Claude for Financial Advisors: How Much Time Comes Back to the Advisor?

The release of Claude for Financial Advisors has prompted plenty of discussion about what it could mean for advisory firms. I wanted to explore one practical question for myself: how much time comes back to the advisor once the output has been reviewed and verified?

I can see the appeal. The plug-in includes skills for meeting preparation, follow-up, prospect intake, portfolio reviews, and estate and tax briefs. Gathering information and preparing that work takes time. Having help could create more room for client conversations.

After my initial tests, that question about time savings is still open. It is what I would encourage firms to test across the plug-in’s skills before a wider rollout.

What I tested

I built a synthetic RIA book of business with made-up clients, households, accounts and planning records. Then I tested Anthropic’s Claude for Financial Advisors plug-in using that information.

My initial tests included the pre-meeting preparation, post-meeting notes and follow-up, and prospect-intake skills. I did not assess every skill.

No real client information was involved. I used an uploaded spreadsheet rather than live connections to a firm’s systems, and I did not measure the time required.

The meeting-preparation skill showed some of the potential. When the records lined up, it brought notes, open tasks, account details and plan information into a prep document with an agenda and talking points. It also called attention to missing information and discrepancies.

There is value in that. An organized draft gives the advisor a useful starting point.

A note that made me pause

In a brief for a fictional client named Elena Marchetti, the agent also wrote: “Elena Reyes is a different client and her data is not included in this.”

I did not see it mixing those clients’ data. It was explicitly saying the other client’s information had been excluded. Still, I wondered why that statement appeared in the brief.

It could have been related to how the agent interpreted the spreadsheet. I have not established the cause. The note alone does not show that the output was wrong or that live connectors would behave the same way.

It did leave me with something to look into during review.

Anthropic is clear that the plug-in prepares briefs, summaries and draft analyses for advisor review. My experience reinforced the importance of including that step when testing the value of its skills.

Test the way your team will work

Not every advisor will use connectors for every task. Some may work from uploaded spreadsheets or documents, or use uploads alongside connected systems. The plug-in’s published instructions support pasting or uploading information when a system is not connected.

That means the upload experience matters, too.

My mock book of business does not tell us how a real firm’s live connectors will perform. But the questions that arose in my tests are worth exploring if your advisors plan to use uploaded information.

A firm should test the way its team actually intends to work.

Include the time to review and verify

Michael Kitces has raised questions about the efficiency of advisor-built workflows and whether usage costs could offset the savings advisors expect. My initial tests raised a practical question within that value equation: how much time is required to get from the first output to reviewed work an advisor is ready to use?

Time savings are one part of the value equation. A firm also needs to weigh the value of that time against the cost of using the tool, including token usage—the charges for the information the model processes and generates. I haven’t measured either in these initial tests.

A meeting brief, a follow-up draft and a portfolio review serve different purposes. For each, I would measure through to the point when the advisor has reviewed the work, resolved outstanding questions and is ready to use it.

The plug-in may save enough time gathering and organizing information to make that review well worth doing. The amount will depend on the skill, the firm’s data and how its team works.

Put the skills to the test

Before a wider rollout, I would take three steps.

Build a synthetic book of business that reflects your firm. Use entirely made-up clients and information, but mirror your data structures, systems, household relationships and everyday challenges. Include shared names, trust-titled accounts, prospects without accounts, missing information and records that don’t line up across systems. Define what a correct answer should contain for each case.

Compare the full workflows. Test each skill you plan to use with the uploads or test connections your team will work with. Time your current process and the plug-in process through to the same point: reviewed work an advisor is ready to use. Include a person’s approval of each client match, verification, follow-up questions and any corrections. Note what still requires returning to a source. Compare the accuracy, completeness and usefulness of the finished work alongside the time.

Use agents to help evaluate repeated runs. Automate checks against known answers and have agents score outputs using the same criteria each time. Those could include accuracy, completeness, whether information belongs to the right client, and statements the source material does not support. Validate the scores against known answers and human-reviewed examples. Once the scoring is reliable, people can focus on flagged results and review a sample of outputs that scored well, rather than manually checking every synthetic test run.

Keep results separate for each skill. A useful meeting-preparation test does not tell you how well a portfolio-review or follow-up skill will work.

Agents can help make testing more manageable. Advisor review remains part of using the resulting work with real clients.

My takeaway

These were early tests to generate observations, and that is the extent of what I tested. I see promise in Claude for Financial Advisors, but I have not established how much time its skills would save in a real practice.

For firms considering the plug-in, my recommendation is to test it against the way your own team works. Include the time to review and verify, and weigh the value of any time saved against the cost of using it.

If you are testing the plug-in in your practice, I’d be interested in hearing what you’re finding—which skills are helping and what work remains before their output is ready to use?

Sources: Anthropic, “Claude for Financial Advisors” and the published plug-in repository. Test observations are from my initial tests using synthetic data.