Hello, everybody. Adam Parks here with another episode of Receivables Podcast. Today, very excited to have OHAD joining me, the founder of TrueAccord, who has been in the space and was one of those first movers when it came to leveraging RPA and all different types of technology. So really excited to finally have the opportunity to chat with you today. Thanks for having me. Excited to do this. Absolutely.
For anyone who doesn't know your backstory, could you tell everyone a little about yourself and how you got to the seat that you're in today? Yeah. The pre-Tramail True Court, I've been in tech for 20 years now, a little bit more than 20 years, always in machine learning for financial services. Started at a company called Fraud Sciences. We did fraud prevention for e-commerce with machine learning back in 2005, 2008. We got acquired by PayPal basically to become part of their risk optimization.
And some spent a few years there, moved to the Bay Area through there. And then I left and started analyzed. Actually, I started working on two companies that both became companies. One is called Signified, still around. It's kind of a fraud sciences 2.0, part of that generation, doing very well, thanks to my founders there. And another called Analyzed that did consumer credit underwriting. And Analyzed got acquired by Klarna, where I was a chief risk officer.
And at Klarna, underwriting a few billions of dollars per year at a time, I learned about the importance of servicing, the importance of collections. And how that area specifically did not get a lot of attention from technologists. I actually got uh some coaching about collections, and a very senior person from a bank told me every quarter you hire people, you fire the bottom performing uh agents, 15% bottom performing. Uh, and it's an art.
You don't exactly know why people collect, why people don't collect. It kind of, I like to say, it tingled my uh my spite sense. It it felt like, well, you know, I I know that I know that we can replace a lot of human judgment in these areas with machine learning if we do it the right way. And so we kind of manage tier one, tier two, tier three issues, and we bring the most complicated stuff to humans where they can actually access judgment. And um, and that's what we did.
We built an engine, a patent engine called Heartbeat, um, that replaces the human-to-human interaction in collections to human-to-machine interaction. Uh, we can we'll talk a little bit about the technology underlying it and so on. But the point is that we started uh selling it first and foremost as a uh fully automated machine learning native collection agency, which is TrueCord, competing with call center-based uh solutions.
And over the years have grown, have proven the model, have proven the model at scale. We're able to be part of the process to change regulation F, to bring Regulation F, change, change the F to CPA, and bring digital to the industry at large and and down to today with a few steps on the way that we can expand on. Okay. And so talk to me a little bit about TrueAccord, TrueML, and what that journey has been like.
You were one of the first movers to start leveraging this technology in the space, and you've kind of built the organization around it versus bolting it onto, say, traditional collections. What's that journey been like as an organization? Wow, so complicated question. I mean that I can answer it in many ways. It's been fascinating. Now, first and foremost, why did we start through a cord versus a technology vendor?
Because when we were raising money for the first time, we're a venture-funded company, we raised almost $150 million. Investors throughout our life, um, always said, just build a SaaS solution, just sell to collection agencies, just sell technology. And I told them, guys, we are educating a market. This is a market where people tell me, I would get phone calls, people telling me that what we do is illegal, uh, what we do is uh will fail. Frankly, not in a mean way. Nobody was mean against me.
It was just uh deeply entrenched belief. And so we needed to prove that this thing works. And the way you prove that this thing works is you create an incumbent and then you show that it works. You don't go and you try and sell and convince people that actually the technology is going to work. So that's where we started with through a cord.
Now, as we scaled and as we have proven the model and grown to our size, step one was lenders coming to us and saying, hey, we're using through a cord for third-party collection and it's post-charge-off. Can we use the same technology, white label, in our own operations for the length of debt? Uh and that was a turning point, a major turning point for us where we said, yes, absolutely. We're big enough so that we can say, hey, trust us that the technology is going to work.
Trust us that it is not just about penetration. It's not about just emailing people every day. It is about the personalization. It is about giving people what they need at the time that they need through the channel that they need. So we could speak with authority based on data, that now when we saw technology, people would actually listen.
Same thing now, uh having acquired ERC in 22, having acquired centric credit last year, that we are expanding into additional areas under the true ML umbrella. Because again, if we can speak with authority about, hey, actually, this is how you negotiate with debt settlement agencies at scale. This is how you actually scale it in an automated fashion. Uh, this is how you do credit reporting in a way that consumers react to and that auditors and regulators don't react to in a negative way.
Interesting. And now we're starting to expand into legal collections as well. Part of it is that lenders and creditors want one vendor to do everything if they can. The other thing is exactly what I said. As we have demonstrated that things work, we can speak with authority and say, hey, by the way, here's a new area where our method and our technology can be applied. How about you try that? And it's been incredibly successful, but we need to build into it.
