Auphonic Audio - Dr. Kate Jarvis ===
[00:00:00] [00:00:01] Hi, I'm Edward [00:00:02] Cohen, and this is [00:00:03] Tangent, a [00:00:04] Commercial Observer [00:00:05] podcast, [00:00:06] [00:00:07] [00:00:08] [00:00:09] coming to you [00:00:10] live from CREtech [00:00:11] New York at [00:00:12] the Javits Center, [00:00:13] the New York City [00:00:14] Real Estate Week. [00:00:15] [00:00:16] Real estate [00:00:17] investment firms make [00:00:18] billion-dollar [00:00:19] decisions on [00:00:20] data that lives in [00:00:21] spreadsheets, [00:00:22] email threads, [00:00:23] PDFs, and [00:00:24] attachments.
[00:00:25] The [00:00:26] intelligence layer has [00:00:27] never existed [00:00:28] until AI came to the [00:00:29] scene. [00:00:30] [00:00:31] Our guest [00:00:32] today is Dr. [00:00:33] Kate Jarvis, [00:00:34] CEO and [00:00:35] co-founder of Fifth [00:00:36] Dimension. [00:00:37] Hi, Kate. How is [00:00:38] New York City Real Estate Week [00:00:39] treating you? [00:00:40] Hello. Thank you for having [00:00:41] me. I think, [00:00:42] as you can tell, there's [00:00:43] a lot of buzz at the [00:00:44] conference here [00:00:45] today, and [00:00:46] I think [00:00:47] obviously New York is one [00:00:48] of the financial [00:00:49] centers of the world.[00:00:50]
I walk around [00:00:51] this city thinking, [00:00:52] "Ah, what an [00:00:53] enormous city built [00:00:54] on debt." [00:00:55] Because that's what I [00:00:56] deal with at Fifth [00:00:57] Dimension, is everyone's [00:00:58] documents and data, [00:00:59] delivering insights [00:01:00] and actions [00:01:01] off the back of them [00:01:02] using agentic [00:01:03] AI. And it's [00:01:04] just amazing to [00:01:05] me that this world in [00:01:06] which we live, [00:01:07] work, and [00:01:08] play has such a [00:01:09] complicated [00:01:10] set of data, [00:01:11] stakeholders, et [00:01:12] cetera, against [00:01:13] it.
And before [00:01:14] I got into this industry [00:01:15] about a decade ago, I [00:01:16] never really thought about [00:01:17] it. But just even the [00:01:18] Javits Center, [00:01:19] right? The Javits Center [00:01:20] itself is a [00:01:21] phenomenal [00:01:22] development, and the way [00:01:23] in which it's used and [00:01:24] commercialized today, we [00:01:25] couldn't be here without [00:01:26] that. [00:01:27] Absolutely. New York City is a [00:01:28] massive playground [00:01:29] of data, [00:01:30] real estate, and, [00:01:31] uh, financial [00:01:32] potential.
[00:01:33] Exactly. [00:01:34] So Fifth Dimension, I [00:01:35] mean, you guys have [00:01:36] been on fire [00:01:37] lately. You raised [00:01:38] $26 million [00:01:39] earlier this year, and [00:01:40] you're working with [00:01:41] firms like Boston [00:01:42] Property Group, [00:01:43] Realty Income, and [00:01:44] Peachtree Group. [00:01:45] Tell us about the [00:01:46] company and the [00:01:47] teams and the firms [00:01:48] that you're building for and [00:01:49] working with.[00:01:50]
Yeah, so [00:01:51] Fifth Dimension [00:01:52] is for investment [00:01:53] managers, asset [00:01:54] managers, and [00:01:55] brokers who are [00:01:56] looking for [00:01:57] a new way of [00:01:58] doing business. [00:01:59] As you mentioned, [00:02:00] this industry [00:02:01] runs on [00:02:02] data, and there have been [00:02:03] many decades of [00:02:04] innovation, many [00:02:05] decades of [00:02:06] attempts at [00:02:07] turning those data [00:02:08] into a commercial [00:02:09] advantage. So [00:02:10] we have, the data [00:02:11] dashboard era in [00:02:12] the 2010s.
[00:02:13] We have our [00:02:14] ERP systems. We [00:02:15] have our CRMs. [00:02:16] There's [00:02:17] software, there's [00:02:18] documents, there's Excel [00:02:19] spreadsheets, there's [00:02:20] PDFs, and all [00:02:21] of it doesn't get [00:02:22] leveraged in the way [00:02:23] that is best for these [00:02:24] businesses. [00:02:25] What you really [00:02:26] want, is for [00:02:27] the way that you do [00:02:28] business [00:02:29] to be [00:02:30] defined [00:02:31] by your data. Your [00:02:32] data are your [00:02:33] alpha.
Your data [00:02:34] inform your buy box, [00:02:35] your data inform [00:02:36] which investments you make, [00:02:37] when you might offload [00:02:38] an asset, [00:02:39] and we're the [00:02:40] platform that helps you see [00:02:41] around corners [00:02:42] in order to make [00:02:43] better decisions. [00:02:44] So in working [00:02:45] with some of the businesses that [00:02:46] you mentioned, [00:02:47] we tend to [00:02:48] have out of the [00:02:49] box a [00:02:50] platform where there's a [00:02:51] workspace, which is a [00:02:52] type of database, [00:02:53] and a set of [00:02:54] out-of-the-box AI [00:02:55] agents and [00:02:56] data connectors [00:02:57] available to run, [00:02:58] and we call that the [00:02:59] playbook.
You [00:03:00] have the playbook [00:03:01] for [00:03:02] acquisitions, which means every [00:03:03] workspace might be a [00:03:04] deal. And we might [00:03:05] have agents within [00:03:06] that workspace that go [00:03:07] into your Outlook [00:03:08] inbox, find all [00:03:09] of the offering memoranda [00:03:10] and property brochures [00:03:11] that have been sent to you over [00:03:12] the, the weekend, [00:03:13] extract data [00:03:14] from them, measure [00:03:15] those against your investment [00:03:16] platform, [00:03:17] and serve those data to [00:03:18] you in a beautiful [00:03:19] spreadsheet ready for your [00:03:20] analysis on [00:03:21] Monday morning.
