SBI Life Insurance develops a document search solution with Amazon Kendra and generative AI, and streamlines call centers with a selfbot

SBI Life Insurance, an SBI Insurance Group company, provides consumers with products such as an affordable mortality insurance plan, which allows customers to revise the sum insured based on their stage of life; a disability insurance plan that features multiple coverable risks; and a medical insurance plan that covers from inpatient treatment to home care. The company also offers group credit life insurance and group credit disability insurance to financial institutions.

The Information Systems Department of SBI Life Insurance is responsible for security response in addition to development, maintenance, and operation. From requirement definition to internal operation, the team continually builds systems in-house to streamline business.

“When it came to compliance or developing new products, our company tended to modify and use the existing applications,” says Toru Ikeyama, the head of Information Systems Department, Director and Executive Officer, SBI Life Insurance. “This approach created technical debt. We therefore decided to migrate our applications to a new technology platform using AWS.”

SBI Life Insurance needed to enhance customer service at its call centers and shorten the training period for operators. For example, the operators sometimes have to answer inquiries about discontinued life insurance products. SBI Life Insurance has products that were discontinued over 10 years ago, each with an enormous variety and volume of related documents, so answering these questions quickly was a challenge for operators. This situation made operator training a lengthy process.

To help solve these challenges, the Information Systems Department started building a solution that let call center workers easily search for documentation on insurance products and policies. Initially, the team considered using Amazon OpenSearch Service, a full-text search and analysis engine.

“We wanted to go beyond simple keyword matching and use machine learning to provide FAQ search results that reflected the user intent,” says Ikeyama. “However, adding these functions to Amazon OpenSearch Service would have increased development workloads and impacted the project timeline.”

In February 2023, Amazon Kendra, an intelligent enterprise search service, became generally available in the Asia–Pacific (Tokyo) Region. Amazon Kendra provides the flexibility to search for information in structured and unstructured content that is scattered across multiple locations, easily bringing convenient semantic searches to users who don’t have specialized machine learning knowledge. The Information Systems Department decided to build a document search system using Amazon Kendra at its core.

In addition to developing the solution, a key part of the project was to convert brochures for old insurance products and procedural documents for contracts into PDF format. Amazon Kendra uses natural language processing technology to read unstructured data in PDF files as a source. It crawls the source data saved in Amazon Simple Storage Service (Amazon S3), which is an object storage built to retrieve virtually any amount of data from anywhere, to create a search index.

The team started to develop the solution in April 2023 and released it to a production environment in July. “Amazon Kendra realized a development in a short period,” says Ikeyama.In September 2023, the team released a selfbot function that summarizes and displays Amazon Kendra search results using generative AI. The operators simply read the summary from the selfbot to respond to customer inquiries. To summarize the information, SBI Life Insurance employs several large language models, one of which is Anthropic’s Claude through Amazon Bedrock, which is a fully managed service that offers a choice of high-performing foundation models from leading AI companies.