That decision engine, I would think, that you're using to drive the consumer down the right channel at the right time with the right path would be really similar to the styling of technology necessary to drive that litigation side of the business as well. I mean, I realize that the processes are quite different, but that decision engine that's pointing each account down the right path at the right time, I would think is a a big competitive advantage. You're absolutely right.
And you've been always on the forefront of technology. But if we look at the industry in general, when we started, we had to deal with, oh, you send emails, we send emails also, you know, or you guys are you guys are the nice solution, or like a misunderstanding, or just not being not being very open to what machine learning is and what technology can do for them. I think the tune has changed in the last two to three years.
Incredibly gratifying, amazing to see it, great to kind of feel the camaraderie around technology and what we can do. But it has not been for the first seven, eight years of the life of the company, it was not very clear that the industry at large was agreeing, was adopting it. Now it's accelerating, and that's incredibly good. And we can ride that wave and continue to expand our services.
Well, early on, I think a lot of organizations were so concerned with laying down the train tracks of technology that they weren't really deeply considering how they were using it. And that, for example, email with a focus on email deliverability create a competitive advantage. With a focus on sending an email, you're commoditized, you're doing the same things as everybody else.
But that decision engine that allows you to kind of control where those accounts are going and then understanding the deliverability of those emails and those types of things, I think is one of those first mover differentiators that kind of puts you guys on the path that you're on now. You know, we're pushing 100 million content items a month.
Uh I can tell you I have I have all the scars to show, you know, uh discovering on a Friday night that Gmail had turned us off and we were not reaching any Gmail uh consumers. I'm talking, you know, 12 years ago, but still I remember because we had yeah, we had to learn all of these things and we had to build a very complex routing algorithm to be able to hit deliverability, which sounds kind of obvious, but it's not. So yeah, absolutely.
Each one of these things is not only needs to be adjusted to collections and how we do things, but also needs to be adjusted again and again when you hit scale. And things break when at times and in ways that you did not you could not anticipate. Such a true statement. And when you start hitting that scale and you have to be able to ingest the information or results in order to improve, creates this cyclone of data that requires constant refinement. There's no end to that ever.
I don't know the things that were final state. Proudest moment, and I agree with you completely. One of our proudest moments was from a technology perspective, was when we we had acquired ERC. And then by the end of that year was the time to flip the switch and move everyone from the ERC entity and to the through a core entity. And over three weeks, we moved 26 and 8 million accounts uh into our system. And when I say moved, I mean moved and started servicing. And the system didn't break.
That was that was that was big. That was something that, you know, we we had experienced uh cases where we moved stuff and things did break. True to experience that and experience that scale and not and not breaking, really being able to dig deeper into the file, reach consumers more effectively, and surprise the clients in the short on a short time frame, that was pretty cool.
There's a lot of elasticity to an organization that can ingest that many accounts without the system overloading, especially when you consider the algorithms behind email deliverability and some of those decision engines. And I keep going back to the decision engines because that seems to be well, one, I did a um a session with Nama Bloom recently, and we were talking a lot about skills and decisioning, which I thought was quite interesting.
But it sounds like as a first mover, you've been looking at these things for a while and looking at the not just the rails in which things are happening, but that ability to control what's happening. And when you can throttle those emails, when you can throttle those text messages and customize or personalize those communications, that seems to be the path, which leads me to the topic that we kind of set off to talk here about tonight.
But I was really interested in learning more about some of the things in that uh, the true ML, true accord story. Talk to me about the heuristic level revolution and what we're seeing right now with consumers becoming more open to technology, they're becoming more open to subscription plans. Like the consumer has trained, has changed significantly since you initially started the company.
How have you been able to keep your finger on the pulse of that changing consumer in order to continue meeting them where they want to be? So that's an excellent question. You know, when I started as a founder, I like to say we sometimes have tense confusion. We talk about things in the future as if they're they're they exist right now. And I used to talk about choose your own adventure.
Um, I'm in my late 40s, so I still remember the books where you had to choose to do this, go to this page, to do that, go to that page. The experience that we wanted for consumers. Our assumption was not that just be nice to people and they'll pay. And of course, not, hey, be starting with people at scale with technology so you reach more people and they'll pay. No, it was about giving the right person the right treatment at the right time.
And furthermore, it was not about looking at Joe Smith and saying uh he owes uh a firm $500 and this is his score and this is going to be his treatment plan. No, it's about with every interaction, deciding what is the next step. And this is why we call the system heartbeat. It beats, it looks at every consumer, it looks at the history of the consumer, it looks at in a multidimensional space, who are the consumers that are similar to this consumer.
What has worked for them in the past, what channel, what communication, what approach, what payment offer, and so on. Through a combination of these behavioral cues, through a combination of the historical data that he has, it that we have, it decides what the next step should be. Now we can inject heuristics into the process. We can teach the system, this is what we call uh feature engineering.
