Jamie Audio Episode === Edward: [00:00:00] I'm Edward [00:00:01] Cohen, and this is [00:00:02] Tangent, a Commercial [00:00:03] Observer [00:00:04] podcast. [00:00:05] [00:00:06] [00:00:07] [00:00:08] Rental [00:00:09] application fraud in [00:00:10] multifamily housing is [00:00:11] not a niche problem. [00:00:12] According to the [00:00:13] National [00:00:14] Multifamily Housing [00:00:15] Council, over [00:00:16] 93% [00:00:17] of housing [00:00:18] providers [00:00:19] experienced fraud in the [00:00:20] past year. [00:00:21] 84% of [00:00:22] them say it has [00:00:23] gotten worse over the [00:00:24] past two years.[00:00:25] [00:00:26] AI-generated document fraud [00:00:27] increased [00:00:28] 500% [00:00:29] in the second [00:00:30] half of last [00:00:31] year. [00:00:32] [00:00:33] Fraud [00:00:34] toolkits, fake [00:00:35] pay stubs, [00:00:36] fabricated bank [00:00:37] statements, and [00:00:38] synthetic Social [00:00:39] Security numbers are being [00:00:40] sold openly on [00:00:41] social media, [00:00:42] and most standard [00:00:43] screening tools are [00:00:44] catching only [00:00:45] 85% to [00:00:46] 90% of [00:00:47] consumer-grade fraud [00:00:48] and even less [00:00:49] against [00:00:50] professional-grade forgeries.[00:00:51] The [00:00:52] result shows up in [00:00:53] bad debt files, [00:00:54] evictions, [00:00:55] lost NOI, [00:00:56] and residents who [00:00:57] were never who [00:00:58] they said they [00:00:59] were. We do [00:01:00] have, in fact, the [00:01:01] perfect guest to talk about [00:01:02] this, Jamie [00:01:03] [00:01:04] Borodin, CEO of [00:01:05] DocuVerus. [00:01:06] Hi, Jamie. Where does this [00:01:07] podcast find [00:01:08] you? Jamie: I am in, [00:01:09] uh, southern part [00:01:10] of New Jersey outside of [00:01:11] Philadelphia Edward: [00:01:12] Great. So [00:01:13] Jamie, you spent [00:01:14] over 30 years [00:01:15] in resident screening [00:01:16] before founding [00:01:17] DocuVerus. [00:01:18] What did you see happening [00:01:19] in the industry that made [00:01:20] you build [00:01:21] this? Jamie: very simply [00:01:22] put, there was a lot of fraud [00:01:23] that just was going [00:01:24] undetected. The tools [00:01:25] simply didn't exist [00:01:26] to [00:01:27] identify and have [00:01:28] any effective way to [00:01:29] mitigate fraud. [00:01:30] So traditional [00:01:31] background checks were [00:01:32] centered around [00:01:33] understanding how somebody [00:01:34] pays their bills, [00:01:35] i.e., [00:01:36] consumer credit [00:01:37] analysis, [00:01:38] checking to see if they've ever [00:01:39] been to housing court [00:01:40] for any kind of [00:01:41] eviction proceeding against [00:01:42] them, [00:01:43] and, , in some [00:01:44] cases, checking for a criminal [00:01:45] background history or [00:01:46] sex offender [00:01:47] history. [00:01:48] So we were looking at [00:01:49] payment [00:01:50] proclivity, [00:01:51] behavior around [00:01:52] repayments. We [00:01:53] weren't looking at [00:01:54] ability to pay [00:01:55] effectively. and [00:01:56] [00:01:57] historically, analyzing [00:01:58] how much money somebody [00:01:59] make, and whether they [00:02:00] were who they said they [00:02:01] were, [00:02:02] was all being done at the [00:02:03] site level, and there [00:02:04] really weren't any [00:02:05] effective tools [00:02:06] to, make the [00:02:07] process [00:02:08] reliable. even [00:02:09] today, Edward, I [00:02:10] think we still [00:02:11] see [00:02:12] many, many apartment [00:02:13] communities and many, many [00:02:14] housing providers that are [00:02:15] still [00:02:16] looking at the online rental [00:02:17] [00:02:18] application. The applicant will [00:02:19] self-report their income, [00:02:20] and they are using that [00:02:21] self-reported [00:02:22] income as a proxy for [00:02:23] determining how much that [00:02:24] person actually [00:02:25] makes. some people [00:02:26] might be using [00:02:27] some form of fraud [00:02:28] detection, which can [00:02:29] catch some forms of [00:02:30] fraud, but not all [00:02:31] forms of fraud. [00:02:32] [00:02:33] More broadly, [00:02:34] they tend to [00:02:35] suffer with [00:02:36] calculation of income, [00:02:37] and that is [00:02:38] one of the areas where we [00:02:39] find that, [00:02:40] there was a hole as [00:02:41] well. And when I [00:02:42] created this service, [00:02:43] I wanted to [00:02:44] fill that gap [00:02:45] in [00:02:46] determining [00:02:47] affordability and [00:02:48] making sure that [00:02:49] when somebody [00:02:50] said they were making [00:02:51] $5,000 a month, [00:02:52] it was actually backed [00:02:53] up with the [00:02:54] documentation that proved it to [00:02:55] be valid and [00:02:56] true. and that's how [00:02:57] the software came to [00:02:58] be. [00:02:59] None of this existed [00:03:00] at the time [00:03:01] our company [00:03:02] was founded back in [00:03:03] [00:03:04] 2017. we had to [00:03:05] pretty much build everything [00:03:06] from the ground up. So we [00:03:07] built all of our own [00:03:08] validation software to be able [00:03:09] to determine [00:03:10] whether [00:03:11] s- whether a document [00:03:12] was [00:03:13] authentically cr-created, [00:03:14] whether it was edited, [00:03:15] or not.[00:03:16] And we also [00:03:17] built one of [00:03:18] the only systems, [00:03:19] and even today it's [00:03:20] still one of the only [00:03:21] systems, that can [00:03:22] automatically [00:03:23] read a [00:03:24] document's contents. [00:03:25] So when somebody [00:03:26] analyzes, [00:03:27] uploads a pay stub, we don't [00:03:28] just authenticate [00:03:29] that pay stub. We can [00:03:30] actually read [00:03:31] [00:03:32] electronically the contents [00:03:33] of the pay stub, read [00:03:34] the [00:03:35] applicant's or the employee's [00:03:36] name, read the [00:03:37] employer name, [00:03:38] read the pay start [00:03:39] date, pay end date. [00:03:40] That's how we determine [00:03:41] pay frequency. [00:03:42] We can also [00:03:43] read gross pay, we [00:03:44] can read net pay, [00:03:45] and that's how [00:03:46] we automatically [00:03:47] calculate income [00:03:48] down to the penny[00:03:49] Edward: Hmm. [00:03:50] Super interesting. I [00:03:51] mean, and, and [00:03:52] income sources [00:03:53] have also gotten [00:03:54] more [00:03:55] varied, [00:03:56] right? The rise of the [00:03:57] gig economy [00:03:58] and people [00:03:59] having multiple jobs, [00:04:00] that has only made [00:04:01] your [00:04:02] job [00:04:03] harder but [00:04:04] your company [00:04:05] more needed, [00:04:06] arguably Jamie: [00:04:07] Yeah, I don't think [00:04:08] gig economy work [00:04:09] has necessarily [00:04:10] made the process [00:04:11] of verification for [00:04:12] gig economy [00:04:13] employment [00:04:14] harder. It's, the [00:04:15] same [00:04:16] platforms that exist for [00:04:17] gig economy [00:04:18] workers [00:04:19] mirror the same kind [00:04:20] of platforms that [00:04:21] exist for W-2 [00:04:22] earners working with a [00:04:23] payroll [00:04:24] vendor. platforms like Uber and [00:04:25] Lyft and [00:04:26] DoorDash all [00:04:27] have portals [00:04:28] that their gig [00:04:29] economy workers can [00:04:30] access. [00:04:31] They just need to [00:04:32] provide the source [00:04:33] documents around that. [00:04:34] Provided they give us the source [00:04:35] documents, again, we can [00:04:36] authenticate that those [00:04:37] documents [00:04:38] came from those [00:04:39] platforms, and [00:04:40] secondarily, we can automatically [00:04:41] read the contents [00:04:42] as a proxy for [00:04:43] determining how much [00:04:44] that person actually [00:04:45] makes from that [00:04:46] source of income. Edward: [00:04:47] Interesting. [00:04:48] so industry [00:04:49] data shows that fraud [00:04:50] has gotten [00:04:51] dramatically worse [00:04:52] for [00:04:53] rental applicants in [00:04:54] multifamily over the [00:04:55] past two years. what [00:04:56] has changed and why is it [00:04:57] accelerating [00:04:58] now? Jamie: I think it [00:04:59] all started with [00:05:00] COVID. I think during [00:05:01] COVID there was lots of [00:05:02] pressure for [00:05:03] people to find [00:05:04] housing or stay in [00:05:05] housing, and, I [00:05:06] think that's when [00:05:07] the [00:05:08] rise of fraud [00:05:09] really took off. I'm [00:05:10] not gonna say that fraud [00:05:11] didn't exist prior [00:05:12] to COVID. [00:05:13] I think it really [00:05:14] exploded during [00:05:15] COVID, and it really [00:05:16] hasn't, [00:05:17] abated over [00:05:18] time. We're still [00:05:19] seeing [00:05:20] very similar rates of [00:05:21] fraud overall. [00:05:22] I think [00:05:23] fraud is a little bit [00:05:24] lower now, mostly [00:05:25] because there are tools that [00:05:26] exist to [00:05:27] mitigate the problem that didn't [00:05:28] exist four or five [00:05:29] years ago. [00:05:30] So it [00:05:31] has reduced to [00:05:32] some degree, [00:05:33] but it is [00:05:34] still a ubiquitous [00:05:35] problem that everybody sort of [00:05:36] faces, and the tools [00:05:37] have gotten [00:05:38] much more [00:05:39] sophisticated. The [00:05:40] kinds of fraud that we [00:05:41] saw four [00:05:42] years ago are very [00:05:43] different than the f- [00:05:44] forms of fraud that we see [00:05:45] now. [00:05:46] And so the tool sets have [00:05:47] have had to [00:05:48] mature and had [00:05:49] to, uh, migrate [00:05:50] with that, to make sure that we [00:05:51] stay on top of [00:05:52] all the newest trends in [00:05:53] fraud at the same [00:05:54] time. Edward: according [00:05:55] to [00:05:56] reports, [00:05:57] AI-generated document fraud [00:05:58] increased [00:05:59] 500% between [00:06:00] April and December [00:06:01] of 2025. [00:06:02] [00:06:03] So certainly, the [00:06:04] trends you mentioned since [00:06:05] COVID and now [00:06:06] also [00:06:07] AI making it [00:06:08] dramatically [00:06:09] cheaper and easier [00:06:10] to [00:06:11] falsify [00:06:12] documents [00:06:13] information is just, [00:06:14] making the problem a lot [00:06:15] worse Jamie: [00:06:16] three or four or five [00:06:17] years ago, the [00:06:18] forms of fraud [00:06:19] were, coming [00:06:20] from the [00:06:21] websites, uh, that were set [00:06:22] up by bad actors. So [00:06:23] you had to buy a [00:06:24] fake document [00:06:25] from a [00:06:26] purveyor, a [00:06:27] website that was [00:06:28] known to be able to produce [00:06:29] these documents. You didn't know [00:06:30] how to do it yourself. You [00:06:31] had to hire [00:06:32] somebody to effectively [00:06:33] do it for you, and they would [00:06:34] build and sell [00:06:35] you the documents. Or you [00:06:36] could buy editing [00:06:37] software like [00:06:38] Adobe Acrobat [00:06:39] Pro or [00:06:40] PDFiller [00:06:41] to [00:06:42] manually change [00:06:43] a PDF [00:06:44] document that somebody else gave [00:06:45] you. Now [00:06:46] you can do all this [00:06:47] using AI. So it, [00:06:48] it is somewhat [00:06:49] democratized, somewhat [00:06:50] commoditized [00:06:51] in your ability to [00:06:52] quickly and [00:06:53] accurately produce pretty [00:06:54] convincingly, [00:06:55] good but fake [00:06:56] documents. Edward: [00:06:57] So [00:06:58] what does a [00:06:59] sophisticated fake [00:07:00] pay stub or bank [00:07:01] statement actually [00:07:02] look like today, [00:07:03] and how does [00:07:04] a leasing team [00:07:05] spot it [00:07:06] without purpose-built [00:07:07] tools? [00:07:08] Or can they?[00:07:09] Jamie: They don't. [00:07:10] I, [00:07:11] I think [00:07:12] that's the [00:07:13] difference between the fraud that [00:07:14] existed four [00:07:15] years ago or five [00:07:16] years ago and the fraud that [00:07:17] exists today. [00:07:18] While some [00:07:19] people still go [00:07:20] the, [00:07:21] what we call the old [00:07:22] path of least resistance, [00:07:23] they go onto a [00:07:24] website and they buy a [00:07:25] pay stub for eight [00:07:26] dollars that is sort of [00:07:27] [00:07:28] noticeably in a [00:07:29] format that doesn't [00:07:30] really match anything [00:07:31] that a [00:07:32] legitimate payroll service [00:07:33] would [00:07:34] use. Now you [00:07:35] can produce [00:07:36] what appears to [00:07:37] be a document [00:07:38] produced by [00:07:39] ADP or ch- or [00:07:40] Paychex [00:07:41] or Dayforce, and [00:07:42] it can look [00:07:43] [00:07:44] convincingly with [00:07:45] really no [00:07:46] visual [00:07:47] discrepancies that was [00:07:48] spotted as being [00:07:49] generated. [00:07:50] So the tools [00:07:51] that some people were [00:07:52] relying on, [00:07:53] again, years ago, to [00:07:54] be able to say, "Yeah, [00:07:55] visually I could tell that [00:07:56] something was off with [00:07:57] this one." [00:07:58] You can still [00:07:59] sometimes visually see [00:08:00] things are off, but not [00:08:01] in the documents. The [00:08:02] documents now [00:08:03] are really [00:08:04] good, [00:08:05] and most people [00:08:06] don't question the [00:08:07] documents that we find [00:08:08] to be fraud these days.[00:08:09] Edward: Right. [00:08:10] Right. And [00:08:11] there's also so much to [00:08:12] review at this [00:08:13] point, right? And it's, [00:08:14] at leasing or at [00:08:15] application is, is the [00:08:16] only point [00:08:17] where these [00:08:18] things can be surfaced [00:08:19] and stopped, once a [00:08:20] bad apple is in, it, [00:08:21] generates bad debt [00:08:22] and can [00:08:23] spoil the, [00:08:24] building [00:08:25] sack Jamie: [00:08:26] 100%. One bad debt [00:08:27] case, uh, [00:08:28] certainly in the city, [00:08:29] e- even in cities [00:08:30] like Dallas, which [00:08:31] is a pretty [00:08:32] easy city to get people [00:08:33] evicted in. You don't have [00:08:34] to, incur [00:08:35] months of bad debt [00:08:36] necessarily to [00:08:37] get somebody out of [00:08:38] the unit. But [00:08:39] certainly in, c- [00:08:40] cities like Philadelphia, [00:08:41] cities like New York [00:08:42] City, cities like [00:08:43] Boston, it can be [00:08:44] six, [00:08:45] nine, 12 [00:08:46] months to get [00:08:47] somebody out of the unit. [00:08:48] So you get one [00:08:49] person in [00:08:50] that otherwise could have [00:08:51] been flagged and should have [00:08:52] been flagged on the front [00:08:53] end, [00:08:54] that can be a [00:08:55] [00:08:56] $100,000 loss. We [00:08:57] have, uh, [00:08:58] housing providers in New York [00:08:59] City that use our [00:09:00] software where [00:09:01] one bad apple can be [00:09:02] a [00:09:03] $100,000 bad debt [00:09:04] case Edward: [00:09:05] Wow. when you go into [00:09:06] a portfolio's [00:09:07] bad debt files, what [00:09:08] are you typically [00:09:09] finding? Jamie: It was very [00:09:10] interesting. That was one of the [00:09:11] things we did early [00:09:12] on, in [00:09:13] our, the evolution [00:09:14] and founding of our [00:09:15] company. We [00:09:16] were lucky enough to be able [00:09:17] to be provided [00:09:18] with lots of [00:09:19] bad debt files from [00:09:20] people that [00:09:21] had [00:09:22] qualified, moved into the [00:09:23] property, and then [00:09:24] ended up defaulting [00:09:25] on rent. [00:09:26] In one [00:09:27] case study [00:09:28] that we did, we were [00:09:29] provided about [00:09:30] 85 bad debt [00:09:31] cases in [00:09:32] just one of [00:09:33] their Mid-Atlantic [00:09:34] portfolios. [00:09:35] of those 85 [00:09:36] cases, [00:09:37] fraud was [00:09:38] definitively determined to [00:09:39] be the root cause [00:09:40] on [00:09:41] 56 of them. [00:09:42] well over [00:09:43] 50% of those bad debt [00:09:44] cases, closer to [00:09:45] 75%, were [00:09:46] all attributed [00:09:47] to undetected [00:09:48] fraud that was [00:09:49] undetected at the [00:09:50] time they applied for the [00:09:51] [00:09:52] apartment Edward: Well, [00:09:53] that's cer-certainly [00:09:54] significant and not [00:09:55] something to be, [00:09:56] deprioritized or, [00:09:57] think of as a [00:09:58] side, uh, opportunity [00:09:59] here. It's-- it should be [00:10:00] core to [00:10:01] processing rentals and [00:10:02] getting the right renters [00:10:03] in place.[00:10:04] Jamie: Absolutely[00:10:05] Edward: Or, I [00:10:06] mean, the renters that can [00:10:07] afford it, right? You also [00:10:08] don't wanna set them up for [00:10:09] failure Jamie: Yeah, and [00:10:10] I think, at the same [00:10:11] time, we have, we talk [00:10:12] about leasing teams and [00:10:13] their [00:10:14] responsibilities. at the end of [00:10:15] the day, they're [00:10:16] not really [00:10:17] hired to be private [00:10:18] investigators and [00:10:19] to look too [00:10:20] hard. They're usually [00:10:21] incentivized to fill [00:10:22] units. They are [00:10:23] pressured to fill the [00:10:24] unit. [00:10:25] So on the one [00:10:26] hand, that they want, [00:10:27] they need to fill the units. [00:10:28] On the other hand, [00:10:29] making one [00:10:30] mistake with the wrong [00:10:31] person, had you [00:10:32] simply waited another day or [00:10:33] two to get somebody who is [00:10:34] truly qualified, [00:10:35] who wasn't [00:10:36] fraudulent, would've been a much [00:10:37] better decision than trying [00:10:38] to force somebody into [00:10:39] the unit that [00:10:40] you, just appeared on [00:10:41] the surface to be good [00:10:42] today, who [00:10:43] ends up costing you [00:10:44] [00:10:45] [00:10:46] 5,000, 10,000, 50,000, $100,000 [00:10:47] later Edward: [00:10:48] To your point about [00:10:49] incentives, because [00:10:50] incentives run the [00:10:51] world, have you noticed a [00:10:52] [00:10:53] discrepancy [00:10:54] between [00:10:55] owner-operators of [00:10:56] multifamily [00:10:57] buildings [00:10:58] that work with [00:10:59] third-party [00:11:00] leasing [00:11:01] companies, versus the [00:11:02] one that have in-house [00:11:03] leasing and [00:11:04] whether [00:11:05] there tends to be [00:11:06] more, fraud, [00:11:07] when working [00:11:08] with one [00:11:09] model over the [00:11:10] other? Jamie: I think [00:11:11] fraud [00:11:12] exists across [00:11:13] both, fee managed [00:11:14] properties and [00:11:15] owner operated [00:11:16] properties. I don't [00:11:17] think the rates of fraud [00:11:18] differ. [00:11:19] I think [00:11:20] there's a [00:11:21] natural thing when [00:11:22] you own your [00:11:23] properties, you [00:11:24] probably just are [00:11:25] gonna dot the Is [00:11:26] and cross [00:11:27] [00:11:28] your [00:11:29] Ts a little more carefully than if you're a third-party [00:11:30] manager. Not to say there aren't [00:11:31] third-party management groups [00:11:32] that do very good jobs [00:11:33] and really do [00:11:34] the best jobs they [00:11:35] can, [00:11:36] but [00:11:37] fundamentally There is a [00:11:38] difference between [00:11:39] managing somebody else's [00:11:40] property where you don't bear [00:11:41] the bad debt risk on the [00:11:42] back end versus [00:11:43] an asset that you [00:11:44] own and operate [00:11:45] yourself. Edward: [00:11:46] Right, right. [00:11:47] New York City maybe [00:11:48] is a, a [00:11:49] unique market [00:11:50] as, as we all know for [00:11:51] good and for bad. [00:11:52] but I guess in New York City [00:11:53] you c- you also [00:11:54] have rental [00:11:55] buildings where you have the owner, [00:11:56] you also have the [00:11:57] third-party manager, [00:11:58] and you also have a [00:11:59] third-party [00:12:00] leasing [00:12:01] brokerage. Jamie: And the [00:12:02] brokers, you know, the funny thing [00:12:03] is we work with a lot of [00:12:04] brokers, and the [00:12:05] brokers, if, if [00:12:06] you really wanna talk [00:12:07] honestly, [00:12:08] they have very [00:12:09] little incentive [00:12:10] to identify and [00:12:11] mitigate bad debt on the [00:12:12] back end. They are [00:12:13] incentivized to fill units. They [00:12:14] make their [00:12:15] money when [00:12:16] somebody moves into a [00:12:17] property. [00:12:18] They never have a [00:12:19] disincentive around [00:12:20] bad debt that [00:12:21] happens to result on the [00:12:22] back end. [00:12:23] So New York is a [00:12:24] very different [00:12:25] market, but I think [00:12:26] that it can [00:12:27] cause some [00:12:28] complexity with [00:12:29] qualification [00:12:30] that might not [00:12:31] exist in a market [00:12:32] that doesn't share that [00:12:33] model. Edward: in [00:12:34] my early days as [00:12:35] a, as an [00:12:36] investment sales and rental [00:12:37] agent here in New York [00:12:38] City, there's no [00:12:39] incentive for [00:12:40] renewals, right? once they [00:12:41] get in, I still [00:12:42] wanna maintain the, [00:12:43] the relationship with [00:12:44] the building [00:12:45] owner. I still want them to [00:12:46] give me, the vacant [00:12:47] units to lease them [00:12:48] up, and I don't [00:12:49] wanna get bad renters [00:12:50] in. However, to your [00:12:51] point, yeah, I, [00:12:52] I get paid [00:12:53] when a [00:12:54] unit gets leased, not [00:12:55] when a [00:12:56] renter, stops [00:12:57] paying for whatever [00:12:58] reason or was [00:12:59] never able to pay in the [00:13:00] first place. [00:13:01] So [00:13:02] for, a property [00:13:03] manager running a midsize [00:13:04] portfolio, how do [00:13:05] they even know how [00:13:06] much of their [00:13:07] bad debt is [00:13:08] fraud versus just [00:13:09] non-payment? Jamie: [00:13:10] Th-they [00:13:11] don't. I think [00:13:12] one of the challenges that [00:13:13] we have is to be able [00:13:14] to convince [00:13:15] people that they [00:13:16] have a [00:13:17] problem that is attributed [00:13:18] to fraud. It's very [00:13:19] different now than it [00:13:20] was six [00:13:21] years ago, five years [00:13:22] ago. Five years [00:13:23] ago, our selling [00:13:24] cycle was [00:13:25] convincing people [00:13:26] that fraud was a [00:13:27] problem. As [00:13:28] you pointed out at the [00:13:29] beginning of this [00:13:30] podcast, people now [00:13:31] know fraud exists, [00:13:32] so fraud is not [00:13:33] this unknown [00:13:34] phenomenon. They [00:13:35] just don't always [00:13:36] know, looking [00:13:37] back at a bad debt [00:13:38] case, whether that [00:13:39] case [00:13:40] was the result of [00:13:41] undetected bad [00:13:42] debt or [00:13:43] not. a lot of [00:13:44] housing providers these days [00:13:45] do have some [00:13:46] form of fraud [00:13:47] detection technology, and [00:13:48] most of the major property [00:13:49] management systems [00:13:50] claim to [00:13:51] have some [00:13:52] form of fraud [00:13:53] detection [00:13:54] technology and [00:13:55] verification software.