Tangent: Conversations with Real Estate & Tech Innovators

Fifth Dimension CEO Dr. Kate Jarvis discusses agentic AI for CRE, claiming 3-10% revenue lift via automation.

27 min Listen to the original Episode page
Commercial Real EstateArtificial IntelligencePropTechInvestment ManagementRisk Management

Summary

Dr. Kate Jarvis, CEO of Fifth Dimension, describes her company's agentic AI platform for commercial real estate investment managers, which structures disparate data into a unified system of record. She argues that unlike reactive chatbots, pre-built agents proactively execute workflows like data extraction and IRR calculation, allowing smaller teams to manage larger portfolios. Jarvis highlights a 'zero-permissions' security framework and deterministic code translation from Excel to Python to ensure auditability and accuracy. She claims clients have seen 3-10% year-over-year revenue increases through improved billing and loss recovery, while emphasizing that human judgment remains critical for final investment decisions.

Key takeaways

  1. 1
    Fifth Dimension raised $26 million in 2026 and serves clients including Boston Property Group, Realty Income, and Peachtree Group.

    Edward Cohen · 2:37

  2. 2
    gain three to 10% ten percent more revenue year on year, as some of our clients do

    Unidentified speaker (speaker_1) · 9:25

  3. 3
    The platform uses 'playbooks' for specific workflows, such as acquisitions, where agents extract data from emails and PDFs to populate spreadsheets for analysis.

    Unidentified speaker (speaker_1) · 3:53

  4. 4
    Jarvis states human judgment is critical in CRE and cannot be fully automated; AI should handle data crunching to free humans for judgment-based tasks.

    Unidentified speaker (speaker_1) · 4:33

  5. 5
    Chatbots are only educational if they cite data and provide audit trails for calculations like IRR; otherwise, they risk making users lazy or misinformed.

    Unidentified speaker (speaker_1) · 6:53

  6. 6
    Fifth Dimension has built an 'atlas' of the 91 most commonly automated processes in CRE to provide pre-built agents and guardrails.

    Unidentified speaker (speaker_1) · 8:05

  7. 7
    SSO RBAC, with Microsoft. So let's say you have SharePoint and you have different teams with different permissions to different data sets. We inherit those from that platform.

    Unidentified speaker (speaker_1) · 13:27

Episode sections

  1. 2:37 – 4:26

    Introduction and Fifth Dimension Overview

    Edward Cohen introduces Dr. Kate Jarvis, CEO of Fifth Dimension, noting the company's $26 million raise and clients including Boston Property Group, Realty Income, and Peachtree Group. Jarvis describes Fifth Dimension as an agentic AI platform for CRE investment managers that structures data from various sources into a unified system of record to drive decision-making.

  2. 4:26 – 7:18

    Human Judgment and AI Collaboration

    Jarvis argues that AI should augment human judgment rather than replace it, allowing smaller teams to manage larger portfolios by automating manual data crunching. She emphasizes that human judgment remains critical for final investment decisions and that AI can serve as an educational tool for junior analysts when properly audited.

  3. 7:18 – 10:25

    Chatbots vs. Agentic AI

    Jarvis claims chatbots are 'old news' and lack ROI because they are reactive and rely on user initiative. She contrasts this with Fifth Dimension's pre-built agents that proactively perform workflows, such as extracting data from offering memoranda and comparing them against investment criteria, ensuring data integrity and human-in-the-loop validation.

  4. 10:25 – 13:08

    Generalist AI vs. Specialized Guardrails

    While generalist AI tools like ChatGPT may suffice for small businesses with internal expertise, large firms managing hundreds of billions in assets require specialized guardrails and a single source of truth. Jarvis asserts that the core value lies in integrating disparate data sources into a system of record for agentic AI, rather than just providing a chat interface.

  5. 13:08 – 14:41

    Zero-Permissions Framework and Security

    Jarvis advocates for a 'zero-permissions' framework where AI agents start with no access and are explicitly granted permissions based on existing user roles (e.g., via SSO/RBAC). She likens onboarding an AI agent to onboarding a human employee, restricting access to specific workspaces and data warehouses to ensure security and compliance.

  6. 14:41 – 18:13

    Audit Trails and Regulatory Scrutiny

    The speaker highlights the importance of audit trails for regulatory compliance, citing SEC scrutiny and potential national regulations on AI in finance. Fifth Dimension provides a full-stack audit trail that logs user actions, models used, and calculations performed, allowing firms to defend their AI usage during audits and avoid liability issues.