We can teach the system to look at actually um look at when their payday is, look at how many communications they've had with us, look at what is the sentiment of their the text when they when they reply to us, and so on. This is an important distinction of a system like ours from uh AI systems that are black box where you don't even know what it's doing.
We teach the system what to look at, much like you would teach a very sophisticated agent, and then it takes all of the data that we have from all of the history of communication with tens of millions of consumers, and it decides how to communicate with a consumer. And the thing is that the scaling with data is almost infinite. What do I mean with that?
As you get more data into your system, there's almost, we have not found, and a lot of Frontier labs have not found, the an end to how much better the model can become if you show it more cases. Because it it still learns differently than the human. It can look at everything, every case, all cases at the same time, all historical data. And the more cases you have, the more edge cases it looks at, the more, the more irregularities it can learn from.
And that adds to our ability, to the system's ability um to treat everyone exactly the way they need to be treated. And on top of that, another topic that's been hot to debated lately, reinforcement learner. That's a system where the system itself is what the result was of its attempts, and it can say, you know, I'm kind of simplifying, but you know, it can say, hey, this worked, let's do this more. This did not work, let's do this less.
And we can actually see when we look at, for example, we started a new client, we look at cohort one, cohort two, cohort three, cohort four, we start seeing the curve of repayments sloping up over time as the system learns that specifically. Correct. It learns from the vintages, like let's bring up the things that work for this type of consumer earlier into the cycle while they are extremely engaged. And that really changes kind of what the vintage looks like.
So that has been, I think, something I have not seen that implemented across uh many solutions at scale. I'm sure there are, but have not seen many. It's not easy to do that at scale. That's a that's a big lift to have the system continually learning from its own actions. And what does that start to look like? Because you're still keeping that human in the loop and you're still looking at the the larger picture to make sure that you're staying on track. But I think it's interesting.
Are there any specific signals that you've found that are more intuitive of a consumer who's going to have a higher repayment? Like, are there any specific data signals that you've started looking at and found that these are more important than others? So the interesting thing is that I would say two things. First and foremost, we don't use demographically that or we use it to a very limited extent. And early on, we made a decision to not use credit score data.
At least now we use it in some areas, but we decided not to because it was expensive and because we wanted our inference to behave differently than the rest of the industry, because at the time most of the industry was using credit scores. Um and so number one is that a lot of our signals are behavioral based. So it's not about determining in advance who is going to engage with us more. It is about because the marginal cost of sending an email is close to zero. Right?
It is about engaging with them and then seeing how they react. Did they open the email? Did they open it a few times? Did they open it different times from different devices? Did they click through the link on the email or the text? Did they browse on the website? Did they play with the widgets? All of these things add back into the system, and the system on its own decides how to optimize basically. And this is this is a very important thing.
We didn't need to teach the system which indicators are better. We just need to teach it where to look and then it optimizes on its own. So that's one very important piece that's very hard to do when you don't have the scale of data that we have. The other element here is that counterintuitively, I think for many of us in the collections industry, true accord is a brand.
It is a recognized brand by consumers that they can research online, they can see recommendations, they can see kind of how people interact with us and so on. As a result, there's a higher level of trust when we contact consumers. And that means that we can ask consumers to tell us how they're doing and they will tell us, they will actually react. And so, for example, we don't need to guess when their paydays, we can make approximations of when they get paid, but we can just ask them.
So when we set up a payment plan, we can ask them when do you get paid? And then we can tailor the recurrent payments and the re and the and the retries if something fails and so on around their payment dates. And so that increases success rates of payment plans.
On the contrast, if we get we have more data about their cash flow and how much work they've done and so on, we can detect whether they have more money in their account or whether they're in a position to make a larger payment, and then we can trigger that and we can encourage them to pay in advance for their for their plan. So as a result, we see double digits reduction in plant breakage because it's about adjusting to what the consumer does. Interesting.
The behavioral signal aspect of it makes a lot of sense because it is hard to predict, but based on the actions and behaviors that they take, it's probably a little easier to categorize and optimize that that journey. As you've looked at these behaviors and you've watched the behaviors change over 14 years, and consumer behavior has dramatically changed in the last 14 years. What do you think the next five years starts to look like from a consumer engagement perspective?
Do you think we see new channels? Do we see deeper channels like RCS and text? Any any insights into what the next five years holds? So I think there's I want to separate the technology that is used for delivery of communications from consumer behavior. Regarding technology, I'm not 100% sure. You know, it could be it could be RCS. That that is definitely where the industry is going. We may find, you know, suddenly the consumers adopt uh WhatsApp. I don't know. Not likely in the US.
But so you know, channels aside, I think the question is what are consumers used to? When we started, consumers were extremely, extremely suspicious of collection uh communication delivered by email or by text. Now they're used to it. But when we started places where where they didn't know if they were talking to a human or a bot and they cared about that very much. Now they don't care. They actually communicate with a bot in a way that's equivalent to how they communicate with the UN.