[00:13:56] And I [00:13:57] think that people get a [00:13:58] false sense of [00:13:59] security [00:14:00] that they, because they [00:14:01] have something through [00:14:02] their property management [00:14:03] software or they [00:14:04] happen to have [00:14:05] some solution in [00:14:06] place, that [00:14:07] that solution is going to [00:14:08] mitigate all [00:14:09] forms of fraud, and they think [00:14:10] they're covered, when in fact [00:14:11] they aren't. [00:14:12] We actually had [00:14:13] one [00:14:14] relatively significant [00:14:15] housing provider that was [00:14:16] using a competitive [00:14:17] service for fraud [00:14:18] detection. [00:14:19] they still were [00:14:20] finding that people were [00:14:21] getting in and bad debt was [00:14:22] incurring, and we [00:14:23] actually challenged them and we [00:14:24] said to them, [00:14:25] "Hey, give us [00:14:26] 10 of your [00:14:27] worst bad debt [00:14:28] cases, people [00:14:29] that owe you the most [00:14:30] money, [00:14:31] people that passed [00:14:32] the other fraud [00:14:33] detection software."[00:14:34] We [00:14:35] challenged them to give it to [00:14:36] us, let us analyze the [00:14:37] same documents that [00:14:38] passed the other [00:14:39] system. [00:14:40] In six [00:14:41] of the 10 [00:14:42] cases that that [00:14:43] provider sent to [00:14:44] us, the pay [00:14:45] stubs, in fact, [00:14:46] were fake. They went [00:14:47] undetected [00:14:48] by the f- so- [00:14:49] fraud software that [00:14:50] they thought was [00:14:51] protecting them. [00:14:52] And that's because, again, [00:14:53] these forms of [00:14:54] fraud [00:14:55] that got [00:14:56] past those other systems [00:14:57] were forms of fraud that [00:14:58] people don't even know [00:14:59] exist. [00:15:00] Fraud is way more [00:15:01] sophisticated than people [00:15:02] think it is. People [00:15:03] somehow think that [00:15:04] fraud is this [00:15:05] amorphous [00:15:06] thing. fraud is [00:15:07] complicated. It takes [00:15:08] many different shapes and [00:15:09] sizes. [00:15:10] It's constantly [00:15:11] evolving, [00:15:12] and so the [00:15:13] tool sets [00:15:14] have to become more [00:15:15] sophisticated. And [00:15:16] the problem is a [00:15:17] lot of the different [00:15:18] fraud detection [00:15:19] platforms out there, [00:15:20] [00:15:21] unlike DocuVerus, where [00:15:22] we build everything we [00:15:23] do from the ground up, [00:15:24] most of [00:15:25] these other services [00:15:26] are reselling. They're [00:15:27] white labeling [00:15:28] third-party tools over [00:15:29] which they have no [00:15:30] control. And whenever you white [00:15:31] label a [00:15:32] third-party solution, you [00:15:33] simply can't control [00:15:34] the recipes. [00:15:35] You can't [00:15:36] control [00:15:37] whether or not that [00:15:38] third-party provider actually [00:15:39] happens to know all the [00:15:40] different forms that fraud [00:15:41] takes, [00:15:42] let [00:15:43] alone whether or [00:15:44] not they know how to [00:15:45] build the recipes [00:15:46] correctly [00:15:47] to be able to mitigate [00:15:48] and detect [00:15:49] each of those [00:15:50] different fraud [00:15:51] [00:15:52] manifestations. Edward: [00:15:53] Fascinating. [00:15:54] according to [00:15:55] your report, the [00:15:56] average [00:15:57] NMHC [00:15:58] respondent [00:15:59] writes off nearly [00:16:00] $4.2 million [00:16:01] in bad [00:16:02] debt annually, [00:16:03] and approximately [00:16:04] 24% of that is [00:16:05] directly [00:16:06] attributable to [00:16:07] fraudulent applicants. Jamie: [00:16:08] I think [00:16:09] people budget for [00:16:10] bad debt, and, you know, a [00:16:11] lot of times people say to [00:16:12] us, "Well, you know, we [00:16:13] have, we have some bad [00:16:14] debt, but it, it's under [00:16:15] control." And I'll say to [00:16:16] them, "Well, [00:16:17] what's your bad debt?" [00:16:18] And they'll say, "Well, it's, you know, [00:16:19] 2%. It's [00:16:20] basically where it's [00:16:21] always been." And I'll...[00:16:22] I'm always [00:16:23] like, "Well, [00:16:24] 2%'s a lot. [00:16:25] in a [00:16:26] 10,000 unit [00:16:27] portfolio, that's [00:16:28] millions of dollars [00:16:29] of bad debt. [00:16:30] [00:16:31] And my pitch [00:16:32] is even if we [00:16:33] chop that down [00:16:34] by 10 basis points, [00:16:35] that would be [00:16:36] hundreds of [00:16:37] thousands of dollars [00:16:38] of, of [00:16:39] savings. And then, and certainly [00:16:40] we've had people that [00:16:41] have used our [00:16:42] software [00:16:43] and lowered their, their bad [00:16:44] debt by one [00:16:45] full percent.[00:16:46] Edward: Well, Jamie: [00:16:47] a monumental [00:16:48] savings when that [00:16:49] happens. I always tell [00:16:50] people you shouldn't budget for [00:16:51] bad debt. You [00:16:52] should really analyze [00:16:53] where the bad debt's [00:16:54] happening and figure out whether there's [00:16:55] a way to mitigate [00:16:56] all the [00:16:57] different [00:16:58] manifestations for it that cause [00:16:59] the bad debt in the first [00:17:00] place. Edward: [00:17:01] Right. Jamie: Don't just [00:17:02] accept it, analyze [00:17:03] it, figure out where it [00:17:04] is, and [00:17:05] then build tools to [00:17:06] mitigate it Edward: the [00:17:07] advice VCs [00:17:08] always give tech founders [00:17:09] is if there's a line item, a [00:17:10] cost line item [00:17:11] existing in a [00:17:12] property, then, like, that's [00:17:13] easier to, [00:17:14] to talk about. You know, it's [00:17:15] less, [00:17:16] imaginative in terms of the [00:17:17] ROI they can get. [00:17:18] But [00:17:19] when a lot of the operators [00:17:20] have just gotten, [00:17:21] resignated to, [00:17:22] "Oh yeah, that's a [00:17:23] line item. We know it [00:17:24] exists." and [00:17:25] yeah, sounds like they've [00:17:26] given up, whether it [00:17:27] is because they're not [00:17:28] incentivized or whether it [00:17:29] is because they [00:17:30] think the speed of [00:17:31] sophistication of [00:17:32] these frauds is something [00:17:33] that, they, it's a [00:17:34] whack-a-mole for them. [00:17:35] but yeah, it, shouldn't be [00:17:36] accepted. Jamie: [00:17:37] Right. And, and again, there is [00:17:38] a false sense of [00:17:39] security. People do think, [00:17:40] "Well, me, my property [00:17:41] management software has, [00:17:42] says they have [00:17:43] verification. They can [00:17:44] connect to somebody's [00:17:45] bank account, and that [00:17:46] protects us." Well, [00:17:47] number one, [00:17:48] bank account linking only [00:17:49] works for a [00:17:50] certain segment of people. [00:17:51] Uh, there's about [00:17:52] a 25% [00:17:53] dropout rate for [00:17:54] people that even want to [00:17:55] link, let alone the [00:17:56] fact that [00:17:57] many people just don't [00:17:58] even like the idea of [00:17:59] linking to, to begin [00:18:00] with. So [00:18:01] linking isn't [00:18:02] fraud detection. [00:18:03] Linking is [00:18:04] a facilitator for [00:18:05] verification [00:18:06] for people [00:18:07] with nothing to [00:18:08] hide, not for [00:18:09] people who are trying to [00:18:10] deceive you Edward: [00:18:11] Yep, [00:18:12] absolutely. [00:18:13] Plaid, as [00:18:14] great of an innovation it [00:18:15] is that you can connect your [00:18:16] credit card, you can [00:18:17] connect your bank [00:18:18] account, renters can [00:18:19] easily [00:18:20] deposit a few thousand [00:18:21] dollars, pass a [00:18:22] verification, [00:18:23] and remove those thousands of [00:18:24] dollars who- [00:18:25] and give them back or, [00:18:26] or something Jamie: [00:18:27] And plus, we allow [00:18:28] people to upload [00:18:29] documents, and we allow [00:18:30] people to link. So [00:18:31] we, we do support [00:18:32] linking for people that wanna [00:18:33] use it. When we [00:18:34] give people an AB [00:18:35] choice, Edward, of [00:18:36] whether or not they wanna [00:18:37] upload their [00:18:38] documents or whether or not they wanna [00:18:39] link, [00:18:40] we find that [00:18:41] 80% choose [00:18:42] documents over linking.