  7. 18:13 – 23:56

    Two Paths to AI Adoption

    Jarvis outlines two adoption paths: task efficiency (digitizing existing processes for incremental time savings) and process reinvention (reimagining workflows to generate new revenue streams). She argues that true value comes from the latter, which requires standardizing data and processes to create a reliable system of record.

  8. 23:56 – 27:19

    Model Library and Deterministic Code

    To address the limitations of LLMs in math, Fifth Dimension uses a 'model library' that translates Excel financial models into deterministic Python code. This ensures consistent, auditable results for complex calculations like IRR, allowing users to run scenario analyses in seconds with guaranteed accuracy, unlike probabilistic LLM outputs.

Claims worth verifying

Fifth Dimension raised $26 million in 2026.

Edward Cohen ·2:37

How to verify: Company press releases or SEC Form D filings for Fifth Dimension. For recent events, verify 2026-08-10 through 2026-11-10. Use another period only when the speaker explicitly names it.

Primary source

Clients include Boston Property Group, Realty Income, and Peachtree Group.

Edward Cohen ·2:37

How to verify: Company website case studies or press releases listing clients. For recent events, verify 2026-08-10 through 2026-11-10. Use another period only when the speaker explicitly names it.

Fifth Dimension has built an atlas of the 91 most commonly automated processes in CRE.

Unidentified speaker (speaker_1) ·8:05

How to verify: Fifth Dimension whitepapers or product documentation detailing their 'atlas' or 'playbook'. For recent events, verify 2026-08-10 through 2026-11-10. Use another period only when the speaker explicitly names it.

Clients gain 3-10% more revenue year-on-year through better recurring billing and loss recovery.

Unidentified speaker (speaker_1) ·9:25

How to verify: Fifth Dimension case studies or client testimonials. For recent events, verify 2026-08-10 through 2026-11-10. Use another period only when the speaker explicitly names it.

The platform operates with SSO and RBAC with Microsoft, inheriting permissions from SharePoint.

Unidentified speaker (speaker_1) ·13:27

How to verify: Fifth Dimension technical documentation or security whitepapers. For recent events, verify 2026-08-10 through 2026-11-10. Use another period only when the speaker explicitly names it.

Primary source

The SEC and other regulatory bodies are thinking about regulation of AI in finance.

Unidentified speaker (speaker_1) ·15:07

How to verify: SEC public statements, rulemaking documents, or industry reports on AI regulation. For recent events, verify 2026-08-10 through 2026-11-10. Use another period only when the speaker explicitly names it.

Primary source

LLMs are bad at math.

Unidentified speaker (speaker_1) ·23:16

How to verify: Independent benchmarks of LLM performance on mathematical tasks (e.g., GSM8K, MATH datasets). For recent events, verify 2026-08-10 through 2026-11-10. Use another period only when the speaker explicitly names it.

Fifth Dimension translates Excel financial models into deterministic Python code.

Unidentified speaker (speaker_1) ·23:56

How to verify: Fifth Dimension product documentation or technical demos of the 'model library'. For recent events, verify 2026-08-10 through 2026-11-10. Use another period only when the speaker explicitly names it.

Changing hold period from seven to nine years in the Python model takes seconds.

Unidentified speaker (speaker_1) ·25:16

How to verify: Fifth Dimension product demo or performance benchmarks. For recent events, verify 2026-08-10 through 2026-11-10. Use another period only when the speaker explicitly names it.

Chelsea's Starred Lehigh building opened in 1931.

Unidentified speaker (UNKNOWN) ·0:00

How to verify: Property records or historical archives for the Lehigh Valley Railroad Terminal (Starred Lehigh) in Chelsea, NYC. For recent events, verify 2026-08-10 through 2026-11-10. Use another period only when the speaker explicitly names it.

