Tavant’s AI-powered Title Analysis solution identifies title risks, extracts relevant data, and
provides lender-ready recommendations—streamlining the closing process and reducing
repurchase risk.
#TouchlessLending #TitleAnalysis #AIinMortgage #MortgageTech
provides lender-ready recommendations—streamlining the closing process and reducing
repurchase risk.
#TouchlessLending #TitleAnalysis #AIinMortgage #MortgageTech
Category
🤖
TechTranscript
00:00Welcome to HousingWire's Demo Day on Demand.
00:03I'm Alison LaForgia, Managing Editor of HousingWire's Content Studio,
00:07and I'm glad that you're joining us.
00:08This is where we spotlight the most innovative technology companies
00:13in housing and mortgage, giving you, our audience,
00:16a front-row seat to real product demos from the teams
00:19building the tools that move our industry forward.
00:22In this session, we're featuring Tavon,
00:25a company that's solving real-world challenges
00:27with smart, scalable solutions.
00:29Stick around for after the demo,
00:32where we'll dive into a short conversation
00:34where I'll ask the presenters questions
00:37that help you connect the dots
00:38between their tech and your business needs.
00:42In this Tavon demo, Kayla and Olav
00:44will be taking us through title analysis touchless lending.
00:48Tavon's AI-powered title analysis solution
00:50identifies title risks, extracts relevant data,
00:54and provides lender-ready recommendations,
00:57streamlining the closing process,
00:59and reducing repurchase risks.
01:02Kayla, Olav, take us through the demo.
01:05Thank you so much, Alison.
01:08So first, we'd like to take a quick moment
01:10to explain what touchless lending is
01:13and how that relates to the title analysis
01:16that we'll be providing.
01:18So Tavon leverages the power of AI technology.
01:23This does include generative AI, agentic AI,
01:27along with multiple OCR strategies,
01:30and advanced large language modules.
01:33This allows us to deliver a truly touchless lending experience
01:38and platform for our customers.
01:40So we intelligently automate complex tasks
01:45such as document classification, data extraction,
01:50decisioning, and borrower analysis.
01:53This helps us to reduce manual effort for our clients
01:57and enhances their speed and efficiency.
02:00It increases their accuracy,
02:02and it also provides compliance
02:04across the life of the loan's lifecycle.
02:07Now, some of the key features include
02:11our touchless document analysis,
02:14decisioning, credit, income, assets, and collateral,
02:19and now we're bringing to you touchless title analysis.
02:24So what is touchless title analysis?
02:28So this, what Tavon has done
02:31is we take AI to read and interpret complex title documents
02:36using intelligent prompts and legal logic.
02:40This also allows us to auto-generate conditions,
02:44conditions for things like resolving liens,
02:47outdated surveys, or unpaid taxes.
02:50And all of this does is to reduce the manual review time,
02:55help our lenders close loans more efficiently,
02:59safer while reducing repurchase risk.
03:02I'd like to take you into the touchless lending platform
03:07so you can see how all of this comes about.
03:11Momentarily, we're going to show you the title analysis
03:14in something called the contract workbench,
03:17and this is our touchless platform where we start today.
03:21Our title analysis is still a concept coming into the platform,
03:25so we're going to show you through the contract workbench
03:28but give you a taste of what it's going to look like
03:31when it enters the platform.
03:36So what you're seeing here is the touchless lending platform.
03:40At the top, you see things that are related to loan details,
03:44just specific information about the loan itself,
03:48including things like the loan number, the loan type,
03:53the number of years on the term of the loan,
03:57the interest rate, appraised values, or borrower's FICO,
04:02all the ratios for the loan like the DTI and the LTV.
04:06We also here are going to see the loan type,
04:08the purpose, what the occupancy is,
04:12along with the purchase price, loan amount,
04:16and borrower's name and subject property address.
04:21Now we can see our analysis for due diligence.
04:26When we click the due diligence link,
04:28this is going to be similar to what you're going to see
04:31with our touchless title.
