What Does AI Implementation Look Like for a Wealth Management Firm?

Most advice on AI for financial services is written from thirty thousand feet. This post goes the other way. It walks through what an engagement with a wealth management firm looks like in order, from the first conversation to the point where the firm runs the thing itself.

The shape is based on an advisory firm we worked with, where six specialized agents draft client follow-ups from approved content and a human approves every message before it reaches a client. The full case study has the setup. Here is the story of how a firm gets there.

Phase one: the argument about where to start

The first meeting is usually not about technology. It is a list of complaints. Advisors are tired of the recap emails after meetings, the document chasing, the notes that have to be rewritten for the CRM, the same explanation typed for the fortieth time.

The temptation is to start with all of it. We push the other way. Pick one workflow, name who does it today, what triggers it, and what a good result looks like. Client follow-up wins this argument most of the time. After a meeting an advisor owes the client a recap, next steps and often a document or two. It is repetitive, written in a consistent voice, and easy to do badly when the calendar is full.

The output of phase one is a paragraph, not a plan. One workflow, one team, one definition of done. If the firm cannot write that paragraph, it is not ready to build, and it is better to find that out now.

Phase two: the content library nobody has written down

This is the phase that surprises people. The agents are only allowed to draft from material the firm has approved. So the question becomes: what has the firm actually approved?

Usually the answer is scattered. Standard explanations of how a particular account type works live in three advisors’ heads and a PDF from 2021. The meeting recap template exists in a few slightly different versions. Phrases that compliance likes are known informally.

We help the firm gather this into one approved set: the meeting notes template, the standard explanations, the tone and wording the firm wants to use. It is slow, unglamorous work, and it pays for itself. The agents do not invent advice. They assemble a message from approved pieces plus the details of the specific meeting. If something is not in the library, the draft says so and a person fills the gap.

Phase three: deciding who approves, before the first draft exists

Before any agent writes a word, the firm decides who reviews what it writes and what a rejection looks like. This is the design choice that makes the rest workable.

In the firm we worked with, every outbound message is approved by a person before it reaches a client. The agent removes the blank page. The advisor keeps the final word. Nothing leaves the firm on the agent’s authority alone.

Settling this early matters for a human reason. Advisors who are asked to trust a system they had no say in resist it. Advisors who designed the approval step use it. We spend a surprising share of the early weeks on this one decision: who sees the draft, how they edit it, what happens when they reject it, and how that rejection teaches the system.

Phase four: the first agent, kept small

Now something gets built. One agent drafts one kind of follow-up. A handful of real meetings go through it. A person approves each draft, edits where needed, and the edits are recorded.

This phase is boring on purpose. The goal is not impressive output. It is evidence: how often does the draft need changes, what kind of changes, and are they changes to the content library or to the agent’s handling of it. Both are fixable, but they are fixed in different places, and you cannot tell which without watching.

Phase five: from one agent to six

Once the first agent earns trust, more are added. In the advisory firm there are six, each specialized, each starting from what the earlier ones taught us. Specialization is the point. An agent with a narrow job is easier to test, easier to approve and easier to correct than a single agent that tries to do everything.

Each new agent follows the same pattern as the first: approved content in, draft out, a person approves. The firm did not add a second approval philosophy as it grew. It added more drafters under the same rule. If you want the longer version of this sequencing, the AI implementation roadmap walks through it stage by stage.

Phase six: handing over the keys

The last phase is the one firms ask about last and should ask about first. Who owns this when we are done?

The firm owns the code, the data and the infrastructure it paid for. That matters in wealth management, where client relationships and process knowledge are the asset. A tool that holds your workflow hostage is a different bet, and who maintains custom AI is the question to ask before you sign anything. If you are still deciding whether to bring in outside help at all, what is an AI implementation partner explains the role.

What did not change

Through all six phases, the firm’s own legal and compliance review stayed with the firm. The controls we build, human approval on every outbound message and a defined content library, do not replace that review. They give it something concrete to inspect.

The advisors still talk to their clients. What changed is that the first draft no longer starts from a blank screen.

The question for your firm is smaller than “how do we adopt AI.” Which one recurring task would your best advisor happily never type again?