Which Company's Case Study?
This case study focuses on Morgan Stanley, one of the world's leading financial institutions. In its wealth management business (asset management and operation services for high-net-worth individuals), the company partnered with OpenAI to build an advanced internal system leveraging generative AI to support the work of financial advisors.
Challenges They Wanted to Solve
Morgan Stanley's financial advisors needed to access a vast amount of proprietary internal content (intellectual property) such as market research reports, investment strategies, and economic analyses to provide optimal recommendations to clients. Rapidly and accurately finding client-specific information from hundreds of thousands of these documents was a significant time burden, reducing the time advisors could dedicate to client interaction and relationship building, which are their primary focus.
How AI Was Used
Morgan Stanley developed an internal chatbot called 'AI @ Morgan Stanley Assistant,' utilizing OpenAI's GPT-4 model. This system allows advisors to pose questions in natural language, searches for relevant information from the company's vast knowledge base, summarizes the content, and generates answers. This is a typical application of a technique called RAG (Retrieval-Augmented Generation), which combines the capabilities of large language models with reliable, proprietary company information to achieve highly accurate responses.
Estimated Architecture
The following is an estimated architecture based on publicly available information and common configurations. It does not definitively describe the actual internal composition.
- ▸Input Data: Natural language questions entered by financial advisors. The search target includes over 100,000 internal documents, such as PDFs and internal websites.
- ▸AI Processing/Model or Search Layer: It is believed that the GPT-4 model is being utilized in a secure environment through the partnership with OpenAI. For the fast retrieval of internal documents related to advisor questions, techniques such as vector search are employed. The search results are then passed as context to GPT-4, and a RAG (Retrieval-Augmented Generation) architecture is adopted to generate the final answer.
- ▸Output to Business System: The advisor's chat interface displays a summarized answer along with links to the internal documents that served as the basis. This allows advisors to quickly verify the validity of the answer.
- ▸Monitoring and Governance: To meet the stringent regulatory and compliance requirements of the financial industry, a framework is established to continuously evaluate the quality of the model's responses. By incorporating a 'Human-in-the-loop' process, where advisors ultimately review and edit the content before presenting it to clients, risks are managed.
Implementation Effects and Key Takeaways
- ▸Increased Advisor Productivity: The time previously spent on information retrieval has been significantly reduced, allowing advisors to dedicate more time to client relationship building and providing deeper insights.
- ▸Deepened Knowledge Utilization: The access rate to internal documents, which was only partially accessed before implementation, dramatically increased from 20% to 80%. The entire organization's knowledge is now instantly accessible to anyone.
- ▸High Adoption Rate: After implementation, over 98% of the advisor teams in the wealth management division use this AI assistant daily, indicating its high valuation in the field.
- ▸Key Takeaway for Readers: The success of this case is not merely due to the introduction of a powerful AI model, but rather the meticulous development, through careful evaluation, of a system that ensures the security and compliance specific to the financial industry while being trustworthy and usable by advisors on the front lines.
What Japanese Companies Can Learn
This case study is relevant for many Japanese companies, regardless of industry. Particularly for companies with a large accumulation of specialized knowledge such as manuals, R&D documents, and past proposals, a similar AI assistant could create significant value. It can enhance employee productivity in various situations, such as handling customer inquiries, assisting sales representatives in proposal creation, and technical problem-solving, transforming often individualized knowledge into organizational strength.