Do your own research

  • Verify Fifth Dimension's $26 million raise via SEC Form D filings or press releases from 2026.
  • Check client case studies from Boston Property Group, Realty Income, and Peachtree Group for independent validation of the 3-10% revenue lift claim.
  • Review the 'atlas' of 91 automated processes to assess if they cover your firm's specific CRE workflows.
  • Evaluate the 'zero-permissions' security model and Microsoft SSO/RBAC integration against your firm's internal compliance and cybersecurity standards.
  • Confirm the accuracy of deterministic Python code translation for complex financial models (e.g., IRR, NPV) against your existing Excel benchmarks.
  • Assess regulatory readiness by reviewing the platform's full-stack audit trail capabilities against current SEC guidance on AI usage in finance.
Full transcriptexpand
0:00
Unidentified speaker (UNKNOWN): When
0:00
Edward Cohen: Chelsea's Starred Lehigh building opened in 1931, freight cars rolled straight inside, and giant elevators carried trucks up to tenants' doors as high as 19 stories in the air. It tackled a problem everyone else accepted: moving goods through a jammed-up city. That's the thing about breakthroughs. In
0:20
Edward Cohen: hindsight, they look obvious. In the moment, a little crazy.
0:24
Unidentified speaker (UNKNOWN): On
0:24
Edward Cohen: December 2nd, we're returning to RXR. RXR's Starred Lehigh building for the 2026 Breakthrough Awards Breakfast, honoring the teams, deals, and technologies that broke through this year. We'll celebrate innovation across multifamily, office, retail, and industrial, plus the year's top technologies and transactions. And the real
0:44
Edward Cohen: value? The room. Meet the owners, founders, developers, investors, lenders, and brokers behind this work. Showcase your brand and connect with new clients face to face. To join us, email us at ecohen at commercialobserver.com or click the link in the show notes. Hope to see you on December 2nd.
1:04
Edward Cohen: Hi, I'm Edward Cohen, and this is Tangent, a Commercial Observer podcast.
1:13
Edward Cohen: Coming to you live from Cretech. CreeTech New York at the Javit Center, the New York City Real Estate League.
1:20
Edward Cohen: Real estate investment firms make billion dollar decisions on data that lives in spreadsheets, email threads, PDFs, and attachments. The intelligence layer has never existed until AI came to the scene.
1:35
Unidentified speaker (UNKNOWN): Our
1:35
Edward Cohen: guest today is Dr. Kate Jarvis, CEO and co-founder of
1:38
Unidentified speaker (UNKNOWN): the COVID.
1:39
Edward Cohen: Fifth Dimension.
1:40
Unidentified speaker (UNKNOWN): Hi
1:41
Edward Cohen: Kate, how is New York City Real Estate Week treating you?
1:43
Unidentified speaker (speaker_1): Hello, thank you for having me. I think as you can tell, there's a lot of buzz at the conference here today. And I think obviously New York is one of the financial centers of the world. I walk around this city thinking, ah, what an enormous city built on debt. Because that's what I deal with at Fifth Dimension, is everyone's documents and data. delivering insights
2:04
Unidentified speaker (speaker_1): and actions off the back of them using agentic AI. And it's just amazing to me that this world in which we live, work, and play has such a complicated set of data stakeholders, et cetera, against it. And before I got into this industry about a decade ago, I never really thought about it. But just even the Javit Center, right? The Javit Center
2:24
Unidentified speaker (speaker_1): itself is a phenomenal development, and the way in which it's used and commercialized today, we couldn't be here without that.
2:30
Edward Cohen: Absolutely. New York City is a massive playground of beta, real estate, and uh financial potential.
2:36
Unidentified speaker (speaker_1): Exactly.
2:37
Edward Cohen: So, fifth dimension, I mean, you guys have been on fire lately. You raised twenty-six million dollars earlier this year, and you're working with firms like Boston Property Group, Realty Income, and Peachtree Peach Tree Group. Tell us about the company and the team. teams and the firms that you're building for and working with.
2:54
Unidentified speaker (speaker_1): Yes, so Fifth Dimension is for investment managers, asset managers, and brokers who are looking for a new way of doing business. As you mentioned, this industry runs on data, and there have been many decades of innovation, many decades of attempts at turning those data into a commercial advantage. So we
3:14
Unidentified speaker (speaker_1): have the data data. dashboard era in the 2010s, we have our ERP systems, we have our CRMs, there's software, there's documents, there's Excel spreadsheets, there's PDFs, and all of it doesn't get leveraged in the way that is best for these businesses. What you really want is for the way that you do business to be
3:34
Unidentified speaker (speaker_1): defined by your data. Your data are your alpha, your data inform your buy box. your data inform which investments you make, when you might offload an asset. And we're the platform that helps you see around corners in order to make better decisions. So, in working with some of the businesses that you mentioned, we tend to have out of the box a
3:53
Unidentified speaker (speaker_1): platform where there's a workspace, which is a type of database, and a set of out-of-the-box AI agents and data connectors available to run. And we call that the playbook. You have the playbook for acquisitions, which means every workspace might be a deal. And we might have agents within that workspace that go into your Outlook inbox, find all of the offering
