New utilization of conversational AI (artificial intelligence) and large language models (LLM) in document indexing by Columbus, Ohio-based Razi Title is already providing new efficiency for clients. In an exclusive interview with The Title Report, Razi Title Chief Technology Officer Robert Zwink outlined these additions to his company’s platform, which were made in December. “In our experience, there’s generally always a backlog of paperwork via title plants or data warehouses or starter management systems,” he said. “It all seems to lead back to there being a stack of paper everywhere. In order to streamline that, you have to know what you’re working with. “The ability to use AI to classify documents right down to the page level can help clear out those stacks.As we clear out paperwork, there’s a consumer benefit. When you think about the wait that can happen, or the variability in closings, it’s often a stack of paperwork that you have to get through. AI can streamline operations an...
New utilization of conversational AI (artificial intelligence) and large language models (LLM) in document indexing by Columbus, Ohio-based Razi Title is already providing new efficiency for clients.
In an exclusive interview with The Title Report, Razi Title Chief Technology Officer Robert Zwink outlined these additions to his company’s platform, which were made in December.
“In our experience, there’s generally always a backlog of paperwork via title plants or data warehouses or starter management systems,” he said. “It all seems to lead back to there being a stack of paper everywhere. In order to streamline that, you have to know what you’re working with.
“The ability to use AI to classify documents right down to the page level can help clear out those stacks.As we clear out paperwork, there’s a consumer benefit. When you think about the wait that can happen, or the variability in closings, it’s often a stack of paperwork that you have to get through. AI can streamline operations and provide more predictability in this regard.”
Zwink said the integration of AI in document indexing revolves around two modern technologies: text vectorization and scaled-dot product attention.
Text vectorization involves preparing text for AI analysis through tokenization, or breaking down text into smaller, manageable units. Vocabulary indexing then adds unique numbers to each textual unit before numerical conversion turns the text into a format that AI systems can process.
Scaled dot-product attention refines AI’s focus on relevant text segments within documents. That includes selective attention that zeroes in on sections pertinent to specific topics. Razi’s tech provides contextual analysis that comprehends the significance of words in their specific context, which is then complimented by efficient information-sorting that prioritizes important text.
“It’s effectively an artificial intelligence-based table of contents for PDFs, which is how we’ve described it to title insurance companies,” Zwink said. “We’ve talked to underwriters and large agents; the main reason they’re interested is that it weaves nicely into their existing operational workflow. You can use AI just to create a smart table of contents. Then downstream, you can look at that table of contents and make your existing processes more efficient.
For example, if you have someone who’s extracting information from a Schedule A, they don't have to scroll through all the pages of a PDF in order to find the Schedule A; you can just zip right to that page and show them the Schedule A.”
In the coming years, Zwink said he and Razi would like to see AI tech aimed at the analysis of search packages.
“Streamlining opening orders by analyzing search packages would be great,” he said. “We use this technology to analyze prior policies, but it can also be used to analyze any type of document. Search packages contain cover sheets, transfer history, tax information and more. To be able to summarize that information and open an order automatically, that’s an area where a lot of title offices are looking to optimize.”
When asked about balancing tech and AI advancements with growing cybersecurity threats looming over the title industry, Zwink said increased automation could help in that “fewer hands” have the chance to make a human error.
Security is important when working with existing technology and with emerging technologies,” he said. “When working with emerging technologies, the hardest part is to go from a proof of concept and the potential for an enhancement to actually getting it into production. Starting small is a great way of going about it.
The fewer hands that touch things, the more secure it can be. So, automation, automation, automation. Set aside moments like this and look for opportunities for automation, in order to create greater security. AI can be a tool in one’s toolkit to achieve that.”