04:33So you will see a review of certain characteristics
04:36from within the title,
04:38and to the right, you'll be able to see
04:40if it passed or failed,
04:42if it's escalated for any reason.
04:45You can scroll down and see all the areas
04:48that we are reviewing.
04:52So one of the things that the touchless title analysis
04:57will do is create conditions based off any items
05:01that are escalated or essentially failed.
05:05So how you will see those conditions appear
05:07will be as tasks or conditions on the screen
05:10for the underwriter to review.
05:13They can simply open it up to get more information
05:15if they would like,
05:17including any recommendations needed to resolve
05:20that condition.
05:22So let's take a look at what the title analysis
05:26will look like.
05:27So OLIP is going to take you through the title workbench,
05:36and this is going to be the report
05:38that you ultimately see.
05:40So AI has been working to compare the data
05:44and give you this information.
05:47Here we can see that the overall risk
05:51of this particular title is escalate.
05:54We're going to go through to see why.
05:58So there's an executive summary of findings
06:00by each type within the title report.
06:04We can see indicators for pass
06:06as well as those for escalate.
06:09We're going to focus our conversation for escalate
06:13since we understand what an item passing is.
06:16For escalate, we have liens and encumbrances,
06:19property taxes, and survey and boundary.
06:22So let's scroll down and get the details on that
06:25and what our conditions would be.
06:29So liens and encumbrances had two encumbrances
06:32identified requiring resolution.
06:35There was a mortgage to Wells Fargo Bank
06:37with an original loan amount of $285,000
06:41that was recorded in March of 2019.
06:46There was also a mechanics lien
06:48by premier pool contractors
06:50with an amount of $12,500
06:52that was recorded in August of 2023.
06:57So here will be the conditions
06:58that you will see displayed
07:00within the touchless lending platform
07:02for title analysis.
07:04The first condition is to achieve a pass status,
07:08you must obtain a payoff statement
07:10for Wells Fargo Mortgage
07:11and ensure closing funds
07:13include the full payoff amount.
07:17The second condition is to resolve the mechanics lien.
07:21For this, we'll require proof of payment
07:23or lien satisfaction
07:24from premier pool contractors prior to closing.
07:29Next, we can take a look
07:31at our property taxes escalation.
07:33Here we show that 2023 property taxes
07:37show a total assessment of $4,875
07:41with only half of that being paid,
07:44which was the first installment,
07:46and the second half still remains unpaid.
07:49We've provided conditions to pass
07:52which reflect to comply with requirements,
07:55ensure second installment payment
07:56of $2,437.50
08:00is collected at closing
08:02or verify the payment prior to closing.
08:06To achieve full compliance,
08:08prorate the taxes through closing date
08:10and collect appropriate amounts from seller.
08:14And then our final escalation
08:15was regarding the current survey.
08:18It was dated June 15th of 2019,
08:22which exceeds the typical lender requirements.
08:26So the conditions to pass include
08:28to meet current standards,
08:29obtain an updated survey
08:31or surveyors' recertification
08:34dated within six months of closing.
08:37To ensure accuracy,
08:39a new survey should verify
08:40no encroachments
08:41or boundary disputes exist.
08:44Of course, you can certainly scroll
08:46through any of the other items
08:48that pass just to review.
08:50And then you can also see
08:52the recommendations
08:53and required actions
08:54at the bottom of the report.
08:55So this is a recap, essentially,
08:59of those conditions
09:00that were provided
09:01for the escalation items.
09:04Now I'm going to turn it over to Olaf.
09:07Thank you, Kayla.
09:09So just to make explicitly clear
09:12what you just saw,
09:13the touchless lending platform
09:15is in fact a project
09:18or a software platform
09:22that Tavant licenses
09:25and has available.
09:27What I'm going to show you here
09:29is a proof of concept
09:31that we're currently developing
09:33for which the title agent
09:35will be implemented
09:37in this touchless lending platform.
