Case Study: First American achieves faster, more accurate data extraction with Databricks

First American Data & Analytics embraces the power of GenAI

First American Data & Analytics (DNA) needed a faster, more accurate way to turn millions of historical title policy images into usable digital records. The team faced the challenge of page-level classification and extracting 100+ data fields from more than 4 million mostly unstructured documents with little metadata, a process that had long depended on manual keying and complex OCR workflows. First American partnered with Databricks and used Databricks Mosaic AI to improve LLM-based data extraction.

Databricks helped First American experiment with multiple LLMs and prompting approaches, then quickly fine-tune a model using domain expertise from the DNA team. The result was improved model accuracy and reduced training time compared with earlier approaches that required large datasets and lengthy fine-tuning. This enabled First American to more efficiently extract and link historical policy data, helping expand its Starter Xchange and accelerate real estate title research and transaction processing.

First American

Paul Hurst

Chief Innovation Officer