Wealthbox AI Agent: What It Automates and What to Review

Wealthbox's AI agents are built to run in the background of your CRM: ranking client touchpoints, creating tasks, drafting outreach and tidying records. They automate preparation, not decisions. Before relying on any output, an adviser should check the facts against the record, read every client-facing draft in full, and confirm the change that was made is the one intended.

The wording matters, because Wealthbox sells several things under one banner: an AI Assistant, Playbooks, Agents, an AI Notetaker and AI for Reports. As of November 2026 the company's own product page still labels Agents "coming soon", even though Wealthbox's March 2026 early-access announcement described them alongside the Assistant and Playbooks. Plan on the basis that what you see in your own workspace is what you have.

Follow me on Instagram@sagnikteaches

Four AI features, one word, different review needs

Wealthbox's pricing page lists AI Assistant, AI Agents, AI Notetaker and AI for Reports as separate rows in the plan comparison. They behave differently, so they need different checks. Treating them as one "AI switch" is the first mistake.

Connect on LinkedInSagnik Bhattacharya

The Assistant and Notetaker are the ones advisers can try now: Wealthbox lists the Assistant on every plan, and the Notetaker is a paid add-on. At the plan prices on Wealthbox's pricing page ($59, $75 and $99 per user a month for Basic, Pro and Premier), the add-on is a further $49 per user a month, which I cover in the Notetaker cost comparison. Each feature below gets the same two questions: what does it do for you, and what must a person check?

Subscribe on YouTube@sagnikteaches

What the Assistant reads, and why that is the review boundary

Wealthbox says the Assistant reads notes, emails, tasks and opportunities directly in the record you are looking at, rather than a synced copy. That is a real advantage over a bolt-on tool, and I compare the two approaches in built-in versus bolted-on CRM AI. But "reads the whole record" also defines the limit: it can only be as right as the record.

That gives you a simple rule. The Assistant is a fast reader, not a source. Anything it tells you about a client is a lead to verify in the underlying note, account page or custodian statement before it goes into advice, a file note or a message to the client.

Worked example: Monday morning on a six-adviser firm

Say a six-adviser firm has 140 open tasks and 18 meetings across the week. An adviser types: "What needs my attention today?" The product page describes the result as a combined view of open tasks, overdue workflow steps and upcoming meetings. Illustrative output:

Today (illustrative):
- 3 overdue workflow steps: annual review pack (Household A), ID refresh (Household C), fee agreement chase (Household F)
- 2 meetings: 10:00 Household B, 14:30 Household D
- 4 tasks due: call back re pension transfer, send valuation, update risk profile, book review

The adviser's review takes about four minutes and has three parts. First, open the three overdue steps and confirm they really are overdue; a task completed on paper but not ticked in the CRM will look overdue. Second, compare the list count with the CRM's own task view for the same filter: if the Assistant says four due tasks and the task list shows six, two are being missed. Third, ask a follow-up question on the item that matters most ("Why is Household F's fee agreement overdue?") and read the note it cites, not just its summary.

Illustrative numbers: if skipping the check saves four minutes and one answer in twenty hides a missed item, the habit is cheap insurance. For a list that drives the whole day, keep the check until you have measured a near-zero miss rate for a month.

Drafts the Assistant writes: the read-it-aloud rule

Wealthbox's page says the Assistant drafts personalised client communications that you review before sending. The review is part of the design, but it is easy to skim. The failure pattern is predictable: a draft is fluent, so it reads as correct.

Use a three-line checklist on every client-facing draft:

1. Names and facts: correct client, correct spouse or partner, correct account type, correct dates. Check each against the record, not against the draft's own earlier sentences.
2. Promises: delete any sentence that commits the firm to a timeline, a return, a product or an action you have not agreed. AI drafts like to reassure.
3. Tone against the relationship: would you say this to this client, with this history? A warm birthday line to a client in the middle of a complaint is a classic miss.

Before and after: a check-in message

Illustrative draft from a prompt of "draft a check-in for [first name] after the market fall":

Hi [first name], I hope you're well. Given recent volatility, I wanted to reassure you that your portfolio is well positioned and your retirement plans remain on track. Let me know if you'd like to talk.

