Morgan Stanley is shaping the future of financial services

Morgan Stanley has spent the past couple of years building AI tools for its financial advisors, most notably an assistant trained on the firm's own research and internal knowledge base. The harder problem was never getting a model to sound confident. It was proving the thing wouldn't quietly get something wrong in front of a client.

That's where evals come in. Rather than shipping a model and hoping, Morgan Stanley built a structured evaluation process to test outputs against real advisor questions, checking for accuracy, relevance, and whether the answer actually reflects the firm's compliance standards. Humans with deep subject knowledge review the results, flag gaps, and feed that back into how the tool gets tuned. It's less glamorous than a product launch, more like quality control on a factory floor, except the product is financial guidance touching billions of dollars in client assets.

The firm treats this as an ongoing discipline, not a one-time checkbox before launch. As the assistant expands into new tasks, summarizing earnings calls, drafting research recaps, prepping advisors for meetings, each new capability gets its own round of scrutiny. Wealth management is a business built on trust accumulated over decades, and a single bad answer repeated across thousands of client conversations could do real damage fast.

There's also a talent angle here. Morgan Stanley has framed evals as a way to bottle the judgment of its most experienced advisors and researchers, encoding what good looks like so newer employees get faster feedback loops. That's a quieter, more interesting use of AI than the usual chatbot pitch: not replacing expertise, but compressing the time it takes to develop it.

What stands out is how unsexy the actual work is. No one is talking about AGI or agents running trading desks. It's spreadsheets of test cases, disagreement logs between reviewers, and slow iteration. In an industry where regulators and compliance officers outnumber enthusiasm, that boring rigor might be exactly what determines whether generative AI sticks around in finance or gets quietly shelved after the first embarrassing headline.