4:13
Unidentified speaker (speaker_1): memoranda and property brochures that have been sent to you over the weekend, extract data from them, measure those against your investment platform, and serve those data to you in a beautiful spreadsheet ready for your analysis on Monday morning. So we're also not coming. cutting the human out of a loop. And I think that's really, really critical. Human judgment is also what makes this industry
4:33
Unidentified speaker (speaker_1): run. You can't hand it over, hand over the keys to the house, just to a computer. And so it's really important to us that people spend their time doing what they're best at, which is using their judgment. And all of this data crunching to figure out how to use past transactions to inform future transactions should be left to the computer.
4:50
Edward Cohen: It is a fascinating time, especially when it comes to that collaboration between humans and technology, if you can call it that. It seems like you fall somewhere in the middle between the spectrum of I mean we we hear now CEOs of tech companies telling a very different message in terms of what AI is doing, whether AI is
5:10
Edward Cohen: replacing humans at their jobs and that's why you should hire them. their technology, or is it AI enhancing those roles until it replaces them? Or will it be a constant collaboration and evolution of the roles and the functions? It seems like you fall somewhere in the middle where alienating the human will only stifle
5:30
Edward Cohen: innovation or stifle the adoption, the pace of adoption.
5:33
Unidentified speaker (speaker_1): I think that's exactly right. AI should work to the best of human capability and augment with computer science, with the ability to look across billions of data points and quickly come up with a recommendation for an investment. But you're never gonna have one click, I want to purchase this asset based on the information that my AI has told me.
5:53
Unidentified speaker (speaker_1): And I think that's important because the future is going to be about, yes, smaller teams that can manage infinitely more money than they can today, because we are replacing. the manual labor, the person hours that are sitting behind that analysis. But what I think that actually enables is kind of full stack teams. You can't spend your day as
6:13
Unidentified speaker (speaker_1): an analyst anymore getting coffee and updating waterfalls. You can actually sit on the investment committee because you yourself have that firsthand access to what's happening across the marketplace within your own transactions history, and you can use that to develop a very informed opinion quickly on what the best decision is. And in the past, that just wasn't possible, right?
6:33
Unidentified speaker (speaker_1): You'd have to spend five, seven, ten, twelve years training up in order to develop an informed opinion. So I think the beauty of AI is actually that it can be leveraged for education. And I hate all of this talk about how chatbots just make people lazy. They only make people lazy if you use them the wrong way, if you just trust what the answer is. from Claude about the valuation of a building
6:53
Unidentified speaker (speaker_1): without checking the answer, absolutely. But if you use technology that can cite its data, have a full auditability trail that goes and gives you the calculation for the internal rate of return and all of the analysis that sits behind it, it actually becomes a really powerful educational tool for people who are just starting out in the industry too.
7:11
Edward Cohen: Seems like you have strong opinions about chatbots just before this interview. you were talking about how people are claiming SAS is dead, but you're also claiming that chatbots are dead and SAS is more back than ever.
7:24
Unidentified speaker (UNKNOWN): That's right.
7:24
Edward Cohen: Say more.
7:25
Unidentified speaker (speaker_1): Chatbots are old news. They're table stakes, they are basically only as powerful as the data that they have access to. And it's just a bad way of doing business. I think about it as if you spend all of your days in meetings, what do you actually get done? All you have you've had is a nice chat. You've gotten a bit of therapy from an AI chatbot where you interrogated whether or
7:45
Unidentified speaker (speaker_1): not you are managing your portfolio well when in fact you still only are really looking at the top five performing assets and the worst performing assets, and all of the assets in the middle are going unexamined, right? The AI chatbot experience relies on the person using it to be driving end end-to-end to end the value of that tool. What we
8:05
Unidentified speaker (speaker_1): do is we bring an informed opinion, we bring a software that allows people to have guardrails that has pre-built agents against workspaces for the 91 most commonly automated processes within commercial real estate. We developed an atlas for this. And that gives you recommendations as to the obvious next steps in any workflow. And it also
8:25
Unidentified speaker (speaker_1): ensures that the data that are sitting against those queries have integrity and have had human-in-the-loop validation against them. So I think chatbots, everyone has one. Nobody cares. The future is actually going to be built on classic machine learning-based software that, as I said, can do things like take your transactions history and use that
8:45
