EY applies search with Elastic to pioneering generative AI experience for finance

For financial institutions, rapid access to trusted information is more important than ever, especially considering increased pressure to comply with evolving regulations like capital adequacy and ESG (Environmental, Social and Governance) standards. For this reason, many banks and their partners are exploring the potential of generative AI to unlock value from vast quantities of unstructured data including pdfs and other documents.

Ernst & Young (EY), one of the largest professional services networks in the world and a leading provider of cutting-edge solutions to the financial sector, is working hard to support banks through these challenges. Vishaal Venkatesh, GenAI Manager, EY, says, "Our primary focus is to champion responsible AI and deliver these solutions effectively to our clients. We ensure they not only understand the technology but also possess the proper framework to implement it successfully."

To achieve this aim, EY built on the Elasticsearch platform using its generative AI solution. "Elastic's cutting-edge work in search and retrieval attracted us," says Venkatesh. "We integrated these capabilities into our AI stack for Retrieval Augmented Generation (RAG), which improves accuracy and accelerates retrieval of insights from unstructured data."

Central to EY's solution is ESRE (Elasticsearch Relevance Engine), a suite of tools that combines machine learning models, data transformation, and a vector database and data search and retrieval.

Venkatesh says, “With ESRE, we can use machine learning models to generate vector embeddings stored at large scale. We used the super-efficient embedding model to build out the RAG component, which significantly reduced our processing time.” EY also took advantage of enhanced indexing and chunking available through the Elastic product suite.

This collaboration enables EY to target two strategically valuable areas of unstructured data in financial services: ESG and financial reports. "Elastic allows us to help banks streamline reporting on their ESG commitments, including internal metrics and details from the full value chain," he says.

For financial reports, EY's AI solution, powered by Elastic, can extract data and insights from complex documents with numerous tables. "Imagine extracting 14 key variables from a 40-page pdf or comparing information across multiple reports from different years. This is where we save clients significant time and resources," says Venkatesh.

In addition to delivering cutting-edge technology, Elastic offers seamless integration with other components in EY's generative AI stack, including LamaIndex, a data framework for building LLM applications, LangChain, and open-source vector databases such as LanceDB.

EY observed significant improvements in three key areas following its deployment of Elastic:

• Approximately 10-15% gains in accuracy across various document and extraction methods;
• Elastic's RAG setup was able to produce results 3 times faster when compared to Native RAG approaches;
• Improved scalability, demonstrating the ability to handle diverse document formats and seamlessly scale across various organizational units.

Speed to value was another crucial advantage. "Traditionally, integrating vendor solutions presents a significant challenge due to lengthy ramp-up periods," says Venkatesh. "However, Elasticsearch's unique infrastructure significantly accelerated our development process, enabling us to realize the benefits much faster."

Venkatesh adds that Elasticsearch's core strength lies in its ability to perform large-scale information retrieval while simultaneously enabling the integration of diverse AI technologies throughout the entire development lifecycle. This integrated approach extends to offering vital observability and security functionalities, ensuring optimal performance across EY's custom-built AI pipeline.

"By leveraging a robust and secure solution like Elastic, we eliminate the need to invest significant resources in building and maintaining these critical components from scratch," Venkatesh says. "Furthermore, relying on a proven and trusted solution like Elastic mitigates the risk of unforeseen issues with homegrown components."

Throughout the project, EY was able to call on the support of Elastic account managers and technical experts whenever necessary. "Our collaboration with Elastic proved to be exceptionally smooth due to the strong foundation established from the outset. The combined efforts of Elastic solution architects and account managers ensured we could swiftly navigate the complexities of generative AI and achieve rapid prototyping," says Venkatesh.

Above all, it helps keep EY at the vanguard of the generative AI revolution. "With Elastic's powerful search and retrieval functionalities, combined with machine learning for vector embedding, we're ideally positioned to equip our clients with responsible generative AI tools that further innovation, unlock value, and mitigate risk."