09:40So we're actively developing this today.
09:42And I'm going to show you
09:44how AI helps to create this.
09:48So Kayla showed the result of it.
09:51And think of it as like
09:52you have a preliminary title report
09:55which is 60 pages long.
09:57It gets processed by an AI
10:00and the artificial intelligence
10:02or the large language model
10:03actually finds all those exceptions
10:08and clauses and whatnot
10:10that need to be taken care of.
10:12Now I'm going to show you
10:14quickly how this is being done.
10:17So we created something for this
10:20called contract workbench.
10:22And if I go into the contract workbench
10:24into the title agent,
10:25you see that it always starts
10:28with uploading your source documents
10:31and your policy documents.
10:32So the idea is twofold.
10:35You're going to be analyzing
10:36the document that you want to analyze
10:41but through the lens
10:43of your own policy documents.
10:45What you're seeing here
10:47is something that develops this,
10:50not something where
10:51the final title report analysis
10:54is being deployed
10:55because that would be
10:56the touchless lending platform.
10:58If we go into the build section of this,
11:03we start with a title
11:06and just like a human would do,
11:08you would consider looking at a title
11:12per topic or per clause or per theme.
11:15And we identified 12 themes
11:19that are, of course,
11:20for everyone who would want
11:23to customize this,
11:24that's entirely possible
11:26to put in either your own categories,
11:30merge categories,
11:31add new categories,
11:33or shuffle them,
11:36treat them differently.
11:37But at the end of the day,
11:38we have this title contract.
11:40And if I just open one,
11:43we have one from a large title contract.
11:47It takes this little piece
11:49of the monetary liens and encumbrances.
11:52This as the legal property description
11:55and so on.
11:58Now, the question here is,
12:00how does the AI extract this?
12:03And to that,
12:04we have, just like you have in ChatGPT,
12:07a prompt.
12:09Now, this prompt shows exactly
12:11what the large language model needs to do
12:14in order to extract the content.
12:17And it starts with,
12:18you're an experienced title researcher
12:20working for a mortgage lender.
12:22And you'll be given a title report.
12:24And your task is to extract
12:25relevant portions of text
12:27and organize them into categories.
12:30And then it goes down
12:31into how to extract
12:34the different components.
12:35And here you see,
12:37effectively,
12:37those categories defined.
12:40And not only defined,
12:41but they start very specifically
12:44to say which elements
12:46need to be extracted.
12:49So the nice thing about this
12:50is you can just edit this.
12:53And then given the title report
12:55that you have,
12:56say, like, I run it
12:58and it will extract the results.
13:01You can inspect that.
13:02And if you're not happy with it,
13:04you can go back
13:05and edit your prompts
13:06in a way just like you do
13:08in ChatGPT.
13:09What we edited here
13:11is that you can use AI
13:15to come up with
13:17a new version of the prompt.
13:19Either you refine your prompt
13:21with a hint
13:22or you completely generate it
13:24with AI.
13:25And I'll come back to that
13:26a bit later.
13:28So once you have
13:29your 12 categories
13:31in this case,
13:32you will end up
13:34basically creating
13:37a large language model step
13:40per theme.
13:42And so if we, for example,
13:44look at non-monetary encumbrances,
13:47we have this piece of text
13:50extracted from the larger contract.
13:52And now we want to look at that
13:54two lenses, right?
13:58And here again,
13:59you see a prompt.
14:00You're a senior mortgage title analyst.
14:05Sorry.
14:06You're a senior mortgage title analyst.
14:08You've been given
14:09the non-monetary encumbrances section
14:11for a title report.
14:13Your task is to determine
14:14whether all non-financial restrictions
14:16are clearly disclosed
14:19and compliant with the lender's policy.
14:21So you have here two lenses
14:24through which you analyze the data.
14:26One is a legal smoke test.
14:29And effectively,
14:30what you're trying to do here
14:31is instruct the large language model
14:33to look at this
14:34and understand
14:36which components
14:38are doubtful
14:40or do not conform legal standards.
