Michinori Kanokogi, Nissay Asset Management - ChatGPT and GenAI: What They Mean for Investment Professionals (S3E50)

Welcome to the 150th edition of the eXponential Finance Podcast, and the last episode of the third season. Whether you listen to us for the first time, or are a regular, we appreciate your spending time with us.

This episode is available on Apple Podcasts, YouTube, Amazon Music, and many other major platforms via our Spotify Podcaster Page.

In this podcast, Michinori Kanokogi, Head of Solution Research at Nissay Asset Management, presents a comprehensive overview of how Generative AI (GenAI) is fundamentally reshaping the asset management industry. He argues that GenAI is not merely an incremental technological improvement but a revolutionary force, a "General Purpose Technology" akin to the steam engine or the internet, that necessitates a complete rethinking of workflows, employee roles, and corporate strategy. Kanokogi details Nissay's proactive, multi-layered approach to implementation, addresses the critical risks and opportunities, and offers a compelling vision for the future role of the human investment analyst in an AI-augmented world.

Key Takeaways

1. GenAI as a Revolutionary "General Purpose Technology": Kanokogi asserts that today’s GenAI is a major break from past AI. Its defining characteristics are usability (anyone who can type can use it, no coding required) and versatility (a single foundation model can perform countless tasks from translation to summarization to programming). This accessibility and broad applicability elevate it to the status of a "General Purpose Technology," a rare innovation with the power to transform entire economies and industries.
2. Successful Adoption Requires a Holistic, Company-Wide Strategy: Effective implementation is not just an IT project. Kanokogi outlines a dual framework: a three-layered technical infrastructure (general chat, custom data integration, specialized apps) and a three-part organizational model of "Leadership, Lab, and Crowd." This requires top-down vision and commitment from leadership, a specialized "Lab" to build and manage the technology, and bottom-up experimentation from the "Crowd"—all employees—to discover and integrate value in their daily work.
3. The Risk of Not Using GenAI Now Outweighs the Risk of Using It: While acknowledging valid concerns around data leakage, copyright infringement, and "hallucinated" inaccuracies, Kanokogi argues these are manageable through robust infrastructure, clear employee guidelines, and a culture of critical verification. He emphasizes a more significant danger: the risk of inaction. Citing Japan's Financial Services Agency, he warns that firms failing to adopt GenAI will face competitive disadvantages, challenges in talent acquisition and retention, and the uncontrolled use of insecure public AI tools by their employees.
4. The Human Analyst's Role Will Shift from Data Processor to Relationship Builder: The emergence of powerful "deep research" AI capabilities will automate the collection and synthesis of public information, tasks that currently consume much of an analyst's time. Kanokogi predicts the value of human analysts will pivot dramatically to the "softer side" of the investment process. Their new focus will be on uniquely human skills: conducting interviews with corporate management to gain private insights, building trust, and exercising nuanced judgment by sensing the "vibe" and "atmosphere" in meetings—abilities far beyond the reach of AI.
5. GenAI Enhances Analysis, Not Final Decision-Making: Kanokogi makes a crucial distinction about where GenAI currently adds the most value. He believes it is a powerful tool for improving the analysis phase of the investment process—information collection, summarization, and idea generation. However, he is deeply skeptical of its current ability to handle portfolio management or make final investment decisions. He cautions that AI-generated recommendations are not grounded in robust financial theory and that relying on them for fiduciary decisions is unwise until the technology matures significantly.

Full Podcast Summary

Introduction: A Paradigm Shift in Artificial Intelligence

Michinori Kanokogi begins by framing the current wave of Generative AI as a fundamental departure from previous iterations of artificial intelligence. While AI has existed for decades, it was traditionally the domain of specialists who needed coding skills to build and deploy purpose-built models for narrow tasks. The advent of Large Language Models (LLMs) has changed everything.

Kanokogi identifies two key differentiators: usability and versatility. "Anyone who can type can now benefit from cutting-edge large language models," he states. This democratization of access is coupled with unprecedented versatility. A single foundation model, like GPT-4 or Claude 3, can handle a vast array of tasks—translation, proofreading, summarization, and even complex programming—that previously would have required multiple, distinct AI systems.

This combination leads him to classify GenAI as a "General Purpose Technology" (GPT), a term economists reserve for a handful of transformative innovations in human history, such as the domestication of plants, the steam engine, and the internet. Like its predecessors, GenAI has the potential to affect an entire economy, and its adoption will be vast and fast. This sets the stage for his core argument: for asset managers, embracing this technology is no longer optional.

Implementing GenAI: Nissay's Three-Layered Approach

To harness this power, Kanokogi explains that a structured, multi-faceted approach is essential. At Nissay Asset Management, they have developed a three-layered infrastructure to integrate GenAI throughout the company.

