Where are all the Portfolio Managers’ Large Language Models?

Imagine it’s 1995 and you come across a portfolio manager who refused to use Excel. Most investment leaders would have considered that a career-limiting decision. Yet today, many investment professionals are approaching large language models (LLMs) with similar skepticism and hesitation.

As an executive coach to investment leaders across the industry, I continue to be surprised by how little AI has penetrated the daily workflow of portfolio managers and analysts. Given the headlines, one might assume AI is already improving every investment process. Walk the halls of many investment firms, however, and a different reality emerges.

There are, of course, investment management companies that have very thoughtfully and usefully deployed AI throughout their investment process, but they remain the exceptions. They also tend to be largest and best-resourced managers. Many are not pure-play asset managers. Quant teams tend to be ahead of the curve, since they are natively digital.

There are two main reasons:

1. Investment performance – LLMs can help your team make better investment decisions. Even a small increase in hit rate, say from 60% to 62% (assuming symmetrical win and loss performance, math by request) increases expected alpha by 20%.
2. Productivity and costs – If introducing LLMs increases productivity by just 2% per year, over 10 years this is a nearly 22%productivity improvement, allowing one in five team members to be re-deployed on new efforts. Imagine walking into the office tomorrow and having 22% more capacity for value-added projects.

We would argue that the greatest barriers to AI adoption are not technological. They are the same human traits that we’ve seen before. Let’s take a deeper dive into a few of the most significant:

1.  The “AI champion” may be the wrong person

Many firms put AI adoption in the hands of technology, data science, operations, or innovation teams. These are professional and well-meaning teams but lack the skills necessary to train PMs on how to use AI effectively.

The message is too often, “Here is a tool. Best of luck in using it in your process.” A more effective approach sounds like, “Here is how this improves the quality of our investment judgment.”

2.  Time scarcity makes experimentation feel expensive

PMs and analysts are overloaded. Earnings seasons, portfolio reviews, client meetings, market events, risk reviews, and internal meetings consume the calendar. LLM adoption is an investment that requires a temporary productivity dip: learning prompts, redesigning workflows, testing use cases, comparing outputs, and building team norms.

The long-term payoff may be substantial, but the short-term cost is real. Busy professionals often reject new tools not because they doubt their long-term efficacy, but because they cannot tolerate the short-term transition cost.

3.  Lack of psychological safety, including around learning in public

LLM adoption requires people to be beginners and show vulnerability. Senior PMs especially dislike feeling this way. They may avoid workshops, demos, or peer learning because they don’t want to tarnish their image of being the wise leader.

This is an EQ issue, not a technical one. Firms that adopt faster will already have psychologically safe environments, encourage experimentation, expect and celebrate failure, and be comfortable with peer learning.

4.  The rewards system doesn’t recognize process improvement

PMs are usually paid for performance, asset retention, client confidence, and commercial outcomes. Analysts are rewarded for good ideas, differentiated work, responsiveness, and credibility with PMs.

Few firms explicitly reward better research workflows, higher-quality debate, more systematic decision reviews, better information sharing, or improved knowledge capture. So LLM adoption becomes an unpaid extra effort. The firm may say AI matters, but the incentive system says: “Keep doing what already gets you paid.”

5.  “Responsible use of AI” often feels risky rather than enabling

Fear dominates many AI conversations: privacy, hallucination, cyber risk, compliance, regulation, model risk. These are all valid concerns in our heavily regulated industry. If the organizational narrative is primarily defensive, however, PMs will conclude that AI use is dangerous, not encouraged.

This directly maps to the behavioral finance concept of the penalties for sins of commission being higher than sins of omission. This occurs when an organization creates an environment where an actual mistake is penalized but a missed opportunity is not. Good investment teams understand this and take steps to ensure it doesn’t occur. One way of doing so is when PMs ensure they are taking sufficient risk in their portfolios to deliver the promised alpha. This same concept should apply to AI – make sure you are taking some risk and recognize employees who try new ways of using the models, even if they fail.

6.  Probably the greatest hurdle of all – process engineering

Process engineering is the discipline of deliberately designing, analyzing, improving, and controlling an investment process so that a desired outcome is produced more reliably, efficiently, and consistently. I would argue that many investment firms do not have an AI problem. They have a process-engineering problem that AI is revealing. Many investment managers don’t naturally think in terms of process-engineering, making it harder for them to see how AI can truly advance their investment process.

Today, when PMs are asked to add LLMs to their process, many look to outcomes:

• Can the LLM find me good stocks?
• Can it tell me what to buy?
• Can it predict earnings?
• Can it generate alpha?

This approach ignores the bigger questions around process. The greatest strides are made when the investment process is examined holistically while considering AI’s role at each step. Ask yourself questions that center on process engineering:

• Where in our investment process is judgment currently bottlenecked?
• Where do we repeatedly lose time?
• Where do we suffer from inconsistent quality?
• Where do we fail to challenge our assumptions?
• Where does information fail to become insight?
• Where do good ideas die because they are poorly framed?
• Where do we rely too heavily on memory, personality, or informal norms?
• Where does our process depend on one person’s tacit knowledge?
• Where are we making preventable decision errors?

1. Stop treating LLM adoption as a technology rollout.
Technology teams are essential, but they cannot own the behavioral transformation. The CIO, director of research, and senior PMs own the investment process, and they need to own its AI transformation.

2. Map your investment process.
Identify every major decision point from idea generation to exit. Then ask where the process could be quicker, needs more consistency, more documentation, or better learning.

3. Pick a small number of high-value use cases.
Avoid generic “use AI more” mandates. Start with concrete practices such as pre-mortems, thesis reviews, investment memo critiques, management-question generation, post-mortem analysis, and decision-journal reviews.

4. Create safe practice environments.
Have teams bring real but non-sensitive examples into working sessions. Let senior people struggle publicly. Normalize the idea that everyone is learning.

5. Make it clear that the LLM is not making the decisions.
The LLM is a tool, the PM owns the decision. The tool’s job is to sharpen thinking, expose blind spots, and improve process quality.

5. Reward the behaviors you want.
You get what you measure. If analysts and PMs are only rewarded for ideas and performance, they will not spend time improving workflows. Recognize and reward people who create reusable prompts, better debate structures, better decision journals, and better knowledge-sharing routines.

7. Use outside facilitation when the real barrier is cultural.
If the barrier is lack of trust, weak debate, unclear incentives, senior resistance, or poor process discipline, then the firm does not primarily need more technology training. It needs leadership, culture, and process work.

At Focus Consulting Group, we believe that investment performance is shaped not only by hard skills, but by culture, leadership, incentives, and the quality of team decision-making. LLMs do not change that belief; they make it more important. The firms that benefit most from AI will not simply be the firms with the best tools. They will be the firms with enough psychological safety to experiment, enough process discipline to identify where AI can help, enough leadership humility to learn in public, and enough strategic alignment to make AI part of how investment judgment actually gets better.