Unidentified speaker (speaker_1): to inform your conversation with a chatbot about whether you should make this next acquisition or not. It's every everything that precedes someone asking a question to a chatbot that matters. Chatbots also, as I mentioned, are extremely reactive. You have to log in, you have to come up with a question, you have to figure out how to make use of it. And frankly, that's why there's no ROI in Enterprise AI,
9:05
Unidentified speaker (speaker_1): because people are just twiddling their thumbs thinking about what to make for dinner, having a conversation with Chat GPT, rather than using it for North Star business metrics. If you have software that has an opinion that is railroaded, that is fit for purpose, it's going to recommend work to you and it's going to pre-process work for you that your business cares about. Whether that's doing
9:25
Unidentified speaker (speaker_1): more deals month on month, whether that's doing better recurring billing and reconciliation against your general ledger, so you can increase your loss recovery, gain three to 10% ten percent more revenue year on year, as some of our clients do. That's really important critical work that actually contributes. to the PL. It's not just having a nice conversation with a nice little chat bot.
9:44
Edward Cohen: Talking about conversations with nice little chat bots, and a word I want to highlight from what you said is the guardrails. So that subject matter expertise where if you just let anyone blast tokens, as some may refer to it, asking questions randomly, enhance Google searches, if you will. That's what a lot of real estate investment teams in the early stage of exploring AI
10:04
Edward Cohen: have started doing, right? Experimenting with Chat GPT or general purpose AI tools. So is this a viable approach for certain types of companies of certain sizes, certain types, or where does it fall short in terms of a generalist versus a specific built fifth dimension with guardrails for real estate investment?
10:24
Unidentified speaker (speaker_1): Good question. So I think for small and medium. sized businesses, having some individual chatbot licenses, Chat GPT, Copilot, Claude can help, but only if you have someone within those businesses who has already identified what use cases should be focused on for the use of those tools and also
10:44
Unidentified speaker (speaker_1): provides training on how to use them. Again, you're not going to get adoption of those tools and therefore get any ROI from. them if people don't know how they should be used to achieve business metrics, not just to play around. So all that usually is too much investment for a business internally. AI is not, right, the commercial special sauce of these businesses, which is
11:04
Unidentified speaker (speaker_1): why it's often better to work with a supplier anyway who has a team that can come in and train on the tool and make sure that that ROI is being delivered. However, I think the real problem at scale. in any organization is the data. So if you are managing at least a couple hundred billion dollars of assets, if you have large-scale funds in an investment
11:24
Unidentified speaker (speaker_1): management space, what you really need is someone to take the reins, provide the guardrails for integrating data sources today in your business that are not currently integrated into a single source of truth, into a system of record, again, to make it available for agencai agentic AI. to actually make use of it.
11:40
Unidentified speaker (UNKNOWN): And
11:40
Unidentified speaker (speaker_1): that's our core value proposition. We do have a chatbot, an orchestration agent, her name is Agent Ellie, but that's kind of again the most liminal, least important part of the experience. What's important to us is, for example, if you are on a financial operations team, that we have an integration with your ERP, we can easily collect data from
12:00
Unidentified speaker (speaker_1): your buildings managers, we understand how you're your general ledger is supposed to work for your different investment platforms. We're able to reconcile an expense that's coded differently in 40 different ways across 100 different spreadsheets. Therefore, you can actually do recurring billing in a way that saves you money and does better loss recovery. So for us, that's the real work of machine learning
12:20
Unidentified speaker (speaker_1): and the real work of AI is in actually creating what you would call a classic data ontology. We need to understand what features are important about your data set, how they relate to each other, and that's what drives the actual return on investment from using an AI platform.
12:35
Edward Cohen: Talking about systems of record, we've seen this playing out already in uh different industries. JP Morgan, Delta, and Amazon blocked AI agents from their platforms just in the past week. Within our industry on the on the leasing side, we see RealPage just did something similar to a lease AI in multifamily
12:55
Edward Cohen: with their agent. So incumbent platforms seem to be drawing lines on what agents can and can do inside the walls. Do you have any thoughts on how this plays out within your sector of the business?
13:07
Unidentified speaker (speaker_1): Absolutely. More famous people in the software industry have said this before, but the basic framework should be that AI agents start with zero. permissions. They start with absolutely nothing and have to be explicitly permissioned to have access to any individual data source and to any individual team member. So in Fifth Dimension, for example, we operate with
13:27