[00:03:22] So we're also not [00:03:23] cutting the human out of the [00:03:24] loop, and I think [00:03:25] that's really, really [00:03:26] critical. Human [00:03:27] judgment is [00:03:28] also what makes this [00:03:29] industry run. [00:03:30] You can't hand it [00:03:31] over, hand over the keys [00:03:32] to the house just to a [00:03:33] computer. [00:03:34] And so it's really important to [00:03:35] us that people spend [00:03:36] their time doing what they're [00:03:37] best at, which is [00:03:38] using their judgment, [00:03:39] and all of [00:03:40] this data crunching to [00:03:41] figure out how to use [00:03:42] past transactions [00:03:43] to inform [00:03:44] future transactions [00:03:45] should be left to the computer.[00:03:46]
It is a [00:03:47] fascinating time, [00:03:48] especially when it comes [00:03:49] to that [00:03:50] collaboration [00:03:51] between humans and [00:03:52] technology, if you can [00:03:53] call it [00:03:54] that. It seems like you fall [00:03:55] somewhere in the middle between [00:03:56] the spectrum of, I [00:03:57] mean, we, we hear [00:03:58] now [00:03:59] CEOs of tech [00:04:00] companies telling [00:04:01] a very different [00:04:02] message in terms of what [00:04:03] AI is doing, [00:04:04] whether [00:04:05] AI is [00:04:06] [00:04:07] replacing humans at [00:04:08] their jobs, and [00:04:09] that's why you should hire [00:04:10] them, [00:04:11] their technology, [00:04:12] or is it [00:04:13] AI enhancing [00:04:14] those roles [00:04:15] until it replaces [00:04:16] them, or will it be [00:04:17] a constant [00:04:18] collaboration and evolution [00:04:19] of the roles and the [00:04:20] functions?
It seems like you [00:04:21] fall somewhere in the middle [00:04:22] where [00:04:23] Alienating the [00:04:24] human [00:04:25] will only [00:04:26] stifle innovation or [00:04:27] stifle the adoption, the [00:04:28] pace of adoption. [00:04:29] I think that's [00:04:30] exactly right. [00:04:31] AI should [00:04:32] work to the [00:04:33] best of human [00:04:34] capability and [00:04:35] augment, with [00:04:36] computer science, with the [00:04:37] ability to look [00:04:38] across billions of [00:04:39] data points [00:04:40] and quickly come up with a [00:04:41] recommendation for an [00:04:42] investment.
But you're [00:04:43] never gonna have [00:04:44] one click, "I [00:04:45] wanna purchase this [00:04:46] asset based on [00:04:47] the information that my [00:04:48] AI has told [00:04:49] me." And I think [00:04:50] that's important [00:04:51] because the future [00:04:52] is going to be [00:04:53] about, yes, [00:04:54] smaller teams that [00:04:55] can manage [00:04:56] infinitely more money than they [00:04:57] can today [00:04:58] because we are [00:04:59] replacing the manual [00:05:00] labor, the [00:05:01] person-hours that are sitting [00:05:02] behind that [00:05:03] analysis.
But what I [00:05:04] think that actually [00:05:05] enables is [00:05:06] kind of full [00:05:07] stack teams. [00:05:08] You can't spend your [00:05:09] day as an analyst [00:05:10] anymore getting [00:05:11] coffee and updating [00:05:12] waterfalls. You can [00:05:13] actually sit on the [00:05:14] investment committee [00:05:15] because you [00:05:16] yourself have that [00:05:17] first-hand access to what's [00:05:18] happening [00:05:19] across the [00:05:20] marketplace within your own [00:05:21] transactions [00:05:22] history, and you can use [00:05:23] that to develop a [00:05:24] very informed [00:05:25] opinion quickly on [00:05:26] what the best decision [00:05:27] is.
And in [00:05:28] past, that just was [00:05:29] impossible, right? You would have to [00:05:30] spend five, seven, [00:05:31] 10, 12 [00:05:32] years training up [00:05:33] in order to develop an [00:05:34] informed opinion. [00:05:35] So I think the beauty of [00:05:36] AI is actually that it [00:05:37] could be leverage for [00:05:38] education, and [00:05:39] I hate all of this [00:05:40] talk about how [00:05:41] chatbots just make people [00:05:42] lazy. They only [00:05:43] make people lazy if [00:05:44] you use them in the wrong [00:05:45] way, if you just [00:05:46] trust what the [00:05:47] answer is from Claude [00:05:48] about the valuation of a [00:05:49] building without [00:05:50] checking the answer, [00:05:51] absolutely.
But if you [00:05:52] use technology [00:05:53] that can cite its [00:05:54] data, have a full [00:05:55] auditability trail that [00:05:56] goes and gives you the [00:05:57] calculation [00:05:58] for the internal rate [00:05:59] of return [00:06:00] and all of the [00:06:01] analysis that sits behind [00:06:02] it, it actually becomes a [00:06:03] really powerful [00:06:04] educational tool for [00:06:05] people who are just starting out in [00:06:06] the industry too [00:06:07] Seems like you [00:06:08] have strong opinions about [00:06:09] chatbots.
just [00:06:10] before this interview, y- [00:06:11] we were talking about [00:06:12] how people [00:06:13] are claiming SaaS is [00:06:14] dead, but [00:06:15] you're also [00:06:16] claiming that [00:06:17] chatbots are dead and [00:06:18] SaaS is more back [00:06:19] than ever. [00:06:20] That's right. Say, say [00:06:21] more. [00:06:22] Chatbots are old news. They're [00:06:23] table stakes. [00:06:24] They are basically [00:06:25] only as powerful as the [00:06:26] data that they have [00:06:27] access to, [00:06:28] and it's just a bad [00:06:29] way of doing [00:06:30] business.
