Capital markets firms are spending millions teaching employees how to “prompt” AI tools, and it’s a mistake.
After a year of major generative AI rollouts, many firms are now scrambling to justify their investments. The results haven’t matched the hype: productivity hasn’t surged, analysts aren’t magically faster, and the ROI dashboards look thin. In response, companies are reaching for an easy fix: prompt training. If employees could just learn to “ask better questions,” maybe they’d finally unlock the value in these powerful models.

They won’t.

The Real Problem Isn’t Prompts

Prompt training is a classic case of solving the wrong problem. Employees aren’t failing to use AI because the tools are too complex, they’re failing because the organization doesn’t understand how AI is actually being used. Most firms don’t have visibility into what their people are prompting, what tasks AI is helping with, or where it’s quietly adding (or wasting) time. Instead of building that insight, companies default to the old learning-and-development reflex: more training.

But AI chat is already intuitive. Every firm has “power users”: analysts, bankers, developers, who have organically figured out how to make AI work for them. They didn’t need a course to learn to prompt. They learned by doing, and by adapting AI to their workflow. These are the people actually moving the productivity needle. The question leaders should be asking isn’t “How do we make everyone better at prompting?” but “What are our best people already doing with AI, and how can we scale that?”

The Complexity Is in the Workflow

The hard part of AI adoption isn’t stringing words together in a chat box.  The hard part is in defining repeatable workflows. The difference between an analyst who saves two hours a day and one who doesn’t isn’t phrasing; it’s process. One has figured out where AI fits in their daily tasks: summarizing market commentary, generating slides, checking regulatory language, and the other hasn’t. Training people on prompt syntax won’t bridge that gap. Building systems that capture and share how AI is used across the firm will.

Invest in Capturing Institutional Intelligence

The real opportunity is to treat every effective prompt and workflow as institutional knowledge. Power users and subject-matter experts are constantly encoding valuable context into their AI interactions, firm-specific language, analytical reasoning, regulatory nuance. Right now, all of that disappears the moment they close a chat window. Instead of running another training series, firms should be building infrastructure that captures, organizes, and redistributes that intelligence.

Imagine a system that learns from the prompts and outputs of your best people: a living knowledge base of proven workflows and contextualized prompts that others can reuse or adapt. Combine that with data integrations so the AI has access to your research, client materials, and internal models, and suddenly you’re not teaching people to prompt — you’re giving them intelligent tools that already know how your business thinks.

The Bottom Line

Generative AI doesn’t need more human training; it needs better organizational memory. Capital markets firms don’t need “better prompters”, they need systems that observe, learn, and scale what their best people are already doing. The ROI won’t come from another workshop. It will come from capturing durable intelligence, embedding it in your tools, and making that knowledge replayable across the firm.

Stop teaching employees to talk to AI. Start building systems that let AI, and your organization, learn from them.

‍