Franklin Templeton directs its investment AI towards risk

The asset manager is revamping its research tools around a "connected intelligence" built with OpenAI. The focus is on diversity of viewpoints, highlighting risks, and a structured transition from experimentation to production.

The company prioritizes diversity of sources and risk alerts

According to Franklin Templeton, certain AI capabilities that work in other sectors are not suitable for asset management. Many tools adapt their functioning to user preferences and tend to reinforce these trends, which, in the realm of financial research, risks generating bias. The company, on the contrary, aims to ensure access to a diversity of research and viewpoints. It also anticipates that its AI systems will emphasize the identification and communication of investment risks. For Katrina Dudley, a solid investment idea is one that is well-argued and whose risks have been analyzed and communicated to managers; identifying risks in a research note is seen as an asset for more informed decision-making. This positioning is accompanied by safeguards: management encourages innovation and experimentation with AI within clearly defined risk limits, while accepting that not all initiatives will succeed, provided they yield useful insights. The transition to production-quality tools includes security reviews, expert support, and data agreements.

An architecture with agents and a semantic layer replaces the initial approach

Initially, the team connected models to an internal database, but did not achieve the expected responses, as the models did not understand the relationships between the data. Results improved after the addition of agents and a semantic layer describing the meaning and links of the company's data. The team relies on Codex and other tools to develop this connected intelligence. Katrina Dudley states that such a layer has been built and that integrating data into these tools allows for the emergence of insights deemed critical.

From hackathons to production workflows, with integrated controls

Franklin Templeton utilized Codex during hackathons that brought together specialists and investment professionals to create their own tools. Specialists in AI applied to investment convert these experimental tools into operational processes ready for deployment, a process that includes security checks, expert support, and data validations. Katrina Dudley presents this movement as a shift from merely augmenting the research process to reimagining the workflow. After each launch of new technologies, practical labs were quickly established to allow teams to test their relevance to their tasks, with the goal of helping them grasp the power of these tools.

Manager-oriented uses and dissemination to all teams

Research, risk, compliance, and trading provide complementary perspectives that Franklin Templeton seeks to bring together for the benefit of portfolio managers. Katrina Dudley asserts that teams are able to aggregate these contributions to produce insights for decision-makers. The ambition is to help managers examine different research perspectives and the associated risks of each idea, relying on Codex among other tools. For Dudley, the strength of OpenAI's tools lies in the fusion of continuous intelligence and human judgment: AI was previously confined to automating processes, but now serves to generate analyses. She notes that teams feel they are at the forefront with OpenAI. In the long term, she aims to make this connected intelligence accessible to all investment teams, from analysts to executives, while encouraging analysts to spend more time in the field with experts, including in private companies, and to share their insights. She emphasizes the essential nature of the human perspective as tools evolve.