I think about it [00:06:31] as if you spend all [00:06:32] of your days in [00:06:33] meetings, what do you actually get [00:06:34] done? All you've [00:06:35] had is a nice [00:06:36] chat. You've gotten a bit of [00:06:37] therapy from an [00:06:38] AI chatbot [00:06:39] where you [00:06:40] interrogated whether or [00:06:41] not you [00:06:42] are [00:06:43] managing your [00:06:44] portfolio well, when in [00:06:45] fact you still only are [00:06:46] really looking at the [00:06:47] top five performing [00:06:48] assets and the [00:06:49] worst performing [00:06:50] assets, and all of the [00:06:51] assets in the middle are [00:06:52] going unexamined, [00:06:53] right?
The [00:06:54] AI chatbot [00:06:55] experience relies on the [00:06:56] person using [00:06:57] it to be [00:06:58] driving end-to-end [00:06:59] the value of that [00:07:00] tool. What [00:07:01] we do is we [00:07:02] bring an informed [00:07:03] opinion. We [00:07:04] bring a software [00:07:05] that allows [00:07:06] people to have [00:07:07] guardrails, that has [00:07:08] pre-built agents [00:07:09] against [00:07:10] workspaces for the [00:07:11] 91 most [00:07:12] commonly automated [00:07:13] processes within [00:07:14] commercial real estate.
We [00:07:15] developed an atlas for [00:07:16] this, and that gives [00:07:17] you [00:07:18] recommendations as to the obvious [00:07:19] next steps in any [00:07:20] workflow, and it [00:07:21] also ensures that [00:07:22] the data that are [00:07:23] sitting against those [00:07:24] queries [00:07:25] have integrity and [00:07:26] have had a human in the [00:07:27] loop validation [00:07:28] against them. [00:07:29] So I think [00:07:30] chatbots, everyone has [00:07:31] one, nobody [00:07:32] cares.
The [00:07:33] future's actually gonna be built [00:07:34] on classic [00:07:35] machine learning based [00:07:36] software that, as I [00:07:37] said, can do things [00:07:38] like take your [00:07:39] transactions history [00:07:40] and use that [00:07:41] to inform [00:07:42] your conversation [00:07:43] with a chatbot about [00:07:44] whether you should make this [00:07:45] next acquisition or [00:07:46] not. It's [00:07:47] everything that precedes [00:07:48] someone asking a [00:07:49] question to a chatbot that [00:07:50] matters.
[00:07:51] Chatbots also, as I [00:07:52] mentioned, are extremely [00:07:53] reactive. You have [00:07:54] to log in, you have to [00:07:55] come up with a question. You [00:07:56] have to figure out how to make [00:07:57] use of it, and [00:07:58] frankly, that's why there's [00:07:59] no ROI in [00:08:00] enterprise AI. [00:08:01] Because people are just [00:08:02] twiddling their [00:08:03] thumbs, thinking about what to [00:08:04] make for dinner, [00:08:05] having a conversation with [00:08:06] ChatGPT [00:08:07] rather than [00:08:08] using it for North [00:08:09] Star business [00:08:10] metrics.
If you [00:08:11] have software that has [00:08:12] an opinion, that is [00:08:13] railroaded, that is [00:08:14] fit for purpose, [00:08:15] it's going to [00:08:16] recommend work to you and [00:08:17] it's going to [00:08:18] pre-process work for you that your [00:08:19] business cares about, [00:08:20] whether that's doing [00:08:21] more deals month on [00:08:22] month, whether that's [00:08:23] doing better recurring [00:08:24] billing and [00:08:25] reconciliation against your [00:08:26] general ledger so you can [00:08:27] increase your [00:08:28] loss recovery, gain [00:08:29] 3 to 10% [00:08:30] more revenue year on year, [00:08:31] as some of our clients [00:08:32] do.
That's [00:08:33] really important, [00:08:34] critical work that actually [00:08:35] contributes to the P&L. [00:08:36] It's not just [00:08:37] having a nice conversation [00:08:38] with a nice little [00:08:39] chatbot. [00:08:40] Talking about [00:08:41] conversations with nice little [00:08:42] chatbots and a [00:08:43] word I want to highlight from [00:08:44] what you said is the [00:08:45] guardrails. That [00:08:46] subject matter [00:08:47] expertise where [00:08:48] if you just let [00:08:49] anyone [00:08:50] blast tokens, [00:08:51] as some may [00:08:52] refer to it, asking [00:08:53] questions randomly, [00:08:54] enhance Google [00:08:55] searches, if you will.[00:08:56]
That's what a lot of [00:08:57] real estate investment [00:08:58] teams in the early stage [00:08:59] of exploring AI [00:09:00] have started [00:09:01] doing, right? [00:09:02] Experimenting with [00:09:03] ChatGPT or general [00:09:04] purpose AI [00:09:05] tools. So is [00:09:06] this a [00:09:07] viable approach [00:09:08] for certain type of [00:09:09] companies of certain [00:09:10] sizes, certain [00:09:11] types? Or where [00:09:12] does it fall [00:09:13] short in terms of [00:09:14] a generalist [00:09:15] versus a [00:09:16] specific built [00:09:17] fifth dimension [00:09:18] with guardrails for [00:09:19] real estate investment?[00:09:20]
Good [00:09:21] question. So I think [00:09:22] for small and [00:09:23] medium-sized [00:09:24] businesses, having [00:09:25] some [00:09:26] individual chatbot [00:09:27] licenses, [00:09:28] ChatGPT, [00:09:29] Copilot, Claude [00:09:30] can help, [00:09:31] but only if [00:09:32] you have someone [00:09:33] within those [00:09:34] businesses who [00:09:35] has already [00:09:36] identified what [00:09:37] use cases should [00:09:38] be focused on for the [00:09:39] use of those tools, [00:09:40] and also [00:09:41] provides training on how to use [00:09:42] them.