14:44And typically,
14:46the mortgage providing
14:48or the title providing company
14:50has to, of course,
14:51legally disclose any issues.
14:54However,
14:54they will do so
14:55in rather covered legal language.
14:58And so they will end up
14:59with blanket statements
15:01or vague references
15:02or no mention
15:04of certain obligations
15:05and whatnot.
15:06And a large language model
15:08is excellent at detecting those.
15:10The second lens
15:11for each of those themes
15:13is reference against the policy.
15:16And here we are providing
15:18or you can provide
15:19your own policy documents
15:21and using the retrieval
15:23augmented generation concept,
15:26you can take the piece of text
15:29or the theme
15:30and basically reference that
15:32against your policy document.
15:34And then this large language model
15:37will basically say,
15:39look,
15:40I have given,
15:41I've been given the title.
15:42These are the rules.
15:44Let me apply the rules
15:45and let me see
15:46if the rules are aligned
15:48with what is being shared
15:49in the title.
15:51Also here,
15:52you can basically
15:53play around with this.
15:55You can edit this prompt.
15:56You can get help from AI
15:59to generate this
16:00and refine this.
16:02And then you test the prompt
16:03and you see the results.
16:05And these were the results
16:06that Kayla showed earlier
16:08as the detailed results
16:10leading up to the final report.
16:13So here you see
16:14water rights,
16:15confidence,
16:16and it's all neatly outlined
16:18what is correct
16:19or what is not correct.
16:20And in this case,
16:21the judgment is called pass.
16:25So it is important to note
16:27that for each theme,
16:29you develop your own specific prompt,
16:34if you will.
16:35And then the final report
16:37also uses exactly that setup.
16:41So if I have,
16:43here we see the final report,
16:45but at the end of the day,
16:47this is created by a prompt
16:49and I can look at the prompt.
16:51And now the prompt is
16:52you're a senior mortgage title analyst
16:54working in title insurance
16:55tasked with synthesizing
16:57individual theme analysis
16:59resulting into a comprehensive
17:01title review report
17:02for underwriting decision support.
17:05And here you then start to specify
17:07what this report needs to look like.
17:10Well, that is in a nutshell
17:12what we do here in this workbench.
17:15It allows you to very quickly
17:18fine-tune a title agent
17:21in this case
17:22to your specific requirements
17:25of how you and your business
17:28analyze contracts
17:30and reference them
17:31against the policy.
17:34All right.
17:35With that, I will yield back.
17:38What an interesting look
17:39at title analysis
17:40and how it's eliminating manual work,
17:43enabling greater efficiency,
17:45I'm really interested
17:47to dig in a little bit more.
17:48Are you guys good
17:49if we jump into Q&A?
17:50Absolutely.
17:53So how does AI evaluate
17:56legal and policy compliance
17:58within the title review process,
18:01especially when conditions
18:03can vary across jurisdictions?
18:06That's a fantastic question.
18:08So you can actually utilize
18:10different policy documents
18:11per jurisdiction.
18:13And that's because AI reads
18:15and interprets complex title documents
18:18using intelligent prompts
18:20and legal logic.
18:23And let's dig in a little bit more
18:26to the benefits.
18:29What benefits does the title workbench
18:32offer to underwriters
18:34and closing teams
18:35in terms of speed and accuracy,
18:37which I think we're all focused on right now?
18:39Absolutely.
18:40So one of the greatest benefits
18:44is the speed and accuracy
18:45because it takes the title review process
18:48down from many minutes to seconds
18:51because the review's done for you
18:53and displays those results
18:54like you saw,
18:55either a pass
18:56or a escalation.
18:59And then the conditions
19:00are automatically provided
19:02for the underwriter and closer.
19:03So when they're in
19:04the touchless lending platform,
19:06they can obviously access
19:08the title policy
19:09to review if they would like
19:11or just to compare,
19:12but they will instantaneously
19:14already have
19:15all of the information
19:16they need
19:16that they glance
19:18at those escalations
19:19and the conditions
19:20are provided for them.
