/

Guides

/

Guide · AI assistants

AI agents for financial advisors: real examples and how to evaluate them

What an AI agent is, what the advisor vendors that describe agents actually say, a four-level autonomy ladder, an eight-point rubric to evaluate any agent, and a 30-day pilot plan.

By the Verlo team

Vendor pages checked: November 2026

Vendor note. Verlo describes parts of its product as agents, so we have an interest in this topic. We describe every other vendor only from its own announcements and pages, checked in November 2026 and linked below, and we say when something was early access or scheduled. The autonomy ladder and rubric are ours, and we have not tested these products hands-on. This guide is part of our best AI software for financial advisors guide, and the method is in how we evaluate AI software for financial advisors. If something here is out of date, tell us at support@verlo.finance.

The short answer. An AI agent carries out a task, or a series of tasks, using your data and tools, instead of only answering a question. Judge any agent on how much it can do without you (its autonomy), what it can read and change, who approves actions, whether you can audit what it did, and how it behaves when data is wrong. Start at the lowest autonomy that saves time, pilot it on real work for 30 days, and expand only on evidence.

“Agent” is the most used and least defined word in advisor technology. Some products that use it draft an email for you to send. Others change records on their own. The difference matters more to your compliance program than to your demo experience, so it helps to have a way to tell them apart.

What counts as an agent

A chatbot answers. An agent acts: it prepares the meeting brief, chases the missing document, fills the form and routes it for review, or updates the CRM. What makes that safe or unsafe is not the label but the design: well-defined tasks with clear inputs, clear outputs and a point where a person reviews the work.

A four-level autonomy ladder

Level

What it does

Example

Main risk

1. Suggest

Answers questions and recommends actions; you do everything

A planning assistant that flags an opportunity

Wrong answers from stale or mismatched data

2. Draft

Prepares work product for you to review and send

A follow-up email or a pre-filled form waiting for your approval

Approving too quickly

3. Act with approval

Makes changes after you approve each one

A CRM update you accept one at a time

Approval fatigue

4. Act within limits

Acts on its own inside rules you set, with logs

A background agent that flags overdue tasks or sends a scheduled reminder

Errors at scale if the rules are wrong

Most of the products described below sit at levels 2 and 3, which is a sensible place to begin.

Real examples

These are the agent-style features vendors describe, with what they say about status and approval.

Vendor and feature

What it is described as doing

Status as described

Approval model

Jump AI Associate

Designed to update CRM records and draft and send client emails

Announced March 26, 2026; Jump’s site lists it as included in all plans

Each action reviewed first

Wealthbox AI Assistant, Playbooks and Agents

A conversational layer over CRM data; saved prompts that run multi-step workflows such as onboarding; background agents that flag overdue work

Announced March 2026, in early access at the time; Wealthbox’s site now says AI Agents are available on all plans (November 2026)

The Assistant proposes updates you review and approve with a click, and agents ask for approval before they add or change anything

Salesforce Agentic Advisor

A background “digital teammate” that prepares meetings, captures notes and next steps, prioritizes the day and flags at-risk clients

Announced June 18, 2026: Meeting Concierge listed as generally available in June 2026; Run My Day and a client details page scheduled for July or August 2026; others “coming soon”. We did not confirm which scheduled items have shipped

Configurable inside Agentforce

RightCapital Iris

An agent in the planning platform that flags data problems, reviews cash flows, suggests opportunities and builds options toward a target probability of success

Launched June 23, 2026 on its Premium and Platinum plans

Outputs come from RightCapital’s calculation engine

Altruist Hazel

An assistant that works from notes, email, CRM data and account data, with tax planning and, from September 2026, a financial planning agent

Tax planning added February 2026; planning agent announced September 1, 2026 and listed as new on Hazel’s pricing page

Hazel’s home page says every decision rests with the advisor and nothing reaches a client without a human’s approval