Unidentified speaker (speaker_1): SSO RBAC, with Microsoft. So let's say you have SharePoint and you have different teams with different permissions to different data sets. We inherit those from that platform. The agent is guardrailed only. within those permission sets. And again, within our workspaces, our individual databases for something like a deal, an asset, portfolio, or a fund.
13:47
Unidentified speaker (speaker_1): It's only restricted to that data warehouse. And so you should treat essentially onboarding an AI agent again like onboarding a human being. You don't give your associate or your analyst access to CEO level business data. That's not how you start. You start with zero permissions, gain trust along the way. And the second most critical piece here for
14:07
Unidentified speaker (speaker_1): any, again, business that's operating at scale, any kind of true upmarket enterprise business, you have to think about being able to understand data provenance and understand who in your business has touched which data in the process of making an investment decision. So if you're acquiring a $300 million asset. you don't
14:27
Unidentified speaker (speaker_1): want it to come up when that asset is underperforming at your board meeting. Oh, how did we end up valuing this asset at $300 million instead of 297? And the answer is, oh, I asked Claude, and that's what Claude's recommendation was. So on Fifth Dimensions platform, there's an entire audit trail for every single action that is taken. So it's not only a
14:47
Unidentified speaker (speaker_1): system of record, it's a system of actions. So you can see. user A logged in at this time, the underlying LLM that was accessed was a Gemini model. This question was asked. The mathematical calculation for the internal rate of return was this, and you can trace the whole full stack. I think that's actually the future because we're gonna see more and more
15:07
Unidentified speaker (speaker_1): regulation of how these AI tools are used. The examples that you gave are businesses just kind of getting ahead of what's probably going to be like next. national level regulation about data localization and how AI should or shouldn't be used for critical financial processes. So I know that the SEC, other regulatory bodies are also thinking about this because you have to be able to pass
15:27
Unidentified speaker (speaker_1): an audit if you're a financial services institution. How do you defend the use of AI and what can you say about your use of AI? It's going is to become a really critical question.
15:35
Edward Cohen: So many tangents we can take from what you just said. and unpack here from AI regulation. I mean, it's taken us 20 years to regulate social media and we're not even there yet. I think it took several decades to regulate tobacco. I'm not saying AI is like social media or tobacco, but it certainly needs some guardrails. And the other
15:55
Edward Cohen: thing that I I'm thinking like a lot of these systems of records didn't have any issue with even to today, but uh up until recently many companies would just hire entire teams abroad to sort through data, look at it. So you had like an army of human agents, if you will, and that was all fine. Of course, not comparing these apples to apples to
16:15
Edward Cohen: AI, but like it seems like it's triggered a not existential necessarily, but like a serious competitive threat or perceived threat for a lot of the incumbents when it comes to what they uh allow their users and partners to do with their data.
16:30
Unidentified speaker (speaker_1): Yeah, that's absolutely true. And I think the landscape hasn't changed that much in terms of what the challenges are. It's that machine learning tools, both large language models and classic machine learning algorithms, are just better at surfacing data conflicts and allowing a human in the loop to validate them. That's the point, right? So you could still be
16:50
Unidentified speaker (speaker_1): outsourcing your lease abstraction to a team in India. You still have quality controls issues. There's still a team in-house that those data get sent to, and they do, in fact, get reconciled line by line if they're critical data sets. And I think the beauty of AI is that you can do that a hundred times faster. Also, again, cited to the page, to the clause,
17:10
Unidentified speaker (speaker_1): to the Excel spreadsheet in a way that makes it really easy to click through with confidence. Oh, yes, that. lease covenant does supersede that lease covenant. Commit to the database covenant A.
17:20
Edward Cohen: You also mentioned liability earlier. I don't even want to open that kind of worms right now because it's a matter of time we're gonna give agents being uh people, like the same as we did with companies, like legally speaking. I don't want to open that kind of worms now, out, do you?
17:35
Unidentified speaker (speaker_1): No, I mean business is gonna get sued. Also, the US is an incredibly litigious culture. There are examples of this already happening in the marketplace. The question is about again, why do you want a human in the loop? Whose judgment do you think actually matters for your critical business processes? If it's me, I would treat an AI agent
17:54
Unidentified speaker (speaker_1): like an analyst more than I would like a CEO. You're gonna have two eyes, you're gonna have four eyes on that work before it gets sent out.
18:01
Edward Cohen: So let's dive deep into fifth dimension. So let's say uh you're working with a real estate firm. So how does that workflow change or look like within the first six months of working with Fifth Dimension?