[00:18:43] So [00:18:44] again, linking [00:18:45] is good for [00:18:46] people that want [00:18:47] to do it, but [00:18:48] the stats [00:18:49] show that over [00:18:50] 80% of [00:18:51] applicants would prefer [00:18:52] the upload method. They [00:18:53] find it to be less [00:18:54] invasive. And [00:18:55] certainly among the [00:18:56] 80% that [00:18:57] are choosing to [00:18:58] upload are all the [00:18:59] fraudsters out there. [00:19:00] Fraud doesn't live [00:19:01] in the linked [00:19:02] accounts. It lives in the [00:19:03] people that don't [00:19:04] wanna link 'cause they know they're, they [00:19:05] would be caught [00:19:06] otherwise Edward: [00:19:07] Are there [00:19:08] specific [00:19:09] geographic markets, [00:19:10] uh, within the [00:19:11] US that have higher [00:19:12] rates of [00:19:13] fraud than [00:19:14] others? Jamie: Yeah, I [00:19:15] mean, I, I always call them [00:19:16] hotspots. So the [00:19:17] hotspots are [00:19:18] cities like [00:19:19] Houston, cities like [00:19:20] Atlanta. We do [00:19:21] see more [00:19:22] fraud [00:19:23] generally [00:19:24] in, some of the [00:19:25] larger [00:19:26] Southern states for [00:19:27] some reason. [00:19:28] Not to say that [00:19:29] fraud [00:19:30] doesn't happen in New [00:19:31] York City, it certainly [00:19:32] does, it [00:19:33] just, the rents [00:19:34] tend to be [00:19:35] higher. you tend to [00:19:36] get more, a different [00:19:37] kind of demographic [00:19:38] in the work- [00:19:39] workforce. So I think the [00:19:40] percentages of [00:19:41] fraud are [00:19:42] lower in a [00:19:43] city like New York, [00:19:44] but [00:19:45] obviously the aftereffects are [00:19:46] larger because [00:19:47] of how long it takes [00:19:48] somebody to get out of the [00:19:49] unit in a city [00:19:50] like New York or [00:19:51] Philadelphia Edward: right. In one you [00:19:52] have more volume [00:19:53] maybe, in the other one you have [00:19:54] more higher ticket [00:19:55] fraud per [00:19:56] fraudster[00:19:57] Jamie: [00:19:58] Absolutely Edward: [00:19:59] So, uh, walk us [00:20:00] through what DocuVerus [00:20:01] does that [00:20:02] standard screening tools [00:20:03] don't[00:20:04] Jamie: Yeah. So [00:20:05] we, build everything [00:20:06] we do from the ground [00:20:07] up, and we do what's [00:20:08] called a [00:20:09] multidimensional metadata [00:20:10] analysis. What [00:20:11] that metadata [00:20:12] analysis [00:20:13] is, is [00:20:14] looking at really [00:20:15] what is as similar [00:20:16] to [00:20:17] an electronic [00:20:18] fingerprint of how a [00:20:19] given document was [00:20:20] created. [00:20:21] So we can [00:20:22] go really [00:20:23] layers deep into [00:20:24] the electronic [00:20:25] properties that define [00:20:26] how a given document [00:20:27] was created.[00:20:28] More [00:20:29] importantly, we [00:20:30] know how to [00:20:31] contextually [00:20:32] analyze and read [00:20:33] the [00:20:34] contents of a document. [00:20:35] So it's not just looking [00:20:36] at the, the [00:20:37] zeros and ones that [00:20:38] underlie how a [00:20:39] given document was [00:20:40] created. It's actually [00:20:41] reading on the [00:20:42] surface what that [00:20:43] pay stub or bank [00:20:44] statement is saying, [00:20:45] and that's how we [00:20:46] can identify [00:20:47] whether or not the [00:20:48] document belongs to the [00:20:49] right person. You [00:20:50] can't do that through [00:20:51] metadata. You can only do that [00:20:52] by actually [00:20:53] reading the contents [00:20:54] of the pay stub or [00:20:55] bank statement. [00:20:56] We can also [00:20:57] read, earnings [00:20:58] details. We can read [00:20:59] tax [00:21:00] details. We can read [00:21:01] withholding [00:21:02] details, and [00:21:03] that allows us to do [00:21:04] [00:21:05] contextualized [00:21:06] validations that [00:21:07] a simple [00:21:08] metadata backend [00:21:09] check would never be [00:21:10] able to do.[00:21:11] Keep in mind, in, [00:21:12] in the market [00:21:13] today, most people are just [00:21:14] analyzing the [00:21:15] zeros and ones, the [00:21:16] metadata [00:21:17] itself. We're [00:21:18] one of the only [00:21:19] companies that actually [00:21:20] mixes that, [00:21:21] but we do that at [00:21:22] a higher level as [00:21:23] well and [00:21:24] using, again, our own [00:21:25] technology. Some [00:21:26] people will try to give a [00:21:27] document [00:21:28] from, June and [00:21:29] July of [00:21:30] 2025, [00:21:31] and the [00:21:32] leasing staff might not [00:21:33] realize that that [00:21:34] is 2025, [00:21:35] not [00:21:36] 2026. [00:21:37] Our [00:21:38] software reads the dates [00:21:39] electronically, so if [00:21:40] somebody tries to pull that [00:21:41] off, that's a very [00:21:42] simple way we can [00:21:43] tell that that document [00:21:44] is not a current document. It [00:21:45] might be a valid [00:21:46] document, but our [00:21:47] software will flag [00:21:48] that because it's not [00:21:49] what we're looking [00:21:50] for in [00:21:51] August of [00:21:52] 2026. Edward: [00:21:53] how do the [00:21:54] AI [00:21:55] companies like [00:21:56] [00:21:57] Claude or [00:21:58] ChatGPT [00:21:59] fare when it [00:22:00] comes to [00:22:01] detecting [00:22:02] fraudulent [00:22:03] documents? And [00:22:04] are you [00:22:05] using them, partnering [00:22:06] with them, or yeah, [00:22:07] how do you see [00:22:08] them [00:22:09] perform in this, [00:22:10] task? Jamie: I think [00:22:11] anybody that has used [00:22:12] ChatGPT [00:22:13] or a like [00:22:14] service finds that there's a lot of [00:22:15] hallucinations that AI [00:22:16] can make. [00:22:17] AI [00:22:18] that is general [00:22:19] AI, like a [00:22:20] model like [00:22:21] ChatGPT isn't really [00:22:22] trained to be [00:22:23] discreetly good [00:22:24] at [00:22:25] detecting falsified [00:22:26] pay stubs or bank [00:22:27] statements. [00:22:28] they might be able to [00:22:29] find, [00:22:30] very limited [00:22:31] forms of fraud. They just [00:22:32] can't detect the [00:22:33] sophistication of fraud [00:22:34] that we can [00:22:35] catch. one of the things I always [00:22:36] tell people is we of [00:22:37] course use AI as [00:22:38] part of our model. What [00:22:39] makes us [00:22:40] unique is the [00:22:41] fact that we [00:22:42] curate the [00:22:43] AI [00:22:44] ourselves. We do [00:22:45] the model [00:22:46] trainings ourselves. [00:22:47] We [00:22:48] train the model [00:22:49] on what makes [00:22:50] something [00:22:51] fake. We don't just [00:22:52] feed it to the a- to [00:22:53] the AI and say, [00:22:54] "Can you tell us why [00:22:55] this is fake?" [00:22:56] We tell the [00:22:57] AI what [00:22:58] makes it fake, [00:22:59] and then we can use [00:23:00] pattern