Finatext and Resona AM have built a platform to automate the preparation of investment trust documents using Snowflake and generative AI. It features 4 functions: natural-language search, an editor, batch change processing and automatic generation of before-and-after comparison tables.
The asset management industry is being pressed to transform itself as Japan works toward becoming an "asset management nation." The preparation of investment trust documents still relies heavily on manual work in Word, while managing unstructured data has created dependence on individual employees and increased the risk of errors. As AI transformation advances, building a platform for accurate document preparation while preventing data fragmentation has become an urgent priority.
The new automation platform uses a database in a Snowflake environment that consolidates investment trust information and centrally manages data extracted from Word files. Its 4 functions are natural-language search, which uses AI to understand context and search for funds; an editor that automatically generates draft proposals based on change instructions; batch changes, in which AI creates replacement rules for changes common to multiple funds and processes the relevant files; and automatic generation of before-and-after comparison tables, which outputs Word documents for submission to the industry based on change differentials.
For development, GitHub Actions and Snowflake CLI are being used for CI/CD to reduce operating costs. The stated policy is to limit AI's role to drafting text and generating rules, while carrying out processes requiring certainty, such as file extraction, through system processing. This approach aims to improve the ability to process large-scale revisions in a short time while maintaining both accuracy and reliability.