Again, [00:09:43] you're not going to get [00:09:44] adoption of those [00:09:45] tools, and therefore [00:09:46] get any ROI from [00:09:47] them, if people don't [00:09:48] know how they should be [00:09:49] used to achieve business [00:09:50] metrics, [00:09:51] not just to play [00:09:52] around. So [00:09:53] all that [00:09:54] usually is too much [00:09:55] investment for a business [00:09:56] internally. AI [00:09:57] is not, right, the [00:09:58] commercial special sauce of [00:09:59] these businesses, [00:10:00] which is why it's [00:10:01] often better to work with a [00:10:02] supplier [00:10:03] anyway who has a [00:10:04] team that can come in [00:10:05] and train on the [00:10:06] tool and make sure that that [00:10:07] ROI is being [00:10:08] delivered.
[00:10:09] However, I think the [00:10:10] real problem at [00:10:11] scale in any [00:10:12] organization is the [00:10:13] data. So if [00:10:14] you are managing at least [00:10:15] a couple [00:10:16] hundred billion dollars of [00:10:17] assets, if you [00:10:18] have large scale [00:10:19] funds in an [00:10:20] investment management [00:10:21] space, what you [00:10:22] really need is [00:10:23] someone to take the reins, [00:10:24] provide the [00:10:25] guardrails for [00:10:26] integrating data sources today [00:10:27] in your business that [00:10:28] are not currently [00:10:29] integrated [00:10:30] into a single source of [00:10:31] truth, into a system [00:10:32] of record, again, [00:10:33] to make it available [00:10:34] for agentic [00:10:35] AI to actually make use of [00:10:36] it.
Mm-hmm. And that's [00:10:37] our core value [00:10:38] proposition. We [00:10:39] do have a [00:10:40] chatbot, an orchestration [00:10:41] agent. Her name is Agent [00:10:42] Ellie. But [00:10:43] that's kind of, again, [00:10:44] the most liminal, [00:10:45] least important part of [00:10:46] the experience. [00:10:47] What's important to us [00:10:48] is, for [00:10:49] example, if you are [00:10:50] on a [00:10:51] financial operations [00:10:52] team, that we have [00:10:53] an integration [00:10:54] with your [00:10:55] ERP.
We can easily [00:10:56] collect data from your [00:10:57] buildings managers. We [00:10:58] understand how your [00:10:59] general ledger is supposed [00:11:00] to work for your different [00:11:01] investment platforms. [00:11:02] We're able to [00:11:03] reconcile an [00:11:04] expense that's coded [00:11:05] differently in 40 [00:11:06] different ways across [00:11:07] 100 different [00:11:08] spreadsheets. [00:11:09] Therefore, you can actually do [00:11:10] recurring billing in a [00:11:11] way that saves you money and [00:11:12] does better loss [00:11:13] recovery.
So for [00:11:14] us, that's the [00:11:15] real work of [00:11:16] machine learning, and the [00:11:17] real work of [00:11:18] AI is in [00:11:19] actually creating what you would [00:11:20] call a classic data [00:11:21] ontology. We [00:11:22] need to understand [00:11:23] what features are [00:11:24] important about your [00:11:25] data set, how they [00:11:26] relate to each [00:11:27] other, and that's what [00:11:28] drives the actual [00:11:29] return on investment from [00:11:30] using an AI [00:11:31] platform.[00:11:32]
Talking about systems [00:11:33] of record, we've [00:11:34] seen this [00:11:35] playing out [00:11:36] already [00:11:37] in, uh, different [00:11:38] industries. JP [00:11:39] Morgan, Delta, and [00:11:40] Amazon [00:11:41] blocked AI [00:11:42] agents from their [00:11:43] platforms just in the [00:11:44] past week. Within [00:11:45] our industry on the, on [00:11:46] the leasing side, [00:11:47] we see [00:11:48] RealPage just did [00:11:49] something similar to [00:11:50] LeaseAI [00:11:51] in multifamily with [00:11:52] their agent.
[00:11:53] So [00:11:54] incumbent platforms [00:11:55] seem to be drawing [00:11:56] lines on what agents [00:11:57] can and can't do [00:11:58] inside the walls. Do [00:11:59] you have any thoughts on how [00:12:00] this plays out [00:12:01] within your [00:12:02] sector of the [00:12:03] business? [00:12:04] Absolutely. more [00:12:05] famous people in the [00:12:06] software industry have said [00:12:07] this before, but [00:12:08] the basic [00:12:09] framework should be that AI [00:12:10] agents start with [00:12:11] zero permissions.[00:12:12]
They start with [00:12:13] absolutely nothing [00:12:14] and have to be [00:12:15] explicitly permissioned to [00:12:16] have access to any [00:12:17] individual data [00:12:18] source and to any [00:12:19] individual team [00:12:20] member. [00:12:21] So in Fifth [00:12:22] Dimension, for example, [00:12:23] we operate with [00:12:24] SSO RBAC [00:12:25] with Microsoft. [00:12:26] So let's say you have [00:12:27] SharePoint, [00:12:28] and you have different [00:12:29] teams with different [00:12:30] permissions to different data [00:12:31] sets.
We inherit [00:12:32] those from that [00:12:33] platform. The agent is [00:12:34] guardrailed [00:12:35] only within those [00:12:36] permission sets, and [00:12:37] again, within our [00:12:38] workspaces, our [00:12:39] individual databases [00:12:40] for something like a [00:12:41] deal, an asset, or [00:12:42] a portfolio, or [00:12:43] a fund, [00:12:44] it's only [00:12:45] restricted to that data [00:12:46] warehouse. [00:12:47] And so you should [00:12:48] treat essentially [00:12:49] onboarding an AI [00:12:50] agent, again, like onboarding [00:12:51] a human being.
You [00:12:52] don't give your [00:12:53] associate or your [00:12:54] analyst access to [00:12:55] CEO-level [00:12:56] business data. [00:12:57] That's not how you [00:12:58] start. You start with [00:12:59] zero permissions, [00:13:00] gain trust along [00:13:01] the way. And [00:13:02] the second most critical [00:13:03] piece here for [00:13:04] any, again, [00:13:05] business that's [00:13:06] operating at scale, [00:13:07] any kind of [00:13:08] true upmarket [00:13:09] enterprise [00:13:10] business, you have to [00:13:11] think about being [00:13:12] able to [00:13:13] understand [00:13:14] data provenance and [00:13:15] understand who in [00:13:16] your business has [00:13:17] touched which data [00:13:18] in the process of [00:13:19] making an investment [00:13:20] decision.
