I recently had the chance to chat with Victor Menasce on The Real Estate Espresso Podcast by Y Street capital to discuss a significant development in the mortgage industry: the integration of Artificial Intelligence (AI) into real estate loans. At Addy AI we're focused on reducing manual workloads for loan officers and private lenders, I’m excited to share how AI is not just a trend but a transformative tool that’s streamlining the entire lending process. Here’s what I discussed and why AI is the future of real estate loans.

Listen to my full conversation on The Real Estate Espresso Podcasts' Spotify link <u>here</u>

The Role of AI in Real Estate Loans

AI technology is becoming increasingly essential in real estate loans, primarily by reducing the time-consuming manual tasks that have traditionally bogged down the lending process. Mortgage loans, for example, can take anywhere from 45 to 60 days to close, with loan officers often only managing one or two loans per month. The bottleneck largely stems from the extensive document review, due diligence, and manual tasks such as title searches that are necessary for each loan.

By leveraging AI, loan officers can significantly reduce the time required for these processes. AI can automate credit policy checks, extract relevant data from documents, and streamline the addition of loans to the lending pipeline. For real estate investors, this means a much smoother experience, while lenders benefit from a more efficient process that allows them to handle more loans in less time.

Streamlining the Mortgage Lending Process with AI

One of the most significant advantages of AI in real estate loans is its ability to automate the initial stages of the mortgage lending process. When a loan scenario comes in, AI can quickly determine whether it meets the lender’s credit policy, saving loan officers the time and effort of manually reviewing each case. For instance, if a loan does not meet the lender’s criteria, the AI system can suggest adjustments to make it eligible, thus increasing the chances of approval.

Moreover, AI can handle complex calculations such as Loan-to-Value (LTV) and Combined Loan-to-Value (CLTV) ratios, which are critical in determining the eligibility of a loan. This automation not only speeds up the process but also reduces the likelihood of human error, ensuring that loan officers can focus on higher-level decision-making.

Document Processing and Data Extraction

In the mortgage industry, loan officers often deal with hundreds of pages of documents that need to be reviewed for specific information. This process can be incredibly time-consuming and prone to errors. AI can revolutionize this aspect of real estate loans by automating the extraction of relevant data from these documents.

For example, when a loan officer needs to verify the entity behind a loan application or find the authorized signer, AI can quickly scan the documents and present the necessary information. This capability drastically reduces the time spent on document review, allowing loan officers to process more loans and provide better service to their clients.

AI and Title Analysis

While AI has made significant strides in automating various aspects of the mortgage lending process, there are still areas where the technology is evolving. Title analysis is one such area. Titles involve complex legal considerations, such as ownership rights, encumbrances, and the history of ownership. While AI can assist in analyzing these aspects, it is not yet capable of handling the entire process end-to-end.

However, AI can still play a crucial role in title analysis by providing insights into the chain of ownership and identifying potential issues that may require further investigation. As the technology continues to develop, we can expect AI to take on more responsibilities in this area, further streamlining the mortgage lending process.

The Next Frontier for AI in Real Estate Loans

As we look to the future, the next frontier for AI in real estate loans will likely involve achieving regulatory compliance and ensuring the accuracy of AI-generated decisions. Currently, AI models are capable of synthesizing large amounts of data and generating valuable insights, but they are not infallible. One of the challenges with AI is its tendency to "hallucinate" or produce incorrect results based on incomplete or outdated information.

To overcome this, future developments in AI will focus on making these tools regulatory compliant. This will involve rigorous testing, quality assurance, and the implementation of safeguards to prevent errors and biases. Additionally, there will be a push towards integrating AI with live data sources, ensuring that the information used to make decisions is current and accurate.

The Impact of AI on Loan Officers and Lenders

The integration of AI into the mortgage lending process offers significant benefits for both loan officers and lenders. For loan officers, AI tools can handle much of the repetitive, time-consuming work, allowing them to focus on providing personalized service to their clients. This can lead to higher job satisfaction and the ability to manage more loans simultaneously.

For lenders, the efficiency gains from AI can result in cost savings, faster loan processing times, and improved customer satisfaction. By automating tasks that were previously done manually, lenders can reduce the risk of errors and ensure that they are complying with regulatory requirements.

Conclusion

The application of AI in real estate loans is not just a trend; it is a fundamental shift in how the mortgage lending industry operates. As AI technology continues to evolve, it will play an increasingly important role in streamlining processes, reducing workloads, and improving the overall efficiency of loan officers and lenders. My experience as the CEO of an AI startup in this space has shown me firsthand the transformative potential of these tools, and I am excited to see how they will continue to shape the future of real estate loans.

As we move forward, the key to success will be ensuring that AI tools are accurate, reliable, and compliant with regulatory standards. By doing so, we can unlock the full potential of AI for real estate loans, creating a more efficient and effective lending process for everyone involved.

Full transcript of my conversation with Victor Menasce

**Victor Menasce:** 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 Menasche. This is the weekend edition where we interview notable people from the world of real estate investing. Today is no exception. We have a great guest all the way from San Francisco, California.

Welcome to the show. Michael Vandi. 

