.png)
Morgan Stanley
Finance · US
Document access: 20% to 80%
98% of Morgan Stanley advisor teams use AI daily to search and apply internal knowledge
Morgan Stanley's AI assistant is not a chatbot. It is a controlled GenAI deployment on 100,000 internal documents, with daily regression testing and an evaluation framework built before any use case went live.
Friction
Financial advisors spent significant time searching for, summarising, and translating internal research, market insights, and product documentation into usable form for client conversations. Speed and accuracy are critical in wealth management.
Breakthrough
A bespoke GPT-4 solution with retrieval over Morgan Stanley's internal content, where answers include direct links to source documents. A second tool, Debrief, transcribes and summarises client meetings with consent, generating CRM notes and draft follow-ups. A pre-deployment evaluation framework tested each use case for quality, reliability, accuracy, and coherence before rollout.
Impact
98%+ of advisor teams actively use the assistant. Document access rose from 20% to 80%. Knowledge available in seconds. Follow-ups that previously took days can now be completed in hours. Debrief generates CRM notes and action items directly from Zoom recordings.
Unlock the full analysis with breakthrough, impact, what made it smart and its technical approach below!
Problem
Morgan Stanley manages some of the most complex financial conversations in the world. Advisors needed institutional knowledge at their fingertips, but 100,000 documents updated constantly could not be searched at conversation speed.
What made it smart
The evaluation framework built before deployment is the hidden key. Rather than launching and iterating publicly, Morgan Stanley tested rigorously internally first. This is what allowed the system to grow from a limited set of queries to covering the full knowledge base without quality degradation.
Technical approach
GPT-4 embedded in internal workflows via the AI Assistant for Q and A and Debrief for meeting summaries. Retrieval runs over controlled, approved Morgan Stanley content. Daily regression testing and evaluation sets monitor quality, compliance, and retrieval accuracy. Debrief uses Zoom recordings with consent to generate summaries, action points, CRM notes, and draft follow-ups.
Strategic lesson
Scaling GenAI in a regulated environment requires building the evaluation infrastructure before building the features. Quality assurance is what makes adoption sustainable.
Reflection question
In your organisation, how much time do experts spend finding information rather than applying it? And what would change if that search time dropped by 80%?