• Layer 1: The General Chat Interface. This is the foundational layer, providing all employees with secure access to leading LLMs like OpenAI's GPT, Anthropic's Claude, and Google's Gemini. It serves as a general-purpose tool for ad-hoc queries, brainstorming, and daily productivity tasks, replacing the need for employees to use public, potentially insecure versions.
• Layer 2: Custom Data Integration. The second layer allows users to upload the firm's internal documents (e.g., Word files, PDFs, research reports) to create a private, secure knowledge base. The AI can then answer questions and generate text based on this proprietary information, a process often facilitated by Retrieval-Augmented Generation (RAG). This ensures that the AI’s responses are grounded in the company's own data.
• Layer 3: Specialized, Custom-Built Applications. This is the most advanced layer, where Nissay's "Lab" develops bespoke applications for specific, high-value business tasks. Kanokogi provides a concrete example: an "ESG Interview Assist" tool. This application analyzes corporate disclosures like annual and sustainability reports, automatically extracting ESG-related information based on Nissay’s proprietary evaluation criteria. It then generates summaries and preliminary evaluations, dramatically accelerating the research process for analysts.

Beyond the technology, successful adoption hinges on an organizational strategy Kanokogi calls the "Leadership, Lab, and Crowd" model.

• Leadership is responsible for setting a clear vision, establishing ethical guardrails, and championing the technology from the top. Kanokogi highlights Nissay’s own initiative of conducting mandatory, hands-on training sessions for all 20 of its top executives to ensure they understand the technology's potential and limitations firsthand.
• The Lab refers to his specialized team, which acts as the engine for digital transformation. They are responsible for building the infrastructure, prototyping new applications, and providing technical expertise.
• The Crowd encompasses all employees. Kanokogi stresses that because GenAI is a tool for everyone, innovation must be a bottom-up process. All employees are encouraged to experiment with AI in their daily roles to find new efficiencies and use cases. To facilitate this, Nissay has appointed 55 "GenAI Navigators"—evangelists embedded within various departments to share knowledge and identify opportunities.

This dual strategy of robust infrastructure and company-wide engagement has been highly effective, with Nissay achieving over 80% monthly active usage of their GenAI platform.

Navigating the Risks: The Imperative to Act

Kanokogi squarely addresses the risks associated with GenAI, categorizing them into three main areas: information leakage, inaccurate or inappropriate content (hallucinations), and copyright violations. However, he argues that each of these can be effectively mitigated.

• Information Leakage: This is managed by using secure, enterprise-grade APIs and proprietary infrastructure, ensuring that sensitive company data is never used to train public models.
• Hallucinations & Inaccuracy: The primary defense is human oversight. Nissay’s internal guidelines mandate that employees are ultimately responsible for any AI-generated output. They must critically review, fact-check, and verify the information as if they had created it themselves.
• Copyright: Guidelines strictly prohibit employees from inputting existing copyrighted works or using prompts designed to mimic a specific artist's or author's style.

With these mitigations in place, Kanokogi makes a powerful counterargument: "The risks of not using GenAI are now much bigger than the risks of using it." He cites a discussion paper from Japan’s Financial Services Agency (FSA) that explicitly encourages financial institutions to embrace the technology. The risks of inaction include falling behind competitors, failing to attract and retain talent who expect access to modern tools, and the proliferation of "shadow IT," where employees resort to using unsecured public AI tools on their personal devices, creating a far greater security threat.

The Future of the Investment Analyst

The most transformative impact of GenAI, according to Kanokogi, will be on the role of the investment analyst. He points to a recent technological leap he calls "test-time scaling" or "reasoning models"—newer LLMs like OpenAI’s O-3 and Claude 3.1 that "think carefully before answering" by dedicating more computational power to complex queries. This has led to a dramatic improvement in reasoning and planning capabilities.

This advancement enables what he calls "Deep Research." An AI agent can now be tasked with a research question, and it will autonomously browse the web for 10-15 minutes, gather information from numerous sources, synthesize the findings, and produce a comprehensive, well-structured report. This capability automates a core function of the traditional analyst: the collection and analysis of public information.

"What's left for the human analyst?" he asks. His answer is a shift to the "softer side" of the investment process, focusing on tasks that machines cannot perform.

1. Accessing Private Information: The human analyst's primary value will be in obtaining information that isn't publicly available online. This means building relationships and conducting insightful interviews with corporate management, suppliers, and customers.
2. Building Trust and Communication: Skills in communication, empathy, and building trust will become paramount in order to elicit candid information during these interactions.
3. Exercising Nuanced Judgment: Analysts will need to use their "five senses"—observing body language, sensing the "atmosphere" in a room, and catching subtle cues that an AI cannot process.

This shift presents a new challenge: how to train junior analysts. The traditional apprenticeship model, where juniors learn by performing research tasks for seniors, breaks down when a senior analyst can get the same information faster from an AI. Kanokogi suggests that AI itself may need to be leveraged as a coaching tool to help develop the next generation of analysts.

Ultimately, Kanokogi sees a clear division of labor. For information analysis, "LLMs can improve a lot." But for the final stages of the investment process, human judgment remains irreplaceable. "I don't see much improvement using LLMs for portfolio management or actually making decisions," he concludes, explaining that AI outputs are not yet grounded in sound financial theory and cannot be trusted with fiduciary responsibility. The future is one of human-AI collaboration, where machines handle the data, freeing humans to focus on what they do best: building relationships, exercising wisdom, and making the final call.