Claude for Financial Advisors

A plugin bundling advisor workflows with connectors to systems such as Wealthbox, Salesforce, Schwab and Docusign

Available as a plugin; Anthropic recommends its Enterprise plan for RIAs

Drafts and CRM updates wait for your approval

Verlo

Multi-step workflows such as onboarding, review preparation and client follow-up across your connected data

Described on Verlo’s site

Drafts go to you for review first

RightCapital’s Iris panel for Plan Builder, with a target probability of success and levers to adjustRightCapital’s Iris panel summarizing a proposed plan and flagging a tax opportunity
RightCapital’s Iris: Plan Builder and Find Opportunities. Source: rightcapital.com, captured November 5, 2026.
Anthropic’s page for Claude for financial advisors, headed Focus on financial advice, not admin, with Get the plugin and Contact sales buttons
Anthropic’s page for Claude for financial advisors. Source: claude.com, captured November 6, 2026. Captured in the site’s dark theme.

An eight-point rubric

Score each agent you consider from 0 to 2 on these. Anything below 1 on the first four is a stop.

  1. Scope and permissions. What can it read, and what can it change? Prefer read-only until you trust it, and permissions that follow your existing access.

  2. Approvals. Who approves each action, and can you set approval by action type?

  3. Data and matching. Which systems does it connect to, and how does it match people and households across them? Ask about the “Bobby Smith” problem.

  4. Audit trail. Can you see and export what it did, when and for whom?

  5. Failure behavior. What happens when the data is wrong, the same person appears twice, or a system is down?

  6. Security and regulation. It is likely to be a service provider under Regulation S-P’s definition if it handles client information. See our five security questions and due-diligence checklist.

  7. Measurable outcome. What task does it replace, and how will you measure time and errors?

  8. Exit. Can you turn it off, take your data and keep your records if you leave?

A 30-day pilot

  1. Week 1: choose one task. Pick something repetitive with a clear check, such as meeting prep or document chasing. Write the baseline: time per task and error rate.

  2. Week 1: set permissions and approvals at the lowest level that still saves time, and record them in your AI-use policy.

  3. Weeks 2 and 3: run it on real work, with a person reviewing every action. Log each correction and each surprise.

  4. Week 4: measure and decide. Compare time and errors with the baseline, review the log, and decide whether to expand, hold or stop.

  5. Keep the evidence in the vendor file, with the pilot results and the decision.

Frequently asked questions

What is an AI agent for financial advisors?

Software that carries out a task or series of tasks using your data and tools, not just answering a question, such as preparing a meeting brief, chasing a missing document or filling a form for review.

Are AI agents safe for client data?

It depends on the design and the vendor. An agent that handles client information is likely to be a service provider under Regulation S-P’s definition, so you need due diligence, written terms and a policy. Limit it to read-only or approval-gated actions at first, and keep an audit trail.

Which advisor tools have AI agents?

Jump, Wealthbox, Salesforce, RightCapital, Altruist, Anthropic and Verlo all describe agent-style features, at different stages and with different approval models. Several were early access or scheduled when we checked.

Should an agent be able to send client emails on its own?

We would not start there. Draft-and-approve (level 2 or 3) captures most of the time savings with far less risk. Several vendors describe actions that wait for your approval.

How do I measure whether an agent is worth it?

Pick one task, record the baseline time and error rate, run the agent on real work for 30 days with review, and compare.

Go deeper

Updates to this guide

  • November 17, 2026: First published.

Sources

Checked November 5, 2026: Jump AI Associate, Wealthbox AI announcement, March 2026, Wealthbox AI Notetaker update, July 15, 2026, Salesforce, Agentic Advisor for Financial Services, June 18, 2026, RightCapital Iris, Altruist Hazel tax planning, February 10, 2026, Altruist financial planning agent, September 1, 2026, Claude for Financial Advisors. Re-read November 15, 2026: Wealthbox AI Agents.