Automate the Workflow, Not the Responsibility

Picture a routine client engagement.

A client uploads a tax return. Software begins pulling out numbers, flagging potential opportunities, and drafting recommendations. The advisor reviews the output, and the client receives advice.

Somewhere in that chain, the software stopped being a filing assistant and started participating in the advice. No one announced the moment, and it may be difficult to point to exactly where it happened.

AI is now showing up in financial planning, tax analysis, meeting notes, client relationship management, and portfolio workflows. Each application is pitched as a time-saver, and in many cases it is.

But efficiency comes with a less visible tradeoff: judgment can begin moving from the advisor’s discretion into a workflow the advisor did not design.

The question, then, is whether an increasingly automated workflow can still fulfill a fundamentally human fiduciary obligation.

What’s Actually Being Automated?

It helps to break the work into four layers:

Data. Analysis. Execution. Communication.

Data includes reading documents and parsing tax returns.

Analysis includes spotting planning opportunities, comparing scenarios, and generating recommendations.

Execution includes creating tasks, updating the CRM, and preparing for meetings.

Communication includes drafting emails and summarizing meetings on the advisor’s behalf.

Hardly anyone objects to a tool that files meeting notes.

But the ethical weight grows as automation moves toward analysis and communication, because those layers increasingly shape what the client is told and why.

Where Judgment Lives

Tools such as Holistiplan and Focal help illustrate the distinction.

Based on how the companies describe their products publicly, Holistiplan analyzes tax documents and surfaces planning opportunities, placing much of its value in the analysis layer. Focal focuses more heavily on automating work surrounding meetings, placing much of its role in execution and communication.

Take the tax example.

The software identifies an opportunity. The advisor decides whether it applies, explains it to the client, and the client ultimately makes the decision.

On paper, the human still controls every step that matters.

But important questions remain.

What does the software actually know?

What assumptions is it making?

Are those assumptions visible to the advisor?

What does the advisor independently verify?

And if the AI misses something, who is responsible?

Who Owns the Decision?

Consider three versions of the advisory relationship.

In the traditional model, the advisor authors the recommendation.

In an AI-assisted model, the advisor and AI effectively co-author it.

In a third model, an AI workflow produces the recommendation and the advisor approves it—closer to an editor signing off on a draft the advisor did not originally write.

From the client’s perspective, those three experiences may look nearly identical.

A fiduciary lens suggests several useful questions.

Care: Does automation improve accuracy, or can it cause busy advisors to skip important checks?

Loyalty: Could the tool favor vendor incentives, platform integrations, or proprietary products—even unintentionally?

Competence: How much must an advisor understand about an analysis before relying on it?

Supervision: Is “I reviewed it” enough, or should the advisor be able to explain why the system reached its conclusion?

The “Human in the Loop” Problem

The usual reassurance is that a human is reviewing everything.

But a human can review a recommendation and still trust it too much, lack the time to verify it, or overlook errors because the output appears plausible.

This is often described as automation bias: the tendency to place too much trust in an automated system, particularly when that system usually performs well.

So the standard should not simply be:

Was a human involved?

A better question is:

Was meaningful human judgment exercised?

Involvement can become a checkbox.

Judgment is a practice.

The Relationship Itself

AI is no longer limited to analyzing portfolios.

It can notice that a client has gone quiet, draft the follow-up, remember personal details, schedule the meeting, and write the meeting summary.

Each step may be reasonable on its own.

Together, however, they raise a larger question:

Is AI managing the advisor’s workflow, or is it beginning to manage the client relationship?

When a note arrives mentioning your daughter’s graduation, it matters who remembered.

AI can imitate personal attention without actually being personal.

These tools may free advisors to spend more meaningful time with clients, which is a genuine benefit. But there is a line between supporting a relationship and performing one.

Data and the Black Box

Financial technology can touch some of the most sensitive information a person has: income, tax returns, estate plans, investment information, and family details.

More data can lead to better recommendations, but it can also create greater privacy exposure.

Consent to financial planning should not automatically be assumed to include every form of third-party AI processing. “They probably wouldn’t mind” is not a strong standard for consent.

Then there is the black-box problem.

If an advisor asks, “Why?” and the answer is simply, “The model identified it,” that creates a problem.

Explainability—whether the advisor can understand how a conclusion was reached—and accountability—whether someone can be held responsible for that conclusion—are not the same thing.

It is possible to have accountability without full explainability.

That is an uncomfortable place for an advisor to stand.

A recommendation an advisor cannot explain is difficult for that advisor to truly own.

A Five-Question Test

Before an AI-generated recommendation reaches a client, an advisor should be able to answer five questions:

1. Accuracy: Is the underlying data correct?

2. Reasoning: Can the advisor explain the “why” without relying entirely on the tool?

3. Fiduciary alignment: Does the recommendation serve the client, and whose interests does it actually favor?

4. Disclosure: Does the client understand the role AI played?

5. Accountability: Is there a clearly responsible human?

If any answer is “no” or “I’m not sure,” the recommendation probably is not ready to go out.

The Bottom Line

AI can automate many of the mechanical parts of wealth management, and that is mostly a good thing.

Advisors buried in paperwork have less time to think, plan, and communicate with clients.

But fiduciary responsibility cannot simply be automated away.

The future is not human versus AI.

It is human judgment plus AI execution.

And the real question is not whether advisors use AI. It is whether they remain accountable for what AI does on their behalf.

One Last Thought

The next time an AI tool promises to “save you hours,” ask a second question:

What decision did it just make for me, and would I have made the same decision myself?

The answer will not always reveal a problem.

But it should always be an answer you know.

From the Digital Kaizen Fellowship

This article is the concluding research project of Roy Zhao, the inaugural Digital Kaizen Fellow.

Over the course of the fellowship, Roy explored how artificial intelligence is changing financial advisory practices, moving from researching individual tools to examining the ethical questions surrounding automation, client trust, and professional responsibility.

His final article, Automate the Workflow, Not the Responsibility, reflects the fellowship's central philosophy: Noise → Signal → Insight → Improvement.

By Roy Zhao | Digital Kaizen Fellow, Summer 2026