matching [00:23:01] to figure out [00:23:02] future cases of the [00:23:03] same form of fraud. [00:23:04] So it's [00:23:05] very different [00:23:06] than, we call [00:23:07] that deterministic [00:23:08] AI, [00:23:09] not what [00:23:10] ChatGPT is, which is [00:23:11] really [00:23:12] probabilistic AI. Edward: [00:23:13] Mm-hmm. this [00:23:14] is also [00:23:15] semi-breaking news as [00:23:16] it was announced [00:23:17] yesterday, and just for [00:23:18] context, we're recording [00:23:19] this on August 11th. [00:23:20] But [00:23:21] Claude announced [00:23:22] yesterday that they [00:23:23] will apply [00:23:24] invisible [00:23:25] watermarks to [00:23:26] AI text and [00:23:27] images. [00:23:28] any early thoughts on [00:23:29] this, and how do you think [00:23:30] this may impact, [00:23:31] resident, [00:23:32] application, and fraud [00:23:33] detection? Jamie: Yeah, it's [00:23:34] hard to say where [00:23:35] that's gonna go. [00:23:36] we have always [00:23:37] wished that [00:23:38] a pay stub could [00:23:39] have some form of [00:23:40] electronic [00:23:41] watermark, or a bank statement [00:23:42] could have some form of [00:23:43] watermark. It [00:23:44] would be, [00:23:45] th-there is a technical [00:23:46] term for it, [00:23:47] and if you could [00:23:48] build that into, [00:23:49] if, if [00:23:50] banks and [00:23:51] payroll services [00:23:52] could incorporate [00:23:53] those into their [00:23:54] documents, it [00:23:55] would be [00:23:56] another [00:23:57] way to be able to [00:23:58] mitigate [00:23:59] and not just the [00:24:00] problem of fraud, but also [00:24:01] really a [00:24:02] facilitator for verification in [00:24:03] general. I [00:24:04] think people think that [00:24:05] linking to your [00:24:06] payroll or bank account [00:24:07] is the way to do [00:24:08] it. The problem [00:24:09] is, the fraudsters [00:24:10] are not trying to [00:24:11] cooperate with you. [00:24:12] They're trying to [00:24:13] figure out ways around your [00:24:14] system, not ways to [00:24:15] cooperate with your [00:24:16] system. So [00:24:17] we're always [00:24:18] trying to think about [00:24:19] the [00:24:20] worst-case [00:24:21] scenario in how we [00:24:22] build our software [00:24:23] to [00:24:24] mitigate. The other thing that we're [00:24:25] really focused on [00:24:26] now is Social Security [00:24:27] fraud. So [00:24:28] b-beyond just [00:24:29] faking [00:24:30] your pay [00:24:31] stubs or your bank statements [00:24:32] as a means to be able to [00:24:33] say that you make more money [00:24:34] than you claim to [00:24:35] or make money in [00:24:36] general that you just [00:24:37] don't make, [00:24:38] people [00:24:39] obviously sometimes have [00:24:40] bad credit that they need to get [00:24:41] around, or they might have [00:24:42] history of eviction that they [00:24:43] need to get around. And the [00:24:44] easiest way to do that [00:24:45] isn't by faking your [00:24:46] ID, [00:24:47] it's by faking your [00:24:48] Social Security number. So [00:24:49] if you apply with [00:24:50] a [00:24:51] fabricated Social [00:24:52] Security number or a, [00:24:53] CPN, [00:24:54] you can [00:24:55] appear to have [00:24:56] credit that is good [00:24:57] and positive when in [00:24:58] fact that is [00:24:59] completely fabricated [00:25:00] credit file. [00:25:01] So [00:25:02] we have [00:25:03] really focused a [00:25:04] lot, over the last couple [00:25:05] years of building out a [00:25:06] model [00:25:07] to [00:25:08] really detect Social [00:25:09] Security fraud, and we [00:25:10] are really the [00:25:11] only service that [00:25:12] has a [00:25:13] visual-based way to [00:25:14] detect and [00:25:15] find [00:25:16] cases of Social [00:25:17] Security [00:25:18] fabrication.[00:25:19] We also are one of the [00:25:20] only companies that has [00:25:21] a connection to the [00:25:22] Social Security [00:25:23] Administration so that [00:25:24] we can actually check to [00:25:25] see if the Social that [00:25:26] you report to be [00:25:27] yours was in [00:25:28] fact issued by the Social [00:25:29] Security [00:25:30] Administration[00:25:31] Edward: Wow, okay. that [00:25:32] seems like a key [00:25:33] differentiator. what's a [00:25:34] CPN for those who [00:25:35] aren't aware? Jamie: [00:25:36] it's called a, credit [00:25:37] privacy number. A [00:25:38] credit privacy [00:25:39] number s- sounds [00:25:40] fancy, actually sounds [00:25:41] legitimate. It is a [00:25:42] made-up [00:25:43] concept, but it is a [00:25:44] concept [00:25:45] that, a credit privacy [00:25:46] number looks like a [00:25:47] Social Security [00:25:48] number. It smells like a [00:25:49] Social Security [00:25:50] number, but it's a [00:25:51] fabricated Social [00:25:52] Security number. And [00:25:53] bad actors [00:25:54] who [00:25:55] need to be able to [00:25:56] create a credit file for [00:25:57] you that is [00:25:58] positive and shows a high [00:25:59] credit score, bad [00:26:00] actors out there [00:26:01] exist that will [00:26:02] set you up with a [00:26:03] CPN. [00:26:04] They will build a credit [00:26:05] file around that [00:26:06] CPN [00:26:07] that shows you as [00:26:08] having very positive [00:26:09] credit, when in fact it's all [00:26:10] been fabricated[00:26:11] Edward: [00:26:12] Well, people are so [00:26:13] creative. [00:26:14] Unfortunately, the [00:26:15] fraudsters are also [00:26:16] creative Jamie: [00:26:17] And so in other words, you [00:26:18] can literally [00:26:19] apply through a [00:26:20] CPA and you can [00:26:21] appear to have a 780 [00:26:22] or [00:26:23] 820 credit score, [00:26:24] which [00:26:25] was entirely [00:26:26] fabricated people [00:26:27] have a false sense of [00:26:28] security. They see a credit [00:26:29] score and they see a [00:26:30] 780 score and they [00:26:31] go, "Wow, person's [00:26:32] great. Very high credit [00:26:33] score, low-risk [00:26:34] person." they don't know how [00:26:35] to see the [00:26:36] signs that that [00:26:37] credit file itself [00:26:38] on which that score [00:26:39] was based [00:26:40] was completely [00:26:41] fabricated Edward: [00:26:42] Yeah, the c- the [00:26:43] credit scoring system [00:26:44] is, is long [00:26:45] overdue for an [00:26:46] overhaul, but that's probably [00:26:47] a, a whole [00:26:48] other podcast [00:26:49] episode Jamie: It's a whole [00:26:50] nother podcast, [00:26:51] and this is [00:26:52] not even a problem with the [00:26:53] algorithms used to [00:26:54] determine a score. This is [00:26:55] just based on the fact [00:26:56] that the credit was, [00:26:57] in fact, [00:26:58] manufactured. [00:26:59] And so [00:27:00] if you look at the credit file, [00:27:01] and even if you weren't [00:27:02] using a credit score, [00:27:03] you'd still think the [00:27:04] person was low [00:27:05] risk [00:27:06] because the [00:27:07] credit, [00:27:08] lines of credit that [00:27:09] underlie that credit [00:27:10] file [00:27:11] are all, show a [00:27:12] positive payment [00:27:13] history and a [00:27:14] well-established [00:27:15] payment history. But [00:27:16] you'll notice [00:27:17] that the credit [00:27:18] file is an authorized [00:27:19] user account. [00:27:20] So what ended up [00:27:21] happening is, [00:27:22] and I'll pick on you, [00:27:23] Edward, just for the sake of [00:27:24] [00:27:25] demonstration, when a [00:27:26] CPN is established [00:27:27] for Edward Cohen, it [00:27:28] might be tied [00:27:29] to Jamie's [00:27:30] credit line. [00:27:31] And what happens is [00:27:32] that, [00:27:33] purveyor, that bad [00:27:34] actor will [00:27:35] effectively [00:27:36] make Edward an [00:27:37] authorized user on [00:27:38] Jamie's credit [00:27:39] line, and that's how you [00:27:40] show as having a [00:27:41] strong line of credit [00:27:42] on that credit [00:27:43] line. It's not actually your line of [00:27:44] credit, but [00:27:45] because you're tied to it, [00:27:46] you get the [00:27:47] benefit from it, and [00:27:48] the scoring models [00:27:49] know that that's [00:27:50] the case Edward: [00:27:51] Well, thank you Jamie for [00:27:52] lending me your, uh, credit [00:27:53] line. [00:27:54] so, uh, let's move [00:27:55] on to [00:27:56] regulation compliance and where [00:27:57] the industry's [00:27:58] heading. Florida just [00:27:59] made rental [00:28:00] application fraud a [00:28:01] third-degree felony [00:28:02] starting in October of [00:28:03] this year. So [00:28:04] what does that [00:28:05] change practically [00:28:06] for multifamily [00:28:07] operators?