Leveraging Language Models for Efficiency and Accuracy in Title Insurance Underwriting

September 13, 2023

By Lili Farhandi and Robert Zwink

Title insurance professionals hold the crucial responsibility of ensuring the validity and accuracy of real estate title documents. Large Language Models (LLM), like ChatGPT, are capable of processing large volumes of data and extracting valuable insights. Processing happens in one of two ways, either by fine tuning a large language model with pre-existing text or by providing the text as “context.” This article explores the efficiencies and risks of LLMs, and dives into the indispensable role that they will inevitably play in the industry.

The Power of Fine Tuning: Gleaning Insights from Pre-existing Text

Picture an artificial intelligence (AI) system capable of responding accurately to inquiries in the English language, for instance, “What risks are associated with 123 Main Street?” Once the question is received, the AI analyzes every pertinent document in the county courthouse, culminating in a succinct and accurate Schedule B.

Achieving this level of precision requires an AI system, like an LLM, that has been fine-tuned on a robust dataset—say, the past 40 years of recorded documents in relevant county offices. This would involve training the AI model on this dataset so that it can learn from the patterns, trends, and anomalies contained therein, and then apply this understanding to new, unseen data.

This isn't science fiction. Today, such technology exists, as exemplified by OpenAI's ChatGPT, or even Razi's TitleGPT. The benefits of fine-tuning a large language model in this way are immense. It can yield highly accurate responses, save time, and alleviate the manual labor involved in scouring through countless documents.

Leveraging Context: The Middle Ground

There is another way to extract value from LLMs, especially when a comprehensive, fine-tuned training on relevant title documents isn't available. Here, the LLM is provided with information in the form of context. The English language questions are formulated in such a way that they contain the necessary text for an accurate response.

Consider this question: “What is the premise address of the property referred to in this policy? <Insert Policy Text>”. The LLM can then correctly answer, “123 Main Street.” Although this approach does not take into account documents beyond the given context, it still has its merits.

Innovative companies like deepset.ai have developed ways of dynamically creating context based on a list of relevant documents. This is called dynamic context, which allows only pertinent documents to be used as context, thereby achieving a middle ground between single document context and the full power of a fine-tuned model.

This method carries significant potential for improving efficiencies and reducing error rates in title insurance underwriting. By leveraging dynamic context, title insurers can draw from a curated pool of relevant documents, improving the LLM's ability to provide accurate responses without the need for comprehensive pre-training.

The Uncharted Terrain: Risks and Mitigation

While leveraging LLMs in title insurance underwriting offers compelling benefits, it is crucial to be mindful of the risks involved. With the automation and algorithmic nature of LLMs, there is a risk of systemic errors, where a single error in the model could result in widespread inaccuracies. Furthermore, AI systems like LLMs might miss nuances and subtleties that a human would easily pick up on. Therefore, mitigating these risks and integrating robust oversight measures is of paramount importance.

In Conclusion: LLMs and the Future of Title Insurance

The title insurance industry is at the brink of a significant transformation, with LLMs providing an opportunity to bring about groundbreaking efficiencies and accuracy. By fine-tuning LLMs on pre-existing text or leveraging context-based techniques, title insurers can extract tremendous value, reducing labor costs and improving productivity.

However, as with all technological advancements, caution should be taken to ensure accuracy, validity, and reliability. Measures should be put in place to mitigate risks and continuously validate and adjust these AI models.

The integration of LLMs in the title insurance industry is more than just an emerging trend—it is an inevitable progression towards the future. As technology continues to evolve and advance, so should our practices, ensuring that we leverage the benefits of these powerful tools while carefully managing the potential risks.

In an industry as vital as title insurance, where the stakes are high and the margins for error are low, the adoption of LLMs represents a powerful step forward. It paves the way for a more efficient, accurate and reliable future—benefiting underwriters, insurers and customers alike.

Lili Farhandi is chief executive officer and Robert Zwink is chief technical officer of Razi Title. As co-founders of the company have expertise in integrating title processes with modern technology. Their combined skills have guided Razi Title's use of Large Language Models to increase efficiency and accuracy in title insurance underwriting. They can be reached at contact@raziexchange.com.

Contact ALTA at 202-296-3671 or communications@alta.org.