Pod Insights The Real Estate Espresso Podcast AI Document Verification with Lisen Kaci About this episode Lisen Kaci is the lead developer of Discrepancy AI, a suite of tools that provide financial document analysis and verification. On today's show we are talking about the state of the art in using AI tools to streamline the financial document due diligence process. To learn more, visit discrepancyai.com.-----------------**Real Estate Espresso Podcast:** Spotify: [The Real Estate Espresso Podcast](https://open.spotify.com/show/3GvtwRmTq4r3es8cbw8jW0?si=c75ea506a6694ef1) iTunes: [The Real Estate Espresso Podcast](https://podcasts.apple.com/ca/podcast/the-real-estate-espresso-podcast/id1340482613) Website: [www.victorjm.com](http://www.victorjm.com) LinkedIn: [Victor Menasce](http://www.linkedin.com/in/vmenasce) YouTube: [The Real Estate Espresso Podcast](http://www.youtube.com/@victorjmenasce6734) Facebook: [www.facebook.com/realestateespresso](http://www.facebook.com/realestateespresso) Email: [[email protected] ](mailto:[email protected] ) **Y Street Capital:** Website: [www.ystreetcapital.com](http://www.ystreetcapital.com) Facebook: [www.facebook.com/YStreetCapital](https://www.facebook.com/YStreetCapital) Instagram: [@ystreetcapital](http://www.instagram.com/ystreetcapital)
Transcript 0:02 Speaker 00 Welcome to the Real Estate Espresso Podcast, your morning shot of what's new in the world of real estate investing. I'm your host Victor Manash. This is the weekend edition where we interview notable people from the world of real estate investing. Today's no exception. We have a great guest all the way from Toronto. Welcome to the show Lisen Casey.
0:17 Speaker 01 Hi Victor, thank you for having me.
0:19 Speaker 00 Well, great to have you here. Well, Lisen, you are in the world of due diligence when it comes to helping landlords and property owners and lenders work with potential clients. And I'm keen to have this conversation because an awful lot has changed on multiple fronts, certainly on the regulatory front in terms of compliance, what you can and cannot ask is very much dependent on jurisdiction. And then how you ask and then what you do with the data, all of those things come into play. But maybe before we do, perhaps give a little bit of your backstory and how you got to this point in your journey.
0:53 Speaker 01 Yeah, well, my background is as a software developer and AI engineer, my background has always been working with companies, building AI solutions and even before ChatGPT, helping build machine learning and AI solutions for different problems. And that's where I was working in the AI document processing space for medical documents. And that's where I started seeing kind of like where the space was, where it could improve, and how right now, how we're dealing and handling and processing these documents and what we're doing with them. After that, I decided I wanted to like solve that problem. So I started discrepancy AI. And I specifically wanted to solve that problem for financial documents. I felt they were the
1:37 Speaker 01 most left behind when it comes to like computer vision, ML analyzing and extracting the data. Things like tables, charts were not working really well with kind of like the latest cutting edge technology for automating that. And there was a lot of human review required, which is expensive and time consuming. So that's why I built discrepancy specifically to automate the entire document processing workflow for financial or property companies, helping them review loans or rental applications or anything where the review of financial data from your customers is important. We can automate that and get it to a really high 99% plus rate where
2:19 Speaker 01 you can be confident it's faster, cheaper, and better than kind of like the current solutions.
2:25 Speaker 00 I think there's a few issues with that whole process. One of them is just the transmittal of information. Often when we're dealing with lenders, they will, you know, give you a list of 20 items that they want. If you send it as attachments, they're asking the same question multiple times because they lost it. And then, of course, if there end up being multiple versions, that you end up with a disease that I call versionitis, where they're referencing the wrong version of the document because there's no version control on their end. Can you talk a little bit about that part before we dive into the details of actually even looking at the documents?
3:04 Speaker 01 For sure. That's super important to me. Kind of the tagline that I use for discrepancy is we want to be the Stripe, but for document processing. We want to build these tools to make it really easy for businesses to capture and process documents from their users. So the entire flow. So like with Stripe, you can copy and paste something on your website and now you can process credit cards. It's the same thing with discrepancy AI. You can paste something on your website and now you can process documents. You can use natural language to kind of specify, hey, I want X, Y, and Z documents. And then we, our paradigm that we're using is we structure at the point of capture. So as the user is interacting with the AI uploading these documents, the analysis and the structure is happening right there.
3:47 Speaker 01 So by the time these documents reach the lender, whatever the customer, you know that all the documents that are required are there. They're all structured, they're all being categorized, ordered, listed, you can search through them. All the text has been extracted. What we're trying to do is that by the time that these documents enter your system, they're already structured for you.
4:08 Speaker 00 Now, I have a tech background, not all of our listeners do. So when you say structured, I know exactly what that means. But you're talking about taking what is in the world of big data, meaning unstructured data, and then putting it in a format that perhaps even a database can recognize. When, are there specific fields that maybe a lender or a property manager might be looking for, looking to populate, extracting from this unstructured data?