So if [00:13:21] you're acquiring a three [00:13:22] hundred million dollar [00:13:23] asset, you don't [00:13:24] want it to come up when [00:13:25] that asset is [00:13:26] underperforming at your board [00:13:27] meeting, "Oh, [00:13:28] how did we end up [00:13:29] valuating this, [00:13:30] asset at three hundred [00:13:31] million instead of two [00:13:32] hundred and [00:13:33] ninety-seven?" And the answer [00:13:34] is, "Oh, I asked [00:13:35] Claude, and that's what Claude's [00:13:36] recommendation was."
[00:13:37] Mm-hmm. So on Fifth [00:13:38] Dimension's platform, [00:13:39] there's an entire [00:13:40] audit trail for [00:13:41] every single action [00:13:42] that is taken, [00:13:43] so it's not only a [00:13:44] system of record, it's a [00:13:45] system of actions. [00:13:46] So you can see [00:13:47] user A logged [00:13:48] in at this [00:13:49] time. The [00:13:50] underlying LLM [00:13:51] that was accessed was [00:13:52] a Gemini [00:13:53] model. This question [00:13:54] was asked. The [00:13:55] mathematical calculation [00:13:56] for the internal rate of [00:13:57] return was [00:13:58] this, and you can trace [00:13:59] the whole full [00:14:00] stack.
I think [00:14:01] that's actually the [00:14:02] future because we're gonna see [00:14:03] more and more regulation [00:14:04] of how these AI [00:14:05] tools are used. The [00:14:06] examples that you gave [00:14:07] are businesses just kind [00:14:08] of getting ahead [00:14:09] of what's probably [00:14:10] going to be, like, [00:14:11] national level [00:14:12] regulation about data [00:14:13] localization [00:14:14] and how AI should or [00:14:15] shouldn't be used for [00:14:16] critical financial [00:14:17] processes.
So I [00:14:18] know that the [00:14:19] SEC, other regulatory [00:14:20] bodies are also [00:14:21] thinking about this [00:14:22] because you have to be able [00:14:23] to pass an audit. [00:14:24] If you're a financial [00:14:25] services [00:14:26] institution, how do you [00:14:27] defend the use of AI, and [00:14:28] what can you say about your [00:14:29] use of AI is gonna [00:14:30] become a really critical [00:14:31] question. [00:14:32] So many [00:14:33] tangents we can [00:14:34] take from what you just said [00:14:35] and unpack here [00:14:36] from [00:14:37] AI [00:14:38] regulation.
I mean- [00:14:39] It's taken us 20 [00:14:40] years to regulate social [00:14:41] media, and we're not even [00:14:42] there yet. I think [00:14:43] it took [00:14:44] several decades to regulate [00:14:45] tobacco. [00:14:46] I'm not saying [00:14:47] AI is like social media [00:14:48] or tobacco, [00:14:49] but it certainly needs [00:14:50] some guardrails. [00:14:51] And the other [00:14:52] thing that I- I'm [00:14:53] thinking, like, a lot of these [00:14:54] systems of records [00:14:55] didn't have any issue [00:14:56] with, even [00:14:57] to today, but [00:14:58] up until recently, [00:14:59] many companies would just [00:15:00] hire [00:15:01] entire teams [00:15:02] abroad to [00:15:03] sort through data, [00:15:04] look at it.
So you had, [00:15:05] like, a army [00:15:06] of human [00:15:07] agents, if you will, [00:15:08] and that was all [00:15:09] fine. Of [00:15:10] course, not comparing these [00:15:11] apples to apples to [00:15:12] AI, but, like, [00:15:13] it seems like it's, [00:15:14] triggered a, not [00:15:15] existential necessarily, [00:15:16] but, like, a serious [00:15:17] competitive threat [00:15:18] or perceived [00:15:19] threat for a lot of the [00:15:20] incumbents when it [00:15:21] comes to [00:15:22] what they [00:15:23] allow their [00:15:24] users and partners to [00:15:25] do with their data.[00:15:26]
Yeah, [00:15:27] that's absolutely true, [00:15:28] and I think the [00:15:29] landscape hasn't changed [00:15:30] that much in terms of what the [00:15:31] challenges are. [00:15:32] It's [00:15:33] that machine learning [00:15:34] tools, both large [00:15:35] language models and [00:15:36] classic machine learning [00:15:37] algorithms, are [00:15:38] just better at [00:15:39] surfacing [00:15:40] data [00:15:41] conflicts and [00:15:42] allowing a human in the [00:15:43] loop to validate [00:15:44] them.
That's the point, [00:15:45] right? So you could still [00:15:46] be outsourcing [00:15:47] your lease [00:15:48] abstraction to a team in [00:15:49] India. You still [00:15:50] have quality [00:15:51] controls issues. There's [00:15:52] still a team [00:15:53] in-house that those data get [00:15:54] sent to, and [00:15:55] they do in fact get [00:15:56] reconciled line by [00:15:57] line if they're critical data [00:15:58] sets. And I [00:15:59] think the beauty [00:16:00] of AI is [00:16:01] that you can do that [00:16:02] 100 times [00:16:03] faster.
Also, [00:16:04] again, cited to the [00:16:05] page, to the [00:16:06] clause, to the Excel [00:16:07] spreadsheet in a way [00:16:08] that makes it really easy to [00:16:09] click through with [00:16:10] confidence. Oh, yes, that [00:16:11] lease covenant does [00:16:12] supersede that lease [00:16:13] covenant. Commit to the [00:16:14] database covenant [00:16:15] A. Mm-hmm. [00:16:16] You also mentioned [00:16:17] liability earlier. I [00:16:18] don't even wanna open that [00:16:19] can of worms right now [00:16:20] because it's a matter of [00:16:21] time we're gonna [00:16:22] give [00:16:23] agents being, [00:16:24] uh, [00:16:25] people, like the same as we [00:16:26] did with companies, like, [00:16:27] legally speaking.[00:16:28]
I don't wanna [00:16:29] open that can of worms. [00:16:30] Al, do you? [00:16:31] No, I mean, [00:16:32] business is gonna get [00:16:33] sued. Also, the [00:16:34] US is an [00:16:35] incredibly litigious [00:16:36] culture. There are [00:16:37] examples of this already [00:16:38] happening in the [00:16:39] marketplace. The question is [00:16:40] about, [00:16:41] again, why do you [00:16:42] want a human in the loop? [00:16:43] Whose judgment do you [00:16:44] think actually [00:16:45] matters for your [00:16:46] critical business [00:16:47] processes?
