‘To The Point’ Bonus Episode: The Future of Fraud Detection w/ Justin Davis

On this episode of ‘To the Point’, our VP of Product Justin Davis dives deep into Point Predictive’s innovations from our Risk and Innovation Lab. Discover the latest advancements in battling fraud, the power of the new tools like AutoPass™️, and insights into dealer risk. If you’re intrigued by how data and technology are reshaping the fight against lending fraud, this episode is a must-listen.

Click below to listen to the full episode or read the full transcript.

Intro:
This is To the Point a podcast from Point Predictive.

Bill Hall:
So, Justin Davis is going to discuss a little bit about, innovations, what Point Predictive is doing from our risk and Innovation lab, what we’ve done, and a couple maybe new things on the horizon.

Justin Davis:
I get the opportunity to share about all the cool things we’re doing with the data and all of the new ways we’re innovating, products at Point Predictive, frank, actually oversees. So we spun up kind of an innovation lab in house that is just strictly focused on innovation, looking on the horizon, what’s happening with technology, with AI. And so I get the opportunity to work with him a lot on how we’re going to incorporate that into our products going forward, and how we’re going to continuously innovate in the fight, against fraud.

So, I will start with just a few things that we’ve been able to accomplish over the last year, and then what is kind of being cooked up in the lab we call, Frank the Mad Scientist. so a lot of the things that he’s working on, so that got very bright. so first off, last year we launched a new tool called Autopass. and as, Donna had mentioned previously, GLBA compliant, that’s Auto fraud manager. That was our flagship solution, primarily used to add friction, remove friction, but not necessarily to decision. and we continuously heard from our lenders and from the market that there needs to be a way to be able to utilize this data in a way to decision loans, to price, out risk, and to just better price the loans in general. And so we were able to actually stand up a consumer office, take, a kind of a snapshot, a copy of the consortium data to be able to be leveraged within that consumer office to power, ah, autopass that can now be used for Auto decisioning, for adverse actioning, for, pricing, and for changing structure. Secondly, we released actually two, products around dealer, risk.

So, we don’t just assess the risk of an application. I mean, from the majority of the conversations I’ve had today, it sounds like dealer risk is actually on the rise. The majority of the people I’ve talked to that’s been the main focus of our conversations is around dealerships. and so what we wanted to do is aggregate that detail up from the application level to understand what risks do the dealerships present themselves, right? So not just, hey, the application risk for this dealer is rising. that might show that, maybe there’s some risk internally, but maybe they’re just being targeted by fraud. Ring. But how can they be seen? What does their performance look like, within the four walls of a lender? And then what does their performance look like across the rest of the consortium to see if is there potentially adverse selection happening, and what might be the reasons for that. And then being able to also, look at very particular fraud trends that might be perpetrated by the dealer compared to what might be perpetrated by the borrower. Right.

Is the dealership, perpetrating these schemes, through maybe the FI office? Right. Whether that be recycling, employment information, recycling PII information. I think there was frank likes to talk about the dealership that had a, ah number of synthetics that were all coming from a address. Right.

I think it was here in Texas, right down the street from the dealer. And that address was actually owned by the F and I manager. So, just spinning up synthetic identities and submitting loans, same, day income deviations, right. In the indirect space that one application is sent to 1020 different lenders, and on every single application, you have differing information. It’s probably not the borrower doing that. The F and I manager is probably just trying to key in some information and try to find the sweet spot to get an approval. So being able to take all that detail and then, put it into a score and attributes and information that can be powered not just to look at your application risk, but the entity that’s actually submitting those applications. we also, within the case manager that Donna was talking about, we announced real time rules, within that. So not just being able to create rules that create cases or send notifications, but actually be able to drive real time decisioning back to an Los.

So, for example, using, the hot list that you had mentioned, right, I know this is a fraudster. If they ever come back in, and submit an application. I want to create a rule that references that hot list that says anytime they come back in, I want to trigger a decline or a review back to my Los. And not just create a case. I’ll create a case for Queuing and for my team to review. But I also want the Los to be automatically updated to say, I don’t want to do business with these people. I’ve seen them commit fraud in the past. So, being able to use that rule engine to utilize any of the input data, any of the output data, to drive real time actioning back to the Los, and then we’re also releasing, new alerts consistently. Right. So we have our team of fraud investigators, fraud analysts internally, that are just trying to look at new types of fraud. How is fraud emerging? How is fraud evolving and changing? And how can we use the data to be able to highlight those trends, highlight those red flags in real time, so that you can take action on them? Right.

So, we’re getting into a more consistent cadence with being able to release these on a quarterly basis as we start to see feedback from lenders, as well as what we’re seeing when we look on a daily basis through applications. And so here’s just, some of the alerts that, we have either released or are going to be releasing at the end of this month. Right. Specific dealer perpetrated fraud, red flags, a fake employer look back. So we did an, analysis on applications where a fake employer was used. If that consumer had been seen in the future, there was a 70% chance that they used a fake employer again. Right. So if we were able to look back, at a consumer, they might not be using a fake employer on this application, but they used one 30 days ago. There’s probably some risk, on that app that you might want to take a look at self employment fraud. So being able to take a look back at, we’ve started to see this quite often, actually, is they stated they were self employed 30 days ago. They got a decline, they submitted a new application, and now they’re a w two earner. Or maybe it wasn’t 30 days ago, maybe it was 30 minutes ago.