[00:28:08] Jamie: I don't think a [00:28:09] ton. I think [00:28:10] it means that in the [00:28:11] state of Florida, you [00:28:12] might have a [00:28:13] much more [00:28:14] cooperative law [00:28:15] enforcement apparatus that [00:28:16] is gonna stand behind you [00:28:17] if you, fall [00:28:18] victim to fraud, [00:28:19] and they might [00:28:20] do more to try to [00:28:21] find the [00:28:22] propagator. Fraud is [00:28:23] not legal. it's [00:28:24] illegal everywhere. It's [00:28:25] not like you are [00:28:26] committing fraud. Fraud [00:28:27] itself is a [00:28:28] crime. [00:28:29] it's just that [00:28:30] nobody-- the law [00:28:31] enforcement agencies [00:28:32] throughout the rest of the [00:28:33] country just don't have the [00:28:34] manpower or the [00:28:35] resources [00:28:36] to [00:28:37] investigate [00:28:38] and do it. [00:28:39] So a lot of the [00:28:40] bad actors that [00:28:41] propagate fraud [00:28:42] just aren't held to [00:28:43] being held to the [00:28:44] account [00:28:45] because law enforcement doesn't [00:28:46] have the time and [00:28:47] resources to pursue [00:28:48] it and to [00:28:49] prosecute it.[00:28:50] Florida's [00:28:51] basically saying, [00:28:52] theoretically, we [00:28:53] have to see if this actually [00:28:54] bears out in, [00:28:55] in action, [00:28:56] but [00:28:57] theoretically, Florida's gonna [00:28:58] take the crime more [00:28:59] seriously and dedicate [00:29:00] more resources to [00:29:01] it. Edward: [00:29:02] Hmm. Jamie: But keep in mind, I [00:29:03] think if you have a [00:29:04] police force that is [00:29:05] dealing with [00:29:06] [00:29:07] kidnapping, they're gonna treat [00:29:08] kidnapping and [00:29:09] drug offenses [00:29:10] much more [00:29:11] heavily than somebody [00:29:12] claiming that [00:29:13] somebody came in [00:29:14] with fake pay stubs. Edward: [00:29:15] [00:29:16] so [00:29:17] screening rules are also [00:29:18] shifting at the federal, [00:29:19] state, and local [00:29:20] levels around credit [00:29:21] and eviction [00:29:22] history and criminal records. [00:29:23] So how [00:29:24] do multifamily [00:29:25] operators stay [00:29:26] compliant while also [00:29:27] protecting themselves from [00:29:28] fraud? Jamie: [00:29:29] I think [00:29:30] everybody thinks [00:29:31] that their [00:29:32] screening [00:29:33] provider [00:29:34] has them covered, and [00:29:35] they don't [00:29:36] realize that the screening [00:29:37] providers, and [00:29:38] with almost all the [00:29:39] laws that exist [00:29:40] that limit [00:29:41] use of criminal [00:29:42] information, [00:29:43] limit use [00:29:44] of eviction [00:29:45] information, limit use [00:29:46] of credit scoring [00:29:47] or, [00:29:48] or the like, they [00:29:49] don't understand that [00:29:50] the screening [00:29:51] company doesn't have to be [00:29:52] compliant.[00:29:53] They are the [00:29:54] ones that have to be [00:29:55] compliant. [00:29:56] And so they think that [00:29:57] because the screening [00:29:58] company says that they are [00:29:59] delivering a compliant [00:30:00] report, They think [00:30:01] they're covered when in [00:30:02] fact they're not. to [00:30:03] give you an example, [00:30:04] Philadelphia, [00:30:05] limits [00:30:06] eviction [00:30:07] record [00:30:08] handling. [00:30:09] So under federal law, [00:30:10] you can look back at [00:30:11] eviction records going [00:30:12] back seven years, and you can [00:30:13] take into account [00:30:14] not just evictions, [00:30:15] [00:30:16] actual people that [00:30:17] left owing money, [00:30:18] but you can take into [00:30:19] account [00:30:20] even cases that were [00:30:21] dismissed. [00:30:22] In [00:30:23] Philadelphia, number one, that [00:30:24] seven-year look back is [00:30:25] reduced to four [00:30:26] years, so anything [00:30:27] over f- four [00:30:28] years simply cannot be [00:30:29] considered at [00:30:30] all. [00:30:31] Secondarily, in [00:30:32] Philadelphia, you can [00:30:33] only [00:30:34] consider an [00:30:35] eviction that [00:30:36] was granted in [00:30:37] favor of the [00:30:38] [00:30:39] landlord. and again, if there was any [00:30:40] kind of settlement [00:30:41] agreement, [00:30:42] that is considered [00:30:43] in favor [00:30:44] of the tenant, not [00:30:45] in favor of the landlord, [00:30:46] and hence you can't [00:30:47] count that. [00:30:48] The only way to really [00:30:49] know that is by looking up [00:30:50] the court docket.[00:30:51] Most screening [00:30:52] companies would never [00:30:53] look up the court docket, [00:30:54] hence they're never gonna be [00:30:55] compliant with that [00:30:56] Philadelphia law. [00:30:57] So people [00:30:58] think that they, that their [00:30:59] screening company [00:31:00] might have them [00:31:01] covered in [00:31:02] Philadelphia. They [00:31:03] don't realize that [00:31:04] the city of Philadelphia [00:31:05] law [00:31:06] is [00:31:07] borne upon the [00:31:08] housing provider [00:31:09] to be compliant [00:31:10] with, not the [00:31:11] screening company they [00:31:12] use. And so the [00:31:13] screening companies are [00:31:14] still gonna report evictions [00:31:15] going back [00:31:16] seven years. [00:31:17] That [00:31:18] housing provider has to [00:31:19] remember, "Oh, I can [00:31:20] only go back four years, [00:31:21] and oh, [00:31:22] I have to look up that [00:31:23] court case [00:31:24] to see that there was no [00:31:25] settlement agreement [00:31:26] made and to see [00:31:27] that it actually was [00:31:28] a judgment [00:31:29] in favor of the [00:31:30] landlord that wasn't [00:31:31] later [00:31:32] settled." Edward: fascinating and [00:31:33] just reminds us [00:31:34] how real [00:31:35] estate is, always a, [00:31:36] a local, [00:31:37] business, a [00:31:38] local issues, [00:31:39] local risks, [00:31:40] local benefits, [00:31:41] and so you guys [00:31:42] cover the entire United [00:31:43] States? Jamie: We [00:31:44] do. and again, we're [00:31:45] not really focused on the [00:31:46] screening piece. We have [00:31:47] partners that [00:31:48] do background screening [00:31:49] because I come [00:31:50] from background screening. [00:31:51] That's what I did before I [00:31:52] started at NTN. I [00:31:53] happen to know a lot [00:31:54] about consumer [00:31:55] credit. I know a lot about [00:31:56] the laws that affect [00:31:57] it, [00:31:58] and we [00:31:59] specialized in eviction and [00:32:00] criminal data, so I know a l- [00:32:01] I know a lot about the [00:32:02] rules that exist. And [00:32:03] frankly, I think [00:32:04] it makes a service [00:32:05] like ours even more [00:32:06] important in the year [00:32:07] 2026 when there [00:32:08] are so many cities and [00:32:09] states that are [00:32:10] seeking [00:32:11] to pass [00:32:12] regulation that [00:32:13] limits, how eviction [00:32:14] records are used [00:32:15] or how, criminal [00:32:16] background records are [00:32:17] used. [00:32:18] laws that didn't exist [00:32:19] 20 years ago, [00:32:20] they are [00:32:21] passing at [00:32:22] a relatively [00:32:23] alarming sta- [00:32:24] alarming rate these [00:32:25] days. New Jersey [00:32:26] has relatively [00:32:27] strict prohibitions [00:32:28] around when you can [00:32:29] run criminal [00:32:30] background. Can't do [00:32:31] criminal background in New Jersey [00:32:32] until you do what's [00:32:33] called a preliminary [00:32:34] notice of [00:32:35] approval, which means you have to fully [00:32:36] qualify the [00:32:37] applicant on [00:32:38] income, on [00:32:39] credit, on [00:32:40] eviction history [00:32:41] first, [00:32:42] make an offer of [00:32:43] qualification to that [00:32:44] person. Only [00:32:45] once you do that [00:32:46] can you actually run [00:32:47] criminal. Edward: [00:32:48] Wow! Jamie: A- and then [00:32:49] New Jersey [00:32:50] prescribes how far you [00:32:51] can look back based on [00:32:52] degree level of [00:32:53] crime. So it's a [00:32:54] very [00:32:55] complicated law to [00:32:56] adhere to, [00:32:57] both in knowing when [00:32:58] you can run [00:32:59] criminal and what you [00:33:00] can and can't deny [00:33:01] for. states like [00:33:02] New York, you cannot [00:33:03] even... [00:33:04] Eviction information is [00:33:05] [00:33:06] practically barred [00:33:07] in New York. [00:33:08] You cannot [00:33:09] consider eviction records [00:33:10] in New York, [00:33:11] which makes [00:33:12] income [00:33:13] verification and [00:33:14] fraud detection [00:33:15] even more [00:33:16] important in a market like [00:33:17] New York [00:33:18] because fundamentally they've [00:33:19] limited your [00:33:20] ability to use [00:33:21] the existing [00:33:22] screening tools in [00:33:23] those [00:33:24] markets. Edward: [00:33:25] fascinating and mind-blowing [00:33:26] at the same time. [00:33:27] so where does fraud [00:33:28] prevention go from [00:33:29] here? I mean, we have [00:33:30] generative AI that keeps [00:33:31] lowering the barrier [00:33:32] to [00:33:33] more sophisticated [00:33:34] document fabrication. [00:33:35] So where does [00:33:36] fraud prevention go from [00:33:37] here? Jamie: a [00:33:38] lot of it is [00:33:39] context-driven, so [00:33:40] AI can [00:33:41] generate pretty [00:33:42] [00:33:43] convincing metadata [00:33:44] validations [00:33:45] that'll [00:33:46] pass very [00:33:47] simplified [00:33:48] metadata checks. [00:33:49] where the AI model [00:33:50] suffer is, you [00:33:51] have to get a lot [00:33:52] right, in order to get [00:33:53] somebody [00:33:54] to pass a fraud [00:33:55] check. and we [00:33:56] do a lot with [00:33:57] [00:33:58] contextualization. So again, when we [00:33:59] see [00:34:00] somebody that's [00:34:01] 20 years old and they're [00:34:02] making [00:34:03] $100,000 a [00:34:04] year, it's not [00:34:05] impossible that that [00:34:06] person actually is [00:34:07] making $100,000 a [00:34:08] year. It's one of those [00:34:09] things that [00:34:10] the algorithm [00:34:11] knows to look a little [00:34:12] more closely at. So [00:34:13] a lot of it is [00:34:14] context-driven [00:34:15] just around, again, [00:34:16] we're never gonna [00:34:17] discriminate based [00:34:18] on age, but we're gonna [00:34:19] maybe look at [00:34:20] source of income [00:34:21] just to make [00:34:22] sure it all makes [00:34:23] sense. we also look [00:34:24] at, uh, to the [00:34:25] degree we know what [00:34:26] somebody's credit history looks [00:34:27] like, we can see if you have a [00:34:28] new credit file. [00:34:29] So [00:34:30] if Edward is [00:34:31] applying and we know [00:34:32] Edward's 45 years old, [00:34:33] and Edward has a credit [00:34:34] file that was [00:34:35] established last year, [00:34:36] we're gonna scratch [00:34:37] our head and go, "Why [00:34:38] does Edward have a [00:34:39] new credit file in [00:34:40] [00:34:41] 2025 when [00:34:42] Edward should have a [00:34:43] credit file that goes [00:34:44] back into the [00:34:45] 1990s or the [00:34:46] year 2000?" [00:34:47] So that's [00:34:48] the kind of [00:34:49] thing that [00:34:50] contextualization allows [00:34:51] for. if we're [00:34:52] just relying on [00:34:53] rudimentary metadata [00:34:54] checks, we [00:34:55] wouldn't necessarily [00:34:56] find fraud in [00:34:57] that. But the [00:34:58] contextual picture [00:34:59] of all the different [00:35:00] pieces of [00:35:01] information [00:35:02] allows us to see [00:35:03] patterns of fraud that [00:35:04] wouldn't exist [00:35:05] otherwise Edward: [00:35:06] Last but not [00:35:07] least, [00:35:08] collaboration superpower. [00:35:09] If you could [00:35:10] partner with one person, [00:35:11] historic or [00:35:12] living, real [00:35:13] or fictional, [00:35:14] who would it be? Jamie: [00:35:15] Oh God. [00:35:16] Um, probably somebody [00:35:17] like Einstein who [00:35:18] just [00:35:19] is seeing [00:35:20] levels [00:35:21] of [00:35:22] facts that are [00:35:23] invisible to [00:35:24] almost everybody [00:35:25] else. Einstein is [00:35:26] a fascinating [00:35:27] character 'cause he just saw [00:35:28] things that nobody else [00:35:29] saw, he [00:35:30] wasn't seeing things that were [00:35:31] fabricated. He saw, [00:35:32] he was seeing [00:35:33] facts and [00:35:34] realities that [00:35:35] before him nobody else [00:35:36] knew to even look [00:35:37] for. [00:35:38] Those people are the [00:35:39] people that fascinate [00:35:40] me. Those are the [00:35:41] people that I wanna be aligned [00:35:42] with Edward: [00:35:43] If, Einstein was alive [00:35:44] today, I'm sure he would be [00:35:45] very [00:35:46] passionate about fraud [00:35:47] detection ' [00:35:48] [00:35:49] so today we covered [00:35:50] why rental [00:35:51] application fraud is one of the [00:35:52] fastest growing [00:35:53] operational risks in [00:35:54] multifamily, and [00:35:55] why the tools most [00:35:56] operators rely on are [00:35:57] not necessarily [00:35:58] built for the fraud [00:35:59] environment [00:36:00] they are operating [00:36:01] in right now.[00:36:02] We also talked [00:36:03] about what [00:36:04] AI-generated documents [00:36:05] actually look like, why [00:36:06] synthetic [00:36:07] identity fraud is so hard [00:36:08] to catch with [00:36:09] standard screening, we [00:36:10] also got into how [00:36:11] DocuVerus approaches the [00:36:12] accuracy gap, [00:36:13] and some of [00:36:14] the regional [00:36:15] intricacies of [00:36:16] fraud [00:36:17] de-prevention, and also [00:36:18] where [00:36:19] the regulation is [00:36:20] going. [00:36:21] Jamie, [00:36:22] thank you so much for coming to [00:36:23] Tanium today. Jamie: Thanks, [00:36:24] Edward. Glad to be [00:36:25] here [00:36:26] [00:36:27] [00:36:28] [00:36:29] Edward: Muchas [00:36:30] gracias. If [00:36:31] you learn something [00:36:32] new or enjoy the [00:36:33] conversation. [00:36:34] Text a friend the link [00:36:35] to this episode [00:36:36] right now. [00:36:37] [00:36:38] You can find links, [00:36:39] resources, and [00:36:40] ways to connect with us [00:36:41] in the show [00:36:42] notes [00:36:43] [00:36:44] [00:36:45] [00:36:46] [00:36:47]