**Michael Vandi:** Thank you so much. Happy to be here. 

**Victor Menasce:** Well, great to have you here. Now, Michael, I'm excited for this conversation because we're going to be talking about something that is certainly making headlines and that's AI,  specifically as applied to a segment of real estate.

But before we do, maybe give a little bit of your backstory and how you got to this point in your journey. 

**Michael Vandi:**  thank you so much, Victor.  so I'm the CEO of  we're bringing AI into the mortgage lending process. I have experience working on enterprise technology,  at Amazon Web Services. And,  I also did artificial intelligence and software engineering at Carnegie Mellon University.

And,  in the start of the whole AI revolution, I spent a lot of time with,  mortgage lenders and mortgage brokers shadowing them and looking at,  their workflows. And that was when I discovered,  an interesting,  hole in the market that AI would fill. And so I built a team of my friends from Carnegie Mellon at the time, and then we started this company.

So my background is specifically in AI. And based on the experience that I have working with mortgage lenders and brokers, we decided to build a company. 

**Victor Menasce:** I love it. Well, maybe after this, we'll talk about some friends that we may have in common in that space. So, There's no question that,  machine learning algorithms can be applied to all kinds of different things.

And especially the type of research that kind of due diligence that a human would do, you can certainly, if the data's out there, it's literally just a scavenger hunt, establishing the right criteria.  for,  for an algorithm to do that heavy lifting, and maybe in some cases, just the donkey work to accelerate that process.

 where have you aimed the technology and where do you see the most value? 

**Michael Vandi:**  so the most value. Is in short shortening the closing time, so it takes anywhere from 45 to 60 days to, you know, close a mortgage loan. And most loan officers actually only do like 1 or 2 loans a month. And,  if you are a real estate investor applying for a loan.

You've gone through like numerous processes where it's like a lot of back and forth with,  the lender or the broker. And,  most of that work on the lending side. Usually it's a bunch of document review, a lot of due diligence,  application review and manual tasks like title searches and a constant followup with brokers.

So where we're aiming the technology is being able to train on all the existing data and then say. This is,  our credit policy that we lend against and being able to filter all of the scenarios that come in and help loan officers with the tools that they need to make decisions faster. So, closing the,  shortening the closing time from like 45 days to half of that.

 so if you are a real estate investor, You get a much smoother experience. And if you're a lender, you get to have a more efficient process. 

**Victor Menasce:** So what aspects of that can you actually streamline with an AI tool? Which parts are you cutting the human manual labor out of the process? 

**Michael Vandi:** Right. So with credit policy checks, you know, when scenarios come in, you have to look at them.

First of all, you check the location, you know, do we lend in this location? And,  What type of loan are they looking for? Is this part of the products that we offer? And if it's not, then you actually have to tell them that, hey, do XYZ if they're ineligible to make them eligible. So AI can handle that initial process.

So doing the automated credit,  credit policy checks. And then when it gets to the point where it looks like this person is eligible, you would want to extract their loan opportunity and then add it to your lending pipeline. So, for instance, you want to get, like, which broker they're working for, what's the loan duration, you know, do things like, you know,  LTV and CLTV.

So AI can do that,  those,   calculations automatically and then extract the relevant data from the documents. And then present it to you, and then you can go ahead and add it to your lending pipeline. Another interesting thing is, a lot of people apply for,  loans on behalf of other people, so you want to look at, like, who's the entity behind it?

Who's, like, the authorized signer? And so if a loan officer has, like, hundreds and hundreds of pages in front of them, and they want to find specific information, they have to spend minutes or hours going through all of that. of that. So all these,  document processing is what you can bring AI into the process for 

**Victor Menasce:** fascinating.

So what you described, I'm going to translate what you said into,  one of our own company's internal processes. And that's in the first phase of due diligence. Excuse me. In the first phase of due diligence, we have what we call a quick kill list. So it's that first set of criteria that will disqualify a project.

It's not everything you need to qualify a project, but if you don't meet these first 15 or 20 criteria, there's no sense looking at the next 80 or a hundred. And, and so it's that, that quick kill list to disqualify a project quickly with a minimum of effort. Is, do I have that right? 

**Michael Vandi:** Yeah. Yeah. You have that right.

 and you pick up a very interesting point where most lenders actually reject the majority of the scenarios that come in. We've worked with lenders that reject like 95 percent of the scenarios they're presented with. So if the loan officers, 95 percent of their time is just spent reviewing scenarios that they're eventually going to reject.

Well, why not? You know, have AI like surface that for them and for instance, in your case, having a kill list and then training an AI model to fully understand what that kill list is and then cross reference that against,  the scenarios that you have and then present that to you. 

**Victor Menasce:** You mentioned something about title and,  this is one of those areas that I think has a lot of potential.

But it also has a lot of complexity. So, for example, you know, title gets described, if you think about the legal description of a, of a property, it requires a lot of actual thought. Like a surveyor needs to look at the title report and determine, well, is this easement, is this utilities easement going to be a proper, A problem or not.

And where is that utilities easement in the hierarchy of other things that might have been recorded either before or after,  there's a lot of complexity. Is that within, in your view, is that within the realm of something that an AI tool can, can do? Easily navigate. 