New synthetic fraud alerts, credit washing alerts. So this was one that was released a little bit ago. And just being able to take a look at the change in a borrower’s credit profile to alert you in the event that there’s potential credit washing, the credit that is being submitted on this application, the credit score that they have is probably not accurate. And then just overall data recycling is the same information being used over and over and over by different consumers or by the same dealership. And then we also started to kind of move into the dealer market. Right. So when we’re trying to fight fraud, we don’t just want to do it on the lender side, we want to move up further into that stream to be able to try to stop fraud further up. and so we’ve been working with dealerships to be able to try to capture better red flags, but also give them insight into employment fraud, insight into income. I know we were having a conversation at the income table a while ago. Is, do dealers perpetrate? Do they just look the other way? Do they just not know? And so by being able to use borrower check, they’re able to actually see where those red flags might be and if they might be kind of manipulating their income information. Let’s see, let me go back. There we go.

All right, so a few things kind of in the hopper, that we’re working on more in the innovation lab is trying to see how, generative AI can be leveraged in the fight against fraud. Right. Fraudsters are using it as Frank had shown. They’re using it in many different ways to create documents to, clone voices. And so we want to see if there’s other ways that it can be used to help fight those frauds. So here’s one that we’re currently kind of working on, and I can actually show a demo after when we have kind of a break, but utilizing something like a chat GPT to, be able to create narratives. To be able to assess risk from an application standpoint of just what’s being read, or, if you want to ask it questions to, look at kind of the consortium data to say, hey, has this phone number ever been seen before? And by how many people? We can be able to spit out the information back to you to say, this phone number has not been seen by this person, but we have seen it six different times for five different people. As an example, we can also use generative AI. I’ve got a demo of this as well, to be able to just look at occupation. and so this is just kind of a quick and dirty review. it’s different than our income pass tool that really does a deep dive into income. This is just taking a look at hey for what this person is stating, where they live, what they make and where they work, what is the low, median and high of, that stated occupation, stated employment in that area, and is it in line? Right?

It’s just a very quick check to make sure that, you can kind of have that litmus test, I guess, of the validity of the income, and then from a rule perspective within our case manager. So you can right now create rules based on any of the input and the output, but that is just based on a drop down menu having to kind of do some greater than, less than equal to and or statements. You can get very complex if you’d like to. But we’re starting to create a rule generator where you just ask it to say, hey, can you create me a rule where the fraud score is high, the income score is high, and there’s synthetic risk, and it’ll write out, fraud score is greater than X, income is greater than X. And here’s all the alerts you want to look at. So instead of having to sit there and try to figure out actually how to create it, you just ask it to do it for you. And boom, there it is. and that is actually all I have.

So I would love to take any questions. Would, love to talk about anything we’re doing now or have released recently. Yes, sir.

Audience Member:
Fraud bot that you were speaking about, the fraud bot. And I, believe one of the questions was, have you seen this phone number? Would, you be able to I mean, would that be able to provide the actual customer information?

Justin Davis:
No.

Audience Member:
You guys keep that, right?

Justin Davis:
Yeah. Okay. The PII level details wouldn’t be displayed. it’s more of the knowledge derived.

Audience Member:
From those details just to flag it.

Justin Davis:
Correct. It’s just like, hey, I want to do a quick check, make sure is there anything that you guys have seen that I can’t see here? Instead of reaching out to us, writing us an email, or chatting us within the case manager, it would just automatically pop it back out.

Audience Member:
Okay, thank you.

Justin Davis:
Yeah, of course. Any other questions?

Bill Hall:
I have one. It’s not necessarily sure. Maybe innovation. You did a study for us about three weeks ago, I think so, on our negative, our employer negative file.

Justin Davis:
Yes.

Bill Hall:
Can you talk a bit a little bit about what we’ve seen from the employer negative file, the amount of money that we’ve seen, and just kind of go into that a little bit m the data insights that we can provide.

Justin Davis:
So I wanted to take a look at trending fake employers from 2020 to today. I wish I would have been able to add that slide into here, but I wanted to look at a look back, because once it’s in the negative file, that’s when it’ll start to flag. Right. But what happened beforehand, we started really building it out in January 2021, December 2020. And Justin Akmuth is the one that did the majority of all of that work. But I wanted to look at how good have we gotten at identifying these fake employers?

And so to look back 2020, before we had the fake employer list, we would still flag some nine nine, nine s here and there based on other red flags. But as that list started to grow, and now we’ve reached over 10,000, we went from having almost a 0% capture of the applications and the exposure to in May of this year, we were able to flag and stop 98% of the attempts of fake employers within the consortium. So, the chart is really cool to see how the graphs start to kind of align based on what was missed to what was caught. And this year in total is about 85% of all attempts were caught and flagged. I think in total, there’s two and a half billion since 2020 attempted, and just over a billion that has actually been stopped.

Bill Hall:
Pretty impressive data set. Yeah. Justin, thank you so much.

Justin Davis:
Thank you.

Bill Hall:
Appreciate m it.

Outro:
Visit pointprodictive.com for the latest insights on lending fraud and much more from Point predictive.