18:13
Unidentified speaker (speaker_1): Yeah, I think there are really two ways of approaching AI adoption. The first is to say, okay, we run this process today by Buy, let's take the example of going into your inbox, looking for property brochures or offering memorandas. memoranda, seeing what an agent has sent you, starting to analyze those data against your buy box, starting to build an investment case,
18:33
Unidentified speaker (speaker_1): and then taking that through to a deal memo. And then you recreate that process digitally. So you have a workspace, for example, on Fifth Dimensions Prep Forum. We go in, we take those data from your inbox, we extract them, we analyze them against your buy box, then you have a nice Excel spreadsheet waiting for you on Monday morning where you can side decide what which deals are interesting to you and which
18:52
Unidentified speaker (speaker_1): deals aren't. That's what most businesses, frankly, are comfortable getting started with. It's a good place to start, but what it delivers is incremental task efficiency gains. So what you've gotten back there are minutes, hours against an analyst or an associate's time and going through those data and putting together that investment case. But if you're working with
19:12
Unidentified speaker (speaker_1): a proper software provider like Fifth Dimension, you've also created a data set. You've created data that wouldn't otherwise have been cataloged anywhere, committed to a system of record, because you opened a PDF, decided that asset wasn't for you, and closed it. When in fact there might be interesting things in there, of course, to inform future transactions. But again, the gains are interesting,
19:32
Unidentified speaker (speaker_1): they're real, but they're sort of minimal, they're contained. The other way of approaching AI adoption is saying, how would we actually do things entirely differently with AI at the center of this process? And there you have to be willing to dream bigger, digitize more of your data, and think about what is currently wrong about the process that you're running in a
19:52
Unidentified speaker (speaker_1): manual fashion. And often that comes down to the fact that there are 10 versions of your master spreadsheet for internal rate of returns for retail assets. Do you need 10 versions? of that spreadsheet? What's the purpose of them? Do you actually want to have standardization across your different offices, across your different geographies, thinking about how that relates to interest rates,
20:12
Unidentified speaker (speaker_1): FX, et cetera? And for us, we really push businesses to think about how you can get your team doing the same thing the same way every time against the same source of truth. And that's just not how most businesses run today, but that's where the real game is. gains are made. Because you're not gaining back minutes, you're gaining back huge amounts of commercially
20:32
Unidentified speaker (speaker_1): valuable data. And you're also turning a process on its head and saying, okay, now that we have a one person team who can do the work of the prior 10, what other revenue streams might we be interested in? Now that I have all of my tenant credit risk data actually available, I might have new investment platforms. I might have
20:51
Unidentified speaker (speaker_1): new products I could launch. That's where it gets really interesting, where AI becomes a revenue generator and you have actual capex investments in AI rather than it being incremental task efficiency gains.
21:04
Edward Cohen: There clearly is an AI rush in in commercial real estate right now. There's also been demo fatigue, especially the mid-size mid-sized to larger companies. A lot of them have seen a lot of demo. demos now. Every week there's a new deal intelligence tool, new data layer, new underwriting platforms. So how do you think about differentiating yourself,
21:24
Edward Cohen: not just today, but going forward as as you get adopted and also stay ahead of the pace of innovation to make sure your clients know that they made the right decision? They signed sign up with the right AI company.
21:36
Unidentified speaker (speaker_1): First of all, I have a lot of empathy with those. who are looking at the AI marketplace right now. I think 2026 has been the year of AI burnout, right? I've tried some things, I've seen a thousand demos, all of this stuff, frankly, looks the same to me. I'm not really sure how to make use of it, but I'm being told that I have to. And so I think there are a
21:56
Unidentified speaker (speaker_1): few key things you need to look for in terms of differentiators for AI native software. The first is does it have a service layer? I cannot emphasize this enough to go back to the point about getting ROI from AI. It doesn't matter how good your software is if nobody uses it and nobody knows how to use it. And that actually takes buy in at an executive
22:16
Unidentified speaker (speaker_1): level from a business that defines what the metrics are you care about in digital transformation. Again, whether that's I want to do more deals month on month, I have a bunch of dry powder, I need to deploy it quickly and competitively, I need to win assets over my competitors, or it's I just need to not leave money on the table for loss recovery, or it's I
22:36
