Morgan Stanley's published Debrief workflow turns a consented client meeting into notes, action items, a Salesforce record and an email draft that the adviser reviews. The example is useful because it connects AI to the work that follows the conversation. A smaller financial services firm can apply the same logic by deciding what should be captured, where the record belongs, who reviews the output and how commitments receive an owner.
A client meeting can go well and still create a messy operating handoff.
Someone needs to record what was discussed, update the client file, identify the next actions and prepare the follow-up. The meeting may end at 11:00. The notes are written later, perhaps after two more calls and an inbox full of new requests.
That delay affects more than administration. A colleague opening the account may see an incomplete record. A promised document may not have an owner. The client may receive a follow-up after the useful moment has passed.
Morgan Stanley Wealth Management designed AI @ Morgan Stanley Debrief around this part of the workflow.
According to the firm's June 2024 launch announcement, Debrief generates notes and surfaces action items from client meetings with the client's consent. After the meeting, it summarizes the key points, saves a note into Salesforce and prepares an email for the adviser to edit and send at their discretion.
The technology matters. The shape of the process is what makes the example useful for other financial services firms.
The meeting creates several records
A meeting rarely produces one simple output.
There is the internal record of what was discussed. There may be commitments, requests, decisions and open questions. Someone may need to update a CRM, create a task, send material to the client or ask another team for information.
These outputs serve different purposes. The client email should be clear and appropriate to send. The internal note may need more context. A task needs an owner and a date. A material instruction may require a separate approval or formal record.
An AI-generated summary can help prepare these items. The workflow still needs to decide where each one belongs and what happens before it becomes authoritative.
Morgan Stanley's published design gives two clear destinations: a note in Salesforce and a draft email for adviser review. That is more useful than leaving the output in a separate chat window for someone to move later.
Consent belongs at the beginning
Morgan Stanley states that Debrief uses meeting recordings with client consent.
That decision sits at the start of the workflow because the firm is capturing a client conversation. A smaller firm considering a similar capability would need to define when capture is appropriate, how consent is obtained and recorded, which participants are covered and what happens when consent is withheld.
The information boundary matters too. A client meeting can include personal data, financial information, commercially sensitive material and advice. The approved tool, storage location, retention period and user access should reflect the firm's obligations and the actual purpose of the workflow.
There should also be a workable path for meetings that are not captured. Otherwise the new process may create a second class of client records whenever a participant declines or the technology fails.
Human review needs a specific job
Morgan Stanley's announcement says that the adviser edits the email and decides when to send it. OpenAI's case study also says advisers review and adjust AI-generated outputs before finalizing them.
That review is a defined step. The adviser is responsible for deciding whether the note and follow-up accurately reflect the conversation and are suitable for the client.
For another financial services workflow, the reviewer may need to check names, figures, instructions, commitments, advice, dates and any language that could be misunderstood. The design should make that check practical by showing the draft alongside the relevant recording, transcript or source material.
A reviewer who has to reconstruct the whole meeting from memory receives very little help. A reviewer who can see the proposed note, the source and the flagged action items can make a more focused decision.
Action items need somewhere to go
Identifying an action item is only part of the work.
The next question is whether it becomes a task with an owner, a deadline and enough context to complete it. Some actions may belong to the adviser. Others may need operations, compliance, investment, finance or client service involvement.
This is where a meeting assistant can either reduce follow-through work or simply produce a better list of things someone still has to organise.
The first implementation does not have to automate every handoff. It can prepare proposed actions for review and let the responsible person assign them. The important point is that the workflow has a visible end state. The client record shows what was agreed, what remains open and who is expected to act.
Test the details people care about
OpenAI reports that Morgan Stanley developed evaluation datasets for different meeting types and tested whether Debrief captured critical action items without introducing errors.
A smaller firm can use the same principle with examples that reflect the conversations the team actually has, including routine reviews, meetings with several participants, unclear requests, changed instructions and conversations where no action is required.
The useful measure may include preparation time, review effort, missing actions, time to follow-up and record completeness. A time saving on the first draft can be valuable, although it should be considered alongside the work required to check and finish the process.
What can a smaller firm implement first?
A smaller asset manager, fund administrator, wealth firm or corporate services business could begin with one meeting type and one team.
The firm can map what currently happens from the moment a meeting is scheduled through to the final note and completed actions. That reveals the systems, information, consent, review and ownership decisions around the work.
The first version might prepare a structured internal note and a follow-up draft without sending or assigning anything automatically. The team can compare the output with real meetings, record corrections and decide whether the workflow is dependable enough to connect to the CRM or task system.
This narrow approach gives the firm evidence. It shows whether the source is usable, whether reviewers trust the draft, which exceptions occur and whether the handoff actually improves.
Six decisions before building the workflow
Which meetings are in scope?
Start with a recognizable meeting type and define which conversations remain outside the first version.
How is consent handled?
Establish how capture is explained, recorded and declined, including the fallback process.
What should be produced?
Separate the internal note, client-facing draft, proposed actions and any formal record that follows another process.
Where does each output belong?
Name the CRM, document location, task system or review queue that should receive it.
Who reviews and approves it?
Give the reviewer a clear check and enough source context to perform it.
How will the firm judge the result?
Measure the quality and completeness of the handoff as well as the time spent preparing and reviewing it.
Follow-through is part of the client experience
The strongest part of the Morgan Stanley example is the connection between the conversation and the work that follows.
The meeting produces a record. The adviser receives a draft rather than a finished client communication. Action items are made visible. The workflow preserves a clear place for consent and human judgment.
When Halyard assesses a similar opportunity, we look at the complete handoff: what should be captured, which system holds the record, how the output is checked, where exceptions go and who owns the next action.
A useful first version can be modest. If the next person opens the client record and can understand what was discussed, what remains outstanding and who is responsible, the workflow is already solving a real operating problem.
Further reading
Founder of Halyard, an AI implementation company helping financial services firms find and implement practical AI safely.