I think if anything, we are going to see a lot more communication that's aided by LLMs and maybe a lot more structured. I think all of us see that in the uh pro se plaintiff wave that's that's clogging up the the courts. I think from a consumer perspective, I think it's beautiful that people who are unsure what to say and when can use a tool that kind of gets them going. On the flip side, it kind of feels like the you know, the the la last place where people still believe in magic.
They think that if they say the right words in the right order, then things are going to change in the world. And that introduces interesting, interesting challenges in terms of how you communicate with people and them thinking that they said a certain word and you have to do it just a different way and so on. So I think there's going to be, weirdly, more structured to our communication with consumers in the baseline because they're going to use LLMs more.
And on our end, we've implemented with tremendous success LLMs and RPA into our back office to allow us to respond to consumer concerns that were historically unstructured in a more structured way. So you can imagine how on our side we use agents, on their side, they use agents. At what point do the agents just speak to one another without humans at all?
That is something that we're thinking about and that we're fascinated by, and kind of we're running some experiments around and uh and kind of we'll see where that takes us. I'm very interested to see where that starts to pan out because I know the debt settlement companies are already working on. Bots. I'm sure that the consumer attorneys are going to be working on bots with the intempt with and which is a totally new attack vector for them. They're going to love that opportunity.
And then you've got the ones that'll be operating for consumers. And what does that mean in terms of third-party disclosure? I think there's a there's a really interesting dynamic in that bot-to-bot situation. And I think being able to identify on our side whether or not we're talking to a bot is going to be important. My understanding to this point is that most of those incoming bots have trouble validating and authenticating as the person, and they get knocked out from that perspective.
But just like our technology continues to get better, I expect their technology will continue to get better. And hopefully the bot war uh, you know, won't start too soon. But I feel like it's something that's already started and it's just gonna ramp up over the coming, it's called 12 to 18 months. We'll expect to see more of that.
I will just say, um, in addition to that, hopefully we will uh we're always ha going to have the um confrontational uh aspect of of lawfare and and kind of and textual warfare. In addition, we can hope that there's going to be a segment, maybe a growing segment, that's going to basically say, hey, look, just talk to my agents. Here's my permissions, here's what the agent knows. Talk to my agent, figure out what needs to be figured out.
And we think that we don't know if this is an imminent thing. It seems a little bit, a little bit further out, but we want to believe that uh there's a world where we basically say via API, via NCP, via whatever, you know, whatever the agents can understand, we can say, hey, consumer, this is what's available to you. Let's let's let agent stock. You don't need to, don't stress about it. You know, we'll figure it out with your agent.
Maybe there's a world like that, and we think it's going to be a pretty cool world. And again, if you can do that at scale, if you have the data to understand what it is what to say to whom, I think you'd be very successful in that world. I think it dramatically changes the debt settlement engagements. Absolutely. We're already seeing a lot more structure in our communication with that settlement, and this is why we can structure and kind of automate those conversations at scale.
Yeah, we we will see where that leads us. I think it'll be interesting. I again, I I know that they're working on the bots as well, and I think everybody's trying to come to you know their piece of the puzzle. I'm I'm curious to see how it evolves over the coming years. You know, look, this has been a great conversation. I'm glad we finally got an opportunity to connect. Is there anything that you wanted to cover today that we haven't covered yet? No, thank you for having me.
I mean, again, I am so happy that we're talking about technology in the debt collection industry. I think that's been long overdue. I encourage everyone who wants to geek out about technology to reach out. I'm always happy to have these conversations. I think that's more than anything, an exchange of ideas, especially the ones that are several years in the future, doesn't hurt anyone. Just helps us all understand the world as we're operating in and what we're seeing.
And I'm excited for what the future brings. I think that the industry, this is a slow-moving industry because of compliance reasons, because of regulatory reasons, uh, but it's it's finally where we wanted it to be. And it feels like the change is accelerating. And I think that really everyone's positively excited about it, myself included. So very cool to talk about it. Well, you've built a great team, and I think that helps to develop some great technology.
I'm really a big fan of just so many people in your organization. So I was very excited for an opportunity to have a chat with you today, OHAD. And thank you so much for joining me. For those of you that are watching, if you have additional questions you'd like to ask OHAD or myself, you can leave those on LinkedIn and YouTube, and we'll get responses out to those. Or if you have additional topics you'd like to see us discuss, you can leave those in the comments below as well.
And hopefully I can get OHAD back here at least one more time to help me continue to create great content for our great industry. But until next time, OHAD, thank you so much for your time. I really do appreciate it. And thank you everybody for watching. We appreciate your time and attention. We'll see y'all again soon. Bye, everyone. Cheers. Thank you.