If [00:16:48] it's me, I would [00:16:49] treat an [00:16:50] AI agent [00:16:51] like an analyst [00:16:52] more than I would like a [00:16:53] CEO. You're [00:16:54] gonna have two eyes, you're gonna [00:16:55] have four eyes on that [00:16:56] work before it gets sent [00:16:57] out. So let's [00:16:58] dive deep [00:16:59] into Fifth [00:17:00] Dimension. So let's [00:17:01] say, uh, you're working [00:17:02] with a real estate [00:17:03] firm. So how does that [00:17:04] workflow change or [00:17:05] look like [00:17:06] within the first six [00:17:07] months of working with Fifth [00:17:08] Dimension?[00:17:09]
Yeah, I think there are [00:17:10] really two ways [00:17:11] of approaching [00:17:12] AI [00:17:13] adoption. The [00:17:14] first is to say, [00:17:15] okay, we run this [00:17:16] process [00:17:17] today by, [00:17:18] let's take the [00:17:19] example of going into your [00:17:20] inbox, looking for [00:17:21] property brochures or [00:17:22] offering memoranda, [00:17:23] seeing what an agent has [00:17:24] sent you, [00:17:25] starting to analyze those [00:17:26] data against your buy [00:17:27] box, starting to build an [00:17:28] investment case, [00:17:29] and then taking that [00:17:30] through to a deal [00:17:31] memo.
And then you [00:17:32] recreate that [00:17:33] process digitally. [00:17:34] So you have [00:17:35] a workspace, for [00:17:36] example, on Fifth Dimension's [00:17:37] prep forum. We [00:17:38] go in, we take [00:17:39] those data from your [00:17:40] inbox, we [00:17:41] extract them, we [00:17:42] analyze them against your buy [00:17:43] box, then you have a [00:17:44] nice Excel spreadsheet [00:17:45] waiting for you on Monday [00:17:46] morning where you can decide wh- [00:17:47] which deals are [00:17:48] interesting to you and which [00:17:49] deals aren't.
That's [00:17:50] what most [00:17:51] businesses, frankly, are [00:17:52] comfortable getting started [00:17:53] with. It's a good [00:17:54] place to start, but what [00:17:55] it delivers is [00:17:56] incremental task [00:17:57] efficiency gains. So [00:17:58] what you've gotten back [00:17:59] there are [00:18:00] minutes, [00:18:01] hours against an [00:18:02] analyst or an [00:18:03] associate's time [00:18:04] in going through [00:18:05] those data and [00:18:06] putting together that [00:18:07] investment case.
[00:18:08] But if you're working with a [00:18:09] proper software provider [00:18:10] like Fifth Dimension, you've [00:18:11] also created a data [00:18:12] set. You've [00:18:13] created data that [00:18:14] wouldn't otherwise have [00:18:15] been cataloged [00:18:16] anywhere, committed to a system [00:18:17] of record, 'cause you [00:18:18] opened a PDF, [00:18:19] decided that asset wasn't [00:18:20] for you, and closed [00:18:21] it, when in fact [00:18:22] there might be interesting [00:18:23] things in there, of course, to [00:18:24] inform future [00:18:25] transactions.
But [00:18:26] again, the [00:18:27] gains are, [00:18:28] interesting. They're [00:18:29] real, but they're sort of [00:18:30] minimal. They're [00:18:31] contained. The [00:18:32] other way of approaching [00:18:33] AI adoption is [00:18:34] saying, how would we [00:18:35] actually do things [00:18:36] entirely differently [00:18:37] with AI at the [00:18:38] center of this [00:18:39] process? [00:18:40] And there you have to be [00:18:41] willing to dream [00:18:42] bigger, [00:18:43] digitize more of your [00:18:44] data, and think [00:18:45] about what is [00:18:46] currently wrong about [00:18:47] the process that you're [00:18:48] running in a manual [00:18:49] fashion.
And [00:18:50] often that comes down to the [00:18:51] fact that there are 10 [00:18:52] versions [00:18:53] of your [00:18:54] master spreadsheet [00:18:55] for internal rate of [00:18:56] returns for [00:18:57] retail assets. Do [00:18:58] you need 10 versions [00:18:59] of that spreadsheet? [00:19:00] What's the purpose of [00:19:01] them? Do you actually [00:19:02] want to have [00:19:03] standardization across your [00:19:04] different offices, across your [00:19:05] different [00:19:06] geographies?
Thinking about [00:19:07] how that relates to [00:19:08] interest rates, FX, [00:19:09] et cetera. [00:19:10] And for us, we [00:19:11] really push [00:19:12] businesses to think about [00:19:13] how can you get [00:19:14] your team [00:19:15] doing the same thing [00:19:16] the same way [00:19:17] every time against [00:19:18] the same source of [00:19:19] truth? And that's just [00:19:20] not how most [00:19:21] businesses run today, [00:19:22] but that's where the real [00:19:23] gains are made, [00:19:24] because you're not gaining back [00:19:25] minutes, you're [00:19:26] gaining back huge [00:19:27] amounts of [00:19:28] commercially valuable [00:19:29] data, and you're [00:19:30] also turning a [00:19:31] process on its head and [00:19:32] saying, "Okay, [00:19:33] now that we have [00:19:34] a [00:19:35] one-person team who [00:19:36] can do the work of the [00:19:37] prior 10, [00:19:38] what other [00:19:39] revenue streams might we [00:19:40] be interested in?
