Asset Management in the Generative AI Era Faces a Mutual Fund Data Lag

2026-08-15

Slow Structuring of Mutual Fund Data

Tetsu Noguchi, founder of Semantic Base, a Tokyo-based company in Chuo that provides AI adoption support for financial institutions, said he is concerned that mutual funds alone could be left out of generative AI-powered asset management services. The backdrop is that most of the more than 5,800 mutual fund data sets have not been converted into XBRL.

Asset managers disclose mutual fund information in the Financial Services Agency's Electronic Disclosure for Investors' NETwork, or EDINET, but the disclosures are limited to items such as fund names, balance sheets and income statements. Key fields such as trust fee rates, portfolio holdings and changes in net assets are not available as comparable structured data. Even when generative AI reads Japanese mutual fund data, it cannot handle it properly and can also cause hallucinations, in which it returns plausible-sounding information that differs from the facts. There have also been cases where it confused the 'dividends' paid by portfolio companies with the 'distributions' paid to customers by mutual funds.

Preparing for an AI-First Model

Behind this delay lies the industry's long-standing reluctance to have mutual fund data directly compared and ranked. A former president of an asset manager under a major bank said, 'Data preparation costs money. If we could not show benefits that outweighed the cost, we could not get agreement even after making a proposal.'

The environment is beginning to change, however, as generative AI spreads. In May, MUFG Bank announced it would offer 'Apps in ChatGPT' using Moneytree's financial data infrastructure, which powers a household account app. Users can ask about connected accounts, balances and spending this month, and then understand their financial behavior through graphs and tables.

A total of 28 companies, including the three major banks, have also started initiatives to support individual asset management with AI. The idea is to let users complete everything from searching for mutual funds to comparing them and signing purchase contracts through dialogue with AI. But if only some mutual fund data are converted into XBRL, it will be difficult to build an optimal portfolio for users. How far XBRL conversion should go will be a key focus going forward.

In the United States, holdings details as well as mutual fund risk and return data are already being converted into XBRL. At the same time, discussion over whether AI-based investment recommendations could fall under investment advice or discretionary investment management is a shared issue in Japan and the US. Semantic Base is working with major online brokerages and asset managers to structure unorganized mutual fund data and integrate it into financial institutions' large language models, or LLMs. Noguchi said he aims to create a 'digital version of EDINET for mutual funds.'

According to Bank of Japan flow of funds statistics, Japanese household financial assets totaled 2,386 trillion yen at the end of March. Of that, stocks and investment trusts amounted to 563 trillion yen, up by nearly 30% in the past year and closing in on the 584 trillion yen held in insurance and pensions. As investing becomes more rooted in household finances, improving data that AI can easily read is essential to advancing personalized investment optimization using AI.

Enjoyed this article? Share it with your network!