**Michael Vandi:** Yeah.  so the technology is not quite there yet to handle the process end to end.

 so title has a lot of legal,  entities to it, like, you know, owner's rights,  like who has the right to sell, who has the right to lease, who has the right to use and like,  you know,  any,  encumbrances as well. So what we do,  I think like. In the future, AI can handle all the legal things that happen around that process, but I still think it's like at least five years away from it handling the process end to end.

So,  one of the things that you mentioned is,   handling the complexities around that. So for instance,  ownership history. So for that, you can have AI, once you have the ownership history, you can have AI like analyze that ownership history. Like you can show like the chain of ownership over time and all these complex analysis that you can do once you have the type as well.

So it's more like title analysis and less so like a complete title searching. 

**Victor Menasce:** And presumably the value proposition For, I mean, your customer is going to be essentially the lender. It's going to be the loan officer that's looking to save time by processing,  a larger number of files on a smaller number of individuals.

And it's that cost savings and quality improvement that is the result of throwing this tool at, at that large data set. 

**Michael Vandi:** Yes. 

**Victor Menasce:** I love it. Where do you see the next frontiers for this technology? I mean, in the world of software development, and I know this isn't software, it's machine learning, which is a little different, right?

But, but there still are,  Let's say formalized releases where you say, okay, the tool has been vetted to a certain standard. And in this release, it's capable of doing X, Y, and Z. Where do you see the next frontier over the next 12, 18 months? 

**Michael Vandi:**  let me think about that question for a little bit. So if we look at.

What AI is really good at,  and by AI, I mean like large language models, it's synthesizing large amounts of data and then trying to, trying to generate whether it's like images, text that's related to that, that data. The good thing about it is that, you know, it can shorten the amount of time that you need to review some data, but the bad thing about it is like, it hallucinates a lot and hallucination, I think, is 1 of the things that, um.

Would stop this technology from being, let's say, regulatory compliance. So I'm thinking next 18, the 18 to 24 months. The next frontier would be getting these tools to be regulatory compliant. So for now, these tools are usually in their infancy, you know,  running pilots and then also using them as,  early adopters who really believe in the technology.

And as a tool start to get mature, then comes the point where you have to do evaluations and, you know, regulatory compliance, like bias and,  and fairness, you know, transparency. And that might look In different ways, it might mean like a human loan officer, human reviewer is actually checking the results against some sort of like quality assurance, or it might be another AI checking the AI to see if the evaluations are correct.

So I think the next frontier actually is checking these tools for accuracy and correctness, whether it's in like the mortgage lending field or it's in like,  finance,  the broader financial services sector. 

**Victor Menasce:** In my experience, one of the areas where. The large language models still struggle a lot is in identifying the boundary of a context.

I'll give you a simple example. I asked an AI tool to edit something that I had written. In fact, it was a podcast and, and what it did is it took something that was a very specific statement and turned it into a generalization because it didn't understand the specific example was I'd made the statement that From 2008 to 2012, property values in Cape Coral had fallen by 60%, which was absolutely a true statement.

It then generalized that to say that property values had fallen by 60 percent across the United States, which was not a true statement. And, and so it, it failed to connect, failed to identify the boundary of the context of the statement, and then simply reapplied it somewhere else where it was not appropriate.

And I think that's one of the areas that is, The tools still struggle tremendously with is identifying the boundaries of those contexts. 

**Michael Vandi:** It's, it's, it's actually just like baked into the,  the architecture of transformers. Because all they do is predict the next word. So when you,  give the example of property values in a particular, let's say, place x,  you know, had gone down, then if you replace x with y, then it's just gonna guess like the next word.

Is property values are going down. It's going to have that context. It's like, Oh, well, I think if property values are going down in this particular place and property values must be going down and, you know, other,  other places as well, which is incorrect. And the way you actually solve that is with a bunch of context and also giving the model access to live data.

 I'm not sure if the system that you use at the time had access to to live data from the Internet, because before presenting the data to the user, you have to cross reference it against,  live Internet data and then checking is if you just come with the model based on the information that it's been trained with, it's going to be very dumb and only have information  backdated to its training time, which could be two years, three years.

So,  making the model come online and having internet data, I think it's one of the ways you could solve that. 

**Victor Menasce:** But then presumably the, the compute power required to solve for a complete solution is, is exponentially much larger. 

**Michael Vandi:** Yes. 

**Victor Menasce:** Well, fascinating. Well, Michael, if,  folks want to connect, if they want to learn more, what's the best way?

**Michael Vandi:**  the best way to learn more would be going on,  our website and booking a demo and seeing firsthand.  how it works and how the future of this technology would look like and,  real estate and mortgage lending, or I can be connected with on LinkedIn. It's just my name, Michael Vandi. I think I'm the 1st,  I'm the 1st name that pops up.

So I'm always happy to chat 

**Victor Menasce:** and the website for your business, 

**Michael Vandi:** it's addy. so. 

**Victor Menasce:** Well, Michael, love what you guys are doing and for the listeners at home, definitely connect with Michael Vandi at addy. so the link 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 again tomorrow.