[00:19:41] Now that I have all of [00:19:42] my tenant credit [00:19:43] risk data [00:19:44] actually available, [00:19:45] I might have new [00:19:46] investment platforms. I [00:19:47] might have [00:19:48] new products I could [00:19:49] launch." That's where [00:19:50] it gets really [00:19:51] interesting, where AI [00:19:52] becomes a revenue [00:19:53] generator and [00:19:54] you have actual [00:19:55] CapEx [00:19:56] investments in AI [00:19:57] rather than it being [00:19:58] incremental task [00:19:59] efficiency gains.[00:20:00]
[00:20:01] There [00:20:02] clearly is an, an AI [00:20:03] rush in, in [00:20:04] commercial real estate right [00:20:05] now. There's also [00:20:06] been demo [00:20:07] fatigue, especially the [00:20:08] mid-size to larger [00:20:09] companies. A lot of them have [00:20:10] seen a lot of [00:20:11] demos now. [00:20:12] Every week there's a new [00:20:13] deal [00:20:14] intelligence tool, new data [00:20:15] layer, new [00:20:16] underwriting platform. [00:20:17] So how do [00:20:18] you think about [00:20:19] differentiating [00:20:20] yourself, not just [00:20:21] today, but going [00:20:22] forward as, as [00:20:23] you get adopted, and [00:20:24] also stay [00:20:25] ahead of the pace of [00:20:26] innovation to [00:20:27] make sure your clients [00:20:28] know that they made the right [00:20:29] decision, [00:20:30] they sign up with the [00:20:31] right AI [00:20:32] company?
First of [00:20:33] all, I have a lot of [00:20:34] empathy with [00:20:35] those who are looking at [00:20:36] the AI marketplace [00:20:37] right now. [00:20:38] I think [00:20:39] 2026 has been the [00:20:40] year of AI [00:20:41] burnout, right? [00:20:42] I've tried some [00:20:43] things. I've seen [00:20:44] 1,000 demos. [00:20:45] All of this stuff [00:20:46] frankly looks the same to [00:20:47] me. I'm [00:20:48] not really sure how to make [00:20:49] use of it, but I'm [00:20:50] being told that I have [00:20:51] to.
And so [00:20:52] I think there are a few [00:20:53] key things you need to look [00:20:54] for in terms of [00:20:55] [00:20:56] differentiators for AI native [00:20:57] software. [00:20:58] The first is, [00:20:59] does it have a service [00:21:00] layer? I [00:21:01] cannot emphasize this [00:21:02] enough to go back to the point [00:21:03] about getting ROI [00:21:04] from AI. [00:21:05] It doesn't matter how [00:21:06] good your software is [00:21:07] if nobody uses it [00:21:08] and nobody knows how to [00:21:09] use it.
And [00:21:10] that actually [00:21:11] takes buy-in at [00:21:12] an executive level [00:21:13] from a business that [00:21:14] defines what the [00:21:15] metrics are you care [00:21:16] about in digital [00:21:17] transformation. [00:21:18] Again, whether [00:21:19] that's I wanna do [00:21:20] more deals month [00:21:21] on month, I [00:21:22] have, you know, a bunch of [00:21:23] dry powder, I need to [00:21:24] deploy it quickly and [00:21:25] competitively, I need [00:21:26] to win assets [00:21:27] over my [00:21:28] competitors, or it's I [00:21:29] just need to not leave [00:21:30] money on the table for [00:21:31] loss recovery, [00:21:32] or it's I can't [00:21:33] stand the fact that we have [00:21:34] 20 people's time [00:21:35] taken up with every [00:21:36] quarterly asset [00:21:37] reporting cycle.[00:21:38]
Make sure you know [00:21:39] what your business cares [00:21:40] about, and work with a [00:21:41] supplier whose team [00:21:42] is willing to spend time [00:21:43] with you to make sure the [00:21:44] technology actually [00:21:45] delivers that [00:21:46] thing [00:21:47] In addition to that, as [00:21:48] you mentioned, right, these [00:21:49] foundation model [00:21:50] labs are gonna be pushing out [00:21:51] new features all [00:21:52] the time.
So [00:21:53] we all would be [00:21:54] foolish if we didn't [00:21:55] assume that [00:21:56] models [00:21:57] themselves will get good [00:21:58] enough to do most [00:21:59] individual [00:22:00] tasks like an [00:22:01] analyst could, whether [00:22:02] that's updating a [00:22:03] waterfall, [00:22:04] whether that's creating a [00:22:05] basic underwriting [00:22:06] model from scratch, [00:22:07] whether that's pulling [00:22:08] data out of a [00:22:09] PDF. All that stuff [00:22:10] is already [00:22:11] happening, and [00:22:12] specificity to a particular [00:22:13] industry is [00:22:14] relatively cheap.
[00:22:15] Anthropic buys up [00:22:16] all of these really [00:22:17] promising [00:22:18] financial services and [00:22:19] commercial real estate [00:22:20] professionals to make [00:22:21] sure they have the talent [00:22:22] in-house to deliver [00:22:23] against those ideals. [00:22:24] So what we [00:22:25] actually think is important, [00:22:26] again, is not [00:22:27] just the [00:22:28] language model, [00:22:29] 'cause large language [00:22:30] models still really [00:22:31] bad, for example, at [00:22:32] math.
It [00:22:33] doesn't matter if you have [00:22:34] ninety-nine or [00:22:35] ninety-eight percent [00:22:36] accuracy from a large language [00:22:37] model. It [00:22:38] matters what happens in the other [00:22:39] one or two percent. [00:22:40] If you're talking about [00:22:41] hundreds of [00:22:42] billions of dollars, [00:22:43] nay, billions of [00:22:44] dollars that are at stake in [00:22:45] transactions and [00:22:46] NOI uplifts year [00:22:47] on year, it [00:22:48] matters when you're wrong, [00:22:49] how you're wrong, and how [00:22:50] you can debug it.[00:22:51]
And so for [00:22:52] us, we actually have a [00:22:53] lot of [00:22:54] deterministic technology under [00:22:55] the hood. We do [00:22:56] not just use [00:22:57] probabilistic technology [00:22:58] because it's not the [00:22:59] right tool for [00:23:00] every single problem. [00:23:01] Mm-hmm. So one of the [00:23:02] innovations out of our [00:23:03] R&D team from our [00:23:04] staff engineers [00:23:05] at Fifth [00:23:06] Dimension, for example, is what [00:23:07] we call our model [00:23:08] library.
