Artificial Intelligence
Franklin Templeton Develops AI Research Workflows With OpenAI

Key PointsFranklin Templeton's first AI setup failed because models lacked a semantic layer…
View more
- Franklin Templeton's first AI setup failed because models lacked a semantic layer explaining how data connects.
- Codex hackathons pair technical experts with investment practitioners so practitioners build their own research tools.
- The firm wants AI to surface diverse research views and flag risks rather than narrow perspectives.
Franklin Templeton is developing artificial-intelligence workflows for equity research using OpenAI’s Codex and other tools, OpenAI said in a customer story published November 5, 2026. The initial objective is to help portfolio managers compare research perspectives and examine the risks behind investment ideas. The disclosure describes an effort still moving from experimentation toward production, not a completed firmwide rollout.
The asset manager’s first AI approach placed data in a database and connected models to it, but the setup did not deliver the answers its teams needed because the models did not understand how the data fit together. The teams subsequently added agents and a semantic layer, defined as a structured model of what the firm’s data means and how it is connected. Codex and other tools are being used to develop that connected intelligence layer.
The planned system is intended to bring together inputs from research, risk, compliance and trading, which provide different perspectives on an investment. Its foundation combines trusted connectors, portfolio and workflow context, and agentic capabilities to surface actionable insights for analysts and portfolio managers. Those elements are meant to make the system reusable and governed while giving investment decisions more context than the original database connection could supply.
Reworking a Sequential Research Process
Katrina Dudley, Franklin Templeton’s senior vice president for public market investments, described the earlier equity-research process as human-led and sequential. Analysts perform first-pass stock analysis, review earnings calls and study competitors by business line as they prepare to pitch new ideas to portfolio managers or provide updates on existing portfolio positions.
“We have moved from augmenting our research process to reimagining the investment research workflow,” Dudley said. The longer-term goal is to make connected intelligence available across investment teams, from analysts to the leaders of each investment group, while retaining human judgment alongside the system’s insights.
Implementation has included Codex hackathons that place technical experts alongside investment practitioners as the practitioners develop their own tools. Investment AI champions are working to turn those experiments into production-ready workflows. Security review, expert support, data controls and data agreements form part of that path. Franklin Templeton has also brought practical labs to teams soon after new technology launches, and Dudley said leadership encourages experimentation within clearly defined risk guardrails while accepting that not every initiative will succeed.
Research Diversity and Risk Shape the Design
The firm’s approach addresses a limitation of preference-learning tools: a system that learns a user’s preference for one research provider can surface more of that provider’s reports, potentially narrowing the perspectives entering the investment process. Franklin Templeton instead wants investment teams to see a diversity of research and opinion.
It also wants AI to emphasize risk alongside the investment thesis. “We need AI to understand that a great investment idea is one that is well reasoned, but it also has analyzed the risks and makes our portfolio managers aware of the risk,” Dudley said. That requirement ties the technology’s output directly to the information portfolio managers need when evaluating an idea.
Dudley also wants analysts to spend more time in the field speaking with experts and professionals, including executives at private companies, and to bring those perspectives back to their investment teams. The work therefore extends beyond adding a single tool: it combines software development, data architecture and investment controls. Its central test is whether role-specific experiments can become governed production workflows while preserving research diversity, explicit risk analysis and the human perspective Dudley considers essential.