Unidentified speaker (speaker_1): can't stand the fact that we have 20 people's time taken up with every quarterly asset reporting cycle. Make sure you know what your business cares about and work with a supplier whose team is willing to spend time with you to make sure the technology actually delivers that thing. In addition to that, as you mentioned, right, these foundation model apps are going to be pushing out new features all the time.
22:56
Unidentified speaker (speaker_1): So we all would be foolish if we didn't assume that models themselves will get good enough to do most individual tasks like an analyst could. Whether that's updating a waterfall, whether that's creating a basic underwriting model from scratch, whether that's pulling data out of a PDF, all that stuff is already happening. And specificity to a
23:16
Unidentified speaker (speaker_1): particular industry is relatively cheap. Anthropic buys up all of these really promising financial services and commercial real estate professionals to make sure they have the talent in-house to deliver against those ideals. So what we actually think is important again is not just the language model, because large language models still really bad, for example, at math.
23:36
Unidentified speaker (speaker_1): It doesn't matter if you have 99 or 90. 98% accuracy from a large language model. It matters what happens in the other one or two percent. If you're talking about hundreds of millions of dollars, nay billions of dollars that are at stake in transactions and NOI uplifts year on year, it matters when you're wrong, how you're wrong, and how you can debug it. And so for us, we
23:56
Unidentified speaker (speaker_1): actually have a lot of deterministic technology under the hood. We do not just use probabilistic technology because it's not the right tool for every single problem. So one of the innovations out of our RD team from our staff engineers at Fifth Dimension, for example, is what we call our model library. And what that is, is it takes a financial model in Excel and it
24:16
Unidentified speaker (speaker_1): translates it into code, into Python. And it breaks it down into its fundamental components. So if you have an internal rate of return model, there are several subcomponents of that that can be used, that can be QA's. QA'd and QC'd against individual changes. And what that means is, let's say you're doing scenario analysis as part of an investment process, and you would normally
24:36
Unidentified speaker (speaker_1): have an analyst run different scenarios and different Excel spreadsheets and say, oh, change the hold period from seven years to nine years. Assume 80 BIPS more inflation over the period that we're gonna going to hold the investment. And when you're doing something like that, that is a process that usually takes hours or days in order to go through and check every mathematical calculation.
24:56
Unidentified speaker (speaker_1): If you do it with a large language model, you're gonna get nonsense. You're not gonna get what you expect in column L. Column L is gonna be formatted differently, you don't know where the data are from. But since we've broken that down into code, which is inherently deterministic, inherently algorithmic, you will get the same results every time by, for example, changing the hold period from seven
25:16
Unidentified speaker (speaker_1): years to nine years, and you'll be able to do it in seconds.
25:20
Edward Cohen: Collaboration superpower, quick round. If you could partner with one person, historic or living, real or fictional, who would it be?
25:29
Unidentified speaker (speaker_1): Oh goodness. So I think one of my personal heroes is Michel de Montaigne. So Renaissance philosopher, and he basically was the only combatant at the time. against religious fundamentalism. And he essentially created the scientific method. So what an absolute icon who was sitting around, you know, getting
25:49
Unidentified speaker (speaker_1): preached to, proselytized to all the time. And he actually wanted to ask fundamental questions about the way the world worked and get back to first principles thinking. That's how I feel like all of us need to navigate a really volatile time like the one we're living in. Ask the right questions and you'll get the right answers about things like AI.
26:06
Edward Cohen: Kate, work in tangent listeners and real estate investment professionals learn more about Fifth Dimension and connect with you.
26:13
Unidentified speaker (speaker_1): So please do check out our website, go online to fifthdimensionai.com. You can also go to my LinkedIn, send me an email at katefifthdimensionai.com, To fifthdimensionai.com. and you can also tune into our podcast and really get a feel for who we are. Because like I said, you can work with any AI supplier by putting your credit card into the internet. You don't know how to trust
26:33
Unidentified speaker (speaker_1): it, you don't know where your data are going. You don't know how to get value out of the platform, but we're a real human team that cares about real results.
26:40
Unidentified speaker (UNKNOWN): And
26:40
Edward Cohen: you can find a link of that in the show notes. Dr. Kate Jarvis, CEO and co-founder of Fifth Dimension. Thank you so much for coming to Tangent today.
26:51
Unidentified speaker (speaker_1): My pleasure. Thanks for having
26:53
Unidentified speaker (UNKNOWN): me.
27:05
Edward Cohen: You can find links, resources, and ways to connect with us in the show notes.
This page was produced automatically: the audio was transcribed with nvidia/parakeet-tdt-0.6b-v3+sortformer-v2.1 and interpreted by Qwen3.6-35B-A3B-UD-IQ3_S.gguf. Transcription and interpretation can contain errors; always confirm against the original episode. Nothing here is investment advice.
All insights