And [00:23:09] what that is is it [00:23:10] takes a financial [00:23:11] model in [00:23:12] Excel, and it [00:23:13] translates it into code, [00:23:14] into Python, [00:23:15] and it breaks it [00:23:16] down into its [00:23:17] fundamental components. So if you [00:23:18] have an internal rate of [00:23:19] return model, there [00:23:20] are several [00:23:21] subcomponents of that that can be [00:23:22] used, that can be QA'd [00:23:23] and QC'd [00:23:24] against individual [00:23:25] changes.
And [00:23:26] what that means [00:23:27] is, let's say you're doing [00:23:28] scenario [00:23:29] analysis as part of an [00:23:30] investment process, [00:23:31] and you would [00:23:32] normally have an [00:23:33] analyst run [00:23:34] different scenarios in [00:23:35] different Excel spreadsheets [00:23:36] and say, "Oh, [00:23:37] change the hold [00:23:38] period from seven [00:23:39] years to nine years. [00:23:40] Assume eighty [00:23:41] bips more [00:23:42] inflation over the [00:23:43] period that we're gonna hold the [00:23:44] investment."
[00:23:45] And when you're doing [00:23:46] something like that, [00:23:47] that is a process that [00:23:48] usually takes hours or [00:23:49] days in order [00:23:50] to go through and check [00:23:51] every mathematical [00:23:52] calculation. If you [00:23:53] do it with a large [00:23:54] language model, you're [00:23:55] gonna get nonsense. [00:23:56] You're not gonna get what you [00:23:57] expect in [00:23:58] column L. Column L [00:23:59] is gonna be formatted [00:24:00] differently.
You don't know [00:24:01] where the data are from. [00:24:02] But since we've [00:24:03] broken that down into [00:24:04] code, which is [00:24:05] inherently [00:24:06] deterministic, inherently [00:24:07] algorithmic, [00:24:08] you will get the same [00:24:09] results every time [00:24:10] by, for [00:24:11] example, changing the hold period [00:24:12] from seven years to nine [00:24:13] years, and you'll [00:24:14] be able to do it in [00:24:15] seconds. [00:24:16] Hmm.
[00:24:17] Collaboration [00:24:18] superpower. Quick [00:24:19] round. If you could partner [00:24:20] with one person, [00:24:21] historic or [00:24:22] living, real or [00:24:23] fictional, [00:24:24] who would it be? [00:24:25] Oh, [00:24:26] goodness. So I [00:24:27] think one of my personal [00:24:28] heroes is Michel de [00:24:29] Montaigne, [00:24:30] so Renaissance [00:24:31] philosopher, and he [00:24:32] basically [00:24:33] was the only [00:24:34] combatant at the time [00:24:35] against religious [00:24:36] fundamentalism, [00:24:37] and he [00:24:38] essentially [00:24:39] created the scientific [00:24:40] method.
So [00:24:41] what an [00:24:42] absolute icon [00:24:43] who was sitting [00:24:44] around, you know, [00:24:45] getting preached to, [00:24:46] proselytized to all the [00:24:47] time, and he [00:24:48] actually wanted to ask [00:24:49] fundamental questions about the [00:24:50] way the world [00:24:51] worked and get back [00:24:52] to first principles [00:24:53] thinking. That's how I [00:24:54] feel like all of us [00:24:55] need to navigate [00:24:56] a really [00:24:57] volatile time like the [00:24:58] one we're living in.
[00:24:59] Ask the right questions, [00:25:00] and you'll get the right answers [00:25:01] about things like [00:25:02] AI. [00:25:03] Kate, where can Tangent [00:25:04] listeners and real estate [00:25:05] investment [00:25:06] professionals learn more [00:25:07] about Fifth Dimension [00:25:08] and connect with you? [00:25:09] So please [00:25:10] do check out our [00:25:11] website. Go online to [00:25:12] [00:25:13] fifthdimensionai.com. You can [00:25:14] also go to [00:25:15] my LinkedIn, send [00:25:16] me an email at [00:25:17] [00:25:18] kate@fifthdimensionai.com, [00:25:19] and you can [00:25:20] also tune into our [00:25:21] podcast and [00:25:22] really get a feel for who we [00:25:23] are.
'Cause like I [00:25:24] said, you can work [00:25:25] with any AI [00:25:26] supplier by putting your [00:25:27] credit card into the [00:25:28] internet. You don't know how to [00:25:29] trust it. You don't know where [00:25:30] your data are going. [00:25:31] You don't know how to get value out [00:25:32] of the platform. [00:25:33] But we're a real [00:25:34] human team that cares [00:25:35] about real results. [00:25:36] And you can [00:25:37] find a link of that in the [00:25:38] show notes.[00:25:39]
Dr. [00:25:40] Kate [00:25:41] Jarvis, CEO [00:25:42] and co-founder of [00:25:43] Fifth Dimension, [00:25:44] thank you so much [00:25:45] for coming [00:25:46] to Tangent today. [00:25:47] My [00:25:48] pleasure. Thanks for having me.
Edward: [00:25:49] [00:25:50] [00:25:51] [00:25:52] [00:25:53] [00:25:54] Muchas [00:25:55] gracias. If you learn [00:25:56] something new or [00:25:57] enjoy the [00:25:58] conversation. Text a friend [00:25:59] the link to this [00:26:00] episode right now.
[00:26:01] [00:26:02] You can find [00:26:03] links, resources, [00:26:04] and ways to connect with [00:26:05] us in the show [00:26:06] notes [00:26:07] [00:26:08] [00:26:09] [00:26:10] [00:26:11]