19:22That also helps
19:23to reduce the human error
19:24piece of it too,
19:25since we're all human
19:26and could, you know,
19:27accidentally miss something
19:28or misread something.
19:30The AI technology's done that.
19:32It's generated conditions.
19:34So now we're into seconds
19:36instead of minutes.
19:38Or however long
19:39it can take you
19:40to go through
19:40a very lengthy title document.
19:43Absolutely.
19:45So this was really cool.
19:48Let's talk a little bit
19:49about if this is available
19:51for other types of documents
19:53or contracts.
19:55It absolutely is.
19:57We talked a little bit
19:58briefly at the beginning
19:59about how we utilize
20:01generative AI,
20:03agentic AI,
20:05large language models,
20:06and multi-OCR.
20:07So these same types of prompts,
20:11how we generate prompts,
20:12and the use of multi-OCR
20:14to review the documents
20:15and take those pieces out
20:17can be used
20:18for other items too.
20:19So you may see coming soon
20:22fraud, for example.
20:24Little teaser there.
20:26I love a teaser.
20:27Right.
20:28And so I would add to that
20:30this is exactly
20:31what this contract workbench
20:33is,
20:34what I showed you.
20:35At the end of the day,
20:37you provide a sample
20:39of the contract
20:40you want to analyze.
20:41And then basically
20:42it's a set of layers
20:44of AI
20:45on top of each other.
20:46If I were to show you,
20:47for example,
20:49how that would look like
20:51for one second,
20:53I can show this.
20:57Here,
20:58if I import a product,
20:59if I start a new project,
21:01for example,
21:02right,
21:02I have a warranty analysis,
21:04I have a title agent,
21:05but a new project here,
21:06you have to provide the domain,
21:08your role,
21:09the document types,
21:10your objective,
21:11and the expected output.
21:12And at first,
21:13you're thinking maybe,
21:15wait,
21:15why do I need to provide that
21:16if I just want to start
21:18a new project?
21:19But this,
21:20in fact,
21:20is exactly what uses,
21:24what the AI uses
21:25to assess
21:26what questions
21:28to ask you
21:29to see the document
21:31so that you can do
21:32this title analysis.
21:34At the end of the day,
21:35you realize that
21:36these are all prompts,
21:38right,
21:38and prompts are,
21:40in a way,
21:40the new programming language,
21:42but many people
21:43do not know yet
21:45how to write prompts,
21:46and let alone
21:47all the other technicalities
21:49that you need
21:50to kind of understand
21:51on,
21:52say,
21:52how to go
21:53to an open AI,
21:55API,
21:56or how to have,
21:57like,
21:58chaining multiple steps
21:59together.
22:00And that's,
22:00in a way,
22:01what this workbench
22:02provides for you
22:03so that,
22:03as a business professional
22:05or a legal professional,
22:07you can focus
22:07on the prompt development
22:09in order to do
22:11what you do best,
22:12namely,
22:13analyze that contract.
22:14And I believe,
22:15we believe that,
22:16indeed,
22:17the legal professional
22:18is capable of,
22:20with the help of AI,
22:22write very good prompts
22:23to enhance their work.
22:26I love when technology
22:29has a great integration
22:31with practical use cases
22:33just like this
22:34that highlight the capability
22:36to make the humans
22:37more impactful
22:38with the tasks
22:40that they should be
22:40spending time on
22:41and reducing time
22:42on some of the tasks
22:44that eat up
22:45a bunch of their time
22:46and make them
22:46a little bit less efficient.
22:49Exactly.
22:49Absolutely.
22:51Thank you so much
22:53for taking us through
22:54Tavon's title analysis,
22:56Touchless Lending.
22:56For our audience,
22:58for more information,
22:59click the link below.
23:01Thank you both
23:02for joining me.
23:04Thank you, Allison.
Be the first to comment