Franklin Templeton 用 AI 研究工作流与 Codex 突出投资风险

Franklin Templeton is developing AI workflows to help portfolio managers compare investment research and examine risks, beginning with equity research, OpenAI said in a case study published November 5, 2026. Codex is one of several tools its investment teams are using.

The work addresses a handoff in the existing research process. Analysts perform first-pass stock analysis, review earnings calls and study competitors in each business line before pitching new ideas or updating portfolio managers on existing positions. Research, risk, compliance and trading each offer a different perspective; Franklin Templeton wants to combine those inputs so decision-makers can assess both the reasoning behind an idea and its risks.

Katrina Dudley, senior vice president for public market investments at Franklin Templeton, described the change as a move beyond automating a human-led sequence. “We have moved from augmenting our research process to reimagining the investment research workflow,” she said.

The first attempt placed data in a database and connected models to it. It did not produce the answers the teams needed because the models did not understand how the data fit together. The approach changed with the addition of agents and a semantic layer, defined as a structured model of what the firm’s data means and how it is connected.

That distinction goes to the core of the research problem: access to information alone did not supply the relationships needed to interpret it. The team is using Codex and other tools to develop a connected intelligence layer that brings data into the tools and surfaces insights for users.

The envisioned result is a governed, reusable foundation combining trusted connectors, portfolio and workflow context, and agentic capabilities that surface actionable insights for analysts and portfolio managers.

Development also involves the people who will use the systems. Codex has been used in hackathons that pair experts with investment practitioners as they build their own tools, while practical labs introduced soon after new technology launches help teams test how the technology fits their work. Investment AI champions are working to turn experiments into production-ready workflows.

Some behavior that makes AI helpful in other settings creates a problem for asset management. Many tools learn a user’s preferences and show more of the same material; for an investor who favors one research provider, that can mean more reports from that provider. Franklin Templeton sees the resulting narrowing as a potential source of bias and instead wants analysts and portfolio managers to access a diversity of research and opinion.

Risk receives similar treatment. The systems are intended to emphasize investment risks alongside the reasoning behind an idea so portfolio managers are made aware of them. Identifying those risks in a research note, Dudley said, “makes for a more informed investment decision.”

Together with the semantic-layer work, those requirements make research exposure part of the system’s design: which perspectives reach the user matters alongside whether information can be retrieved. The intended workflow therefore connects technical architecture with an investment-specific requirement to preserve differing views and make downside analysis visible.

Guardrails, Adoption and Human Fieldwork

Leadership encourages innovation and experimentation with AI within clearly defined risk guardrails, accepting that not every initiative will succeed. Dudley said each should nevertheless generate insights that strengthen capabilities, inform future decisions and drive continuous improvement. The controls around that ambition include security review, expert support and data agreements on the path from experimentation to production-grade tools.

Dudley’s longer-term goal is to make connected intelligence accessible across investment teams, from analysts to the leaders of each investment group. She also wants analysts to spend more time in the field speaking with experts, professionals and executives at private companies, then bring what they learn back to their teams. That human perspective remains essential as the tools develop.

The project’s significance lies in that combination of automation and restraint. Franklin Templeton is trying to surface insights and connect more of the context around an investment while keeping differing opinions, explicit risk analysis, security controls and human judgment inside the process. The central question is whether those elements can work together in a governed, reusable system without narrowing the research exposure the system is meant to expand.