4:36 Speaker 01 Yeah, so the the number one most kind of like across the many industries is income verification. So that is something that's incredibly important that a lot of different industries look for. And what makes our kind of system unique is that it can, again, I'm going to get just a little technical, is it can give you custom JSON. So you can specify the structure that you need the data to be in. So and the AI will actually calculate anything it needs to calculate and return your document in that structure. So it can be handled by an API, by your database. If you have a multi-step, like if you're a lender, you have probably like many, many steps in the analysis of those documents.
5:17 Speaker 01 And the whatever that other API that you're using has a certain JSON that it expects. Our tool can can kind of be customized to return any JSON. That way it can it can seamlessly kind of integrate into any system or any other flow that that these companies have set up.
5:34 Speaker 00 Now, I'm going to go out on a limb here. I'm going to imagine for a moment that I'm a loans officer. By the way, I would never be a loans officer, so it's going to take a stretch for a moment. But imagine for a moment that I'm a loans officer and I've issued a commitment letter that might have, I don't know, 20 preconditions. Will your software literally check the box and say, this document here satisfies requirement number eight? And this other document fully satisfies requirement number 11 and so on? Is it going that far?
6:05 Speaker 01 Correct, yeah, it will it will look through those, see if those requirements are satisfied. And then actually give you the exact structured data for where those requirements are. So if you say, I only want a customer with an income of 100K. It will tell you that a requirement that you made and their income is 150K. So it will verify, categorize, classify and structure the data. You just specify what you want the return structured JSON, I'm doing air quotes, JSON data to be like. Then it will return it to that. So the difficulty there is kind of like, I work kind of closely with a lot of our customers to to make sure that that structured output is as expected.
6:47 Speaker 01 And our no-code platform, you can go on there, you can like play and test it. And then we have like no-code snippets that match kind of like what you tested. So you always can make sure that it matches your your needs, your requirements.
7:01 Speaker 00 Now, I think about complex cases where, for example, income might be coming from multiple sources. Maybe someone has a full-time job, maybe they're making money on the side by delivering pizza at night, or whatever it might be. Are you able to assemble that together to produce a complete picture?
7:18 Speaker 01 Yeah, so right now, we can handle multiple documents and up to 600 pages in one analysis. So if it's within 600 pages, even if it's like 10 documents, right, like you can, you can submit all of those at once, and you can look through those, and you can even return kind of like, hey, you're missing this, you're missing that, and then here is the structured output from these 10 documents based on your requirements. So, so yeah, I can handle very complex stuff. And in fact, I built this specifically to handle tables, charts, financial data, which has been kind of like left behind. The, the state-of-the-art technology is called OCR. So with current OCR technologies for reviewing documents, kind of financial
8:00 Speaker 01 documents do not work as well. So that's why I wanted discrepancy AI to really solve that problem.
8:06 Speaker 00 Yeah, character recognition is just image to text, which is not really applying a lot of intelligence to the problem. You're talking about taking it a step further and actually making value judgments based on the content of the data, not just converting it into something machine-readable. Now, when I think about a property manager that might be accepting, they have two vacancies and they get 100 applicants, that's a lot of data to sift through. And how do you choose? And in fact, what criteria are you using to choose? Is it just, all right, out of the 100 applicants, 10 meet the criteria, and then a human gets involved? Or how do these decisions get made?
8:50 Speaker 01 That's a really great question. And the answer to that, no, we do not choose. Our solution is set up where our customers can set up these pass fail requirements based on their needs. And we never make that selection. We rate them, we rank them, we provide kind of like a pass or fail based on the requirements that they have set up. Our AI is never kind of like making these determinations on whether someone is appropriate or not appropriate. We're simply organizing the documents that those people provided, based on the requirements that that customer gave.
9:22 Speaker 00 Terrific. Well, Leeson, I love what you're doing, and I think this is actually fulfilling a very real need in the marketplace. Have you had any discussions with major property management firms about integrating your capability within those platforms?
9:36 Speaker 01 Yeah, yeah, absolutely. I'm currently in talks actually with integrating into several PMS systems. It's something that a lot of landlords do not want to go on a different system on a different platform. They're already using one property management system that they love. So they're like, yeah, you're definitely solving my need, but like, can it work in my PMS? So that's currently coming up in the works and that's coming up in the next few weeks.
10:04 Speaker 00 Awesome. Well, Lisen, if folks want to connect, if they want to learn more, what's the best way?
10:09 Speaker 01 Yeah, you can go to discrepancyai.com. You can try our product for free and you can just try the AI on our website without even making an account, really. You can just upload a document right on the front page. You can search me on LinkedIn at Lisen Casey, L-I-S-E-N-K-A-C-I. Yeah, hit me up, send me a message, and I look forward to connecting with everyone.
10:33 Speaker 00 Fantastic. Well, Lisen, great to connect and love what you're doing. And for the listeners at home, definitely connect with Lisen Casey at discrepancyai.com. The links will be in the show notes. And in the meantime, have an awesome rest of your weekend. Go make some great things happen. We'll talk to you again tomorrow.
Previous Hotels Versus Short Term Rentals Next Security Methods That Work Page updated November 15, 2026 · transcript and summary are machine-generated