A number that usually gets credited to the tool
Morgan Stanley's wealth management arm built an AI assistant for its financial advisors. By the account OpenAI has published of the project, more than 98% of advisor teams actively use it. It began able to answer about 7,000 questions. A senior member of the firm's AI team now says it can "effectively answer any question from a corpus of 100,000 documents."
The detail that travels
The size is not what other organizations can copy. How the firm decided the assistant was good enough is. Advisors and prompt engineers graded the AI's responses for accuracy and coherence, and the team refined the assistant from what they found. The people who would put a client's money behind the answers were the people who marked them.
A different order
Most rollouts run the other way. A committee selects a tool, the company announces it, and whoever complains becomes the quality check after the fact. Here the advisors did the checking while the assistant was being refined, and adoption is now close to universal.
The checking also teaches. People who mark the answers learn where the assistant is weak before they depend on it. A committee approving a tool from a demo gets no equivalent signal.
A note on the source
The account comes from OpenAI, the vendor, describing a customer. It carries no publication date, and the adoption figure is reported as OpenAI states it, not independently measured. The pattern is still worth taking seriously, because it does not depend on the figure.
Before your next rollout
Ask who marks the first hundred answers. If it is the people who will rely on them, you are doing the part that mattered here. If it is a committee or the vendor, expect adoption to wait on a trust nobody measured.
Build the grading step into the rollout before it launches: Design the rollout around who marks the answers →
Keep reading: The reviewer folds when you press hardest. That is the moment you need it most. · You are treating two different failures as one
This is the thinking behind a Trusted AI Culture.
Sources: OpenAI, "Morgan Stanley uses AI evals to shape the future of financial services" (customer case study). Vendor-authored and undated.
Written with AI in the loop: my idea, AI drafted and sharpened, my judgment on the way out. Every word is mine to stand behind. — Darren