Mike, you've you've been you've been working with these collectors for a long time now. What are you seeing as they've started to make that migration to real time? Like as we see that data connectivity and ability to learn now versus tomorrow. I think it's just it's just way more productive and it's way more, it sounds crazy. It's more personalized collection journeys, right? Like when I was first started, when you filled out every RFP and you had this is our early out strategy, this is our and it's this many attempts, and it's this and it's this, and this is what happens, broken promise, this is the way we structure it, that's all gone. You know, now the treatment strategy is Adam Parks, Dan Yakamenko, Mike Walsh, right? Like it is, it is personalized, and it seems crazy that a machine gets more personalized, but even you know, it works with people too, right? Like again, I think what agencies like is they use the digital, they get um behavior point. It's usually a behavior because that's what is the real time. Like that's the juice, right? That's what yeah, payment. Well, payments and and behavior, yeah. Right, right, right, right. So even yeah, obviously, yeah. So then payment is a behavior. So correct. It's a great one, it's the best one. It makes us happy. But even if it's not a payment and I see something, and then I can call that person, or I can then say, okay, this person isn't reacting. Let's send a letter with a settlement because maybe they're old school, like you said, maybe they're the like my mom who walks into the bank and talks to people, right? Like a total waste. But I think that's where I think the the marriage of tech with people who've been doing it collections for you know, have experience with different portfolios. They understand, like, when you look at utility, right? Let's take a Texas utility, for example. Their bills are a lot bigger in summer, right? And they have a lot more cutoffs. Why? Because it's a hundred degrees and everybody's sweating and their air AC is cranked up. You know, that's not the same thing in Maine and New Jersey, where I'm from, right? Winter is cold and they have high bills, right? Like, so that knowledge and that expertise mixed with the tech, and now you can adjust the tech. Like the tech is not this is what it does, and that's it. This is we send, you know, the text on this date, the email on this date, that's gone. We send the text when Mike Walsh needs the text. We send, we make the call when Mike Walsh should get a call. We make, you know, that's the technology. We send email to support the text to work to follow up the phone call. It all works together. And it's all it's you're buying a brain, you're buying an orchestration engine that never forgets, that knows every activity and lack of activity. That's what we're all trying to get for the most efficiency, the most valued use of our collector time. Right? Like, let's not waste our time on outbound calls that 3% of people are gonna answer, if that.

SPEAKER_02 41:48

You know, like let's let's let them take their own 30% out.

SPEAKER_01 41:55

Yeah, yeah. So let's let them take the inbound stuff and the and the hard stuff that the AI is gonna shoot to them. Like, I think that's it's a different age and it's uh it's easier on the consumer. You know, I I've always said uh Dan always uses the healthcare examples because I go through healthcare hell all the time. Like I have so many doctors for my son and and go through things. You don't even know what you're paying for, like 90% of the time, right? So making it easier, simpler, faster, you know, you're the Amazoning or door dashing of debt collections is just it's what you want. I think it's great for everybody. The creditor, debt or slash debt buyer of the agency.

SPEAKER_00 42:39

Well, I guess it's already showed success in marketing. Why wouldn't it show success in collections? Right. It is showing it to us, but just think yourself, if you don't believe us, if it shows success in one industry, which is way harder because the customer doesn't owe anything to the marketologist, right? They plug him stuff if he doesn't need it, he still buys it. Right.

SPEAKER_01 42:57

Well, and and they already bought what they're paying for now, right? Like they already received the item, the motivation is not the same as new shopping, but the human response and behavioral patterns to those stimulus works. Marketing, I would my master's degree is marketing, not debt collection. But let me tell you, those two things are very, very similar. Well, very much. The intersection of e-commerce and debt collection is coming, whether we like it or not, because that's I think what a lot of the future is going to hold, especially with an increase in the volume of accounts to manage, a decrease in the liquidity of those counts. We're going to have to find more ways to work efficiently when we can't hire folks, and at the same time, the FTC wants to stop offshore calling. So you've got a lot of challenges happening for the industry at the same time. But this has been a fantastic conversation, guys. Any final advice for our audience who is at the beginning of their AI journey and living in a world of siloed data? I would say Just keep looking, keep talking to people. The one thing about this industry is we share information. You know, we've been in it forever, since 96. Talk to people, get information, get demos, take a look. Your data, if you're collecting, if you have a human being on the phone collecting on an account, you have enough data to use AI. You do. Because that's what the AI is trying to replicate in different ways, but pretty much the same thing. So you can do it. It's gonna change your business for the better when you start the process, when you do a little bit of learning. Again, coming from the least technical guy ever at an AI company in the history of the world, it seems like a really big and hard process. It's not once you you'll catch up, you'll get a little bit of information. There's some information out there, there's a couple good LinkedIn learnings about AI, just to get the basics and then generative and then AJ. Take those, go out, talk to some vendors. You'll find a solution that fits.

SPEAKER_02 45:02

Doctor? Thank you, Professor.

SPEAKER_00 45:04

So I already said that, but just to summarize. Well, firstly, record everything, get track of everything you can in your collection cycle. Sometimes companies think, why would I like web stats, right? Google stats, for instance, on the portal. Why would I need it? I'm not using it. You're not using it today, you definitely will be using it, if not next year, in five years, right? Or you will just die out because you will not survive the competition. So record everything you can, uh including the portal data, the mobile application data if applicable, and of course the whole delirial funnel, the communication logs, the statuses of the accounts, like promise to pay broken products, all that needs to be at least recorded, whether you use it or not. Secondly, make sure that all the decisions in your, not only collections, and your whole business cycle are made based on the data. No experts, no experience. Experience only in terms of working with the data to make use of it, not replacing database decisions with some expert thinking. Okay, I believe this worked better, so I'll do it again, and it will work better again. Well, that's that's a very faulty approach. And lastly, uh, digitize. We're living in the era when like if you're and I'm not saying immediately jump into AI and start using AI voice and you know, replace humans with that. Eventually you'll get there, but start with something simpler. Start with using digital channels for the customer outreach. Start applying some uh like simplistic segmentation with some programs based on the data again, using those digital tools. Right? If you cannot do it in-house, which is not difficult these days, but if you don't want to hire the right talent, outsource, but do it sooner because time to go live with that is way more valuable than you will save a couple bucks and do it yourself, but two years later, by hiring the right people, and doing it, and failing and then doing it again. And then lots of our clients that did not go with our tool by saying that we will build it ourselves, they're still building and they haven't gone live.

SPEAKER_01 47:10

The build it yourself thing. Mike and I have gone down way that right way down that rabbit hole. Building it is what I mean, that is a much bigger project than people give it credit for.

SPEAKER_00 47:21

Yeah, time to go live is more important because speed to market is the key, not only in collections across the board, and business people will understand. So these would be the key recommendations. Store the data, make use of that data in your decisioning, and digitize.

SPEAKER_01 47:35

Well, gentlemen, this has been an insightful discussion, as every conversation with the two of you really is. I appreciate all of your time. I mean, just so much to cover here, but I think listeners will really get a great taste of the challenges that most organizations are facing, in that there is really a light at the end of the tunnel with the right partners, the right tools. You can bring the data together, you can make it happen. But Mike, this is my favorite podcast now because my co-host is always so much fun. So I appreciate everything from you. And Dan, your insights are always excellent, man. Uh just so well spoken. I appreciate your time today. Thank you everyone. Thanks, Mike. And thank you everybody for watching. We appreciate your time and attention.

SPEAKER_02 48:21

We'll see you all again soon. Bye.