Two phrases make claims the adviser has not verified: "well positioned" and "on track". The edited version reads:

Hi [first name], I know the last few weeks have been unsettling. I'm reviewing the household's plan this week and will send you a short note by Friday. If you'd like to talk it through sooner, reply and I'll find a time.

The edit removes a suitability-style statement, adds a dated, checkable promise the adviser can keep, and gives the client a next step. Whoever owns your compliance rules should confirm whether drafted messages need to be archived with the edits or only the final version; Wealthbox's page does not settle that for you.

Playbooks: review the recipe once, then sample the output

A Playbook is a request you save once and run again. Wealthbox's page says it runs the same way every time, and that the library over time becomes the firm's operating manual. The risk shifts from the output to the instruction. A mistake in the saved request is copied into every run.

Treat a Playbook the way you would treat a new firm procedure.

1. Write it down in plain words first: what it reads, what it produces, what it must not do.
2. Run it on three households you know well and compare the result with what you would have written.
3. Name an owner, give it a version date, and review it every quarter or after any change to your process.

A filled-in Playbook specification

Name: New client onboarding check (v1, owner: operations lead)
Reads: household record, open tasks, onboarding workflow
Produces: list of missing items (ID, fee agreement, risk profile, data consent)
Must not: send anything to a client, mark a step complete, or edit notes
Review: operations lead checks the first 10 runs, then 1 in 5

Notice the "must not" line. A saved request that only reads and lists is low risk. One that also changes fields is a different thing and deserves a slower rollout. If you are choosing what to automate first, the line between administration and advice in what AI agents can do for advisers without advising is a useful guide.

Agents: what Wealthbox lists and what is still unproven

Wealthbox's own page describes Agents that watch the book and act without being asked. The list: client monitoring, task creation and assignment, communication drafting for birthdays and check-ins, compliance workflows, CRM hygiene and daily planning. Kitces's April 2026 AdvisorTech roundup, later reprinted by InvestmentNews, gave one concrete example: an agent set up once to deliver a daily list of the most important client touchpoints, built from each client's contact and meeting history. The same piece pointed out that several of these jobs (overdue-interaction reports, birthday reminders, onboarding workflows) can already be built with ordinary CRM automation, and questioned how much time a natural-language set-up really saves.

That framing is useful for a small firm. For each agent job, ask whether a plain rule would do it. A birthday reminder is a rule. "Who has gone quiet and matters?" is a judgement, so an agent has a case.

Because an agent acts without a prompt, the review step has to move from "check this answer" to "check this log". Wealthbox's launch announcement claims every action is logged, auditable and compliant by default, and says the Assistant always asks for confirmation before it acts. Neither that announcement nor the Kitces coverage describes approval settings for Agents, which by design act without being asked. Ask Wealthbox, in writing, what the log shows (who or what acted, what changed, the before and after values) and whether you can require approval before an agent edits a record or creates a task for a colleague.

Example: an agent that creates tasks, and the duplicate-task mistake

Imagine a four-adviser firm switches on an agent that creates a follow-up task after each meeting. The Notetaker also suggests tasks, and an adviser creates a third by hand. One client meeting now produces three versions of "send valuation". The symptom is task counts creeping up and colleagues ignoring the task list. The fix is to decide in advance which source owns task creation after meetings (one of them, not all three), then do a weekly check of duplicate task titles for the first month.

The Notetaker: the one place a wrong detail lasts years

Wealthbox's Notetaker page says it joins Zoom, Teams, Webex and Google Meet calls, also records in-person meetings, phone calls and uploaded audio, then produces a searchable transcript, a structured summary and suggested tasks it calls Ideas. It can prepare a brief from the past 12 months of activity, and it drafts a follow-up email and CRM note that you can edit. The page says you start and stop it for each meeting and that a mid-conversation restart is marked in the record.

The record is where review matters most, because a client file note is the document you may be asked to produce later. The page does not describe retention or archiving rules in text, so confirm those with Wealthbox and your compliance lead before the first client meeting. I cover consent wording in the consent and template guide and the wider question in AI note takers for financial advisers.

A five-point check on each meeting summary

1. Numbers: every amount, percentage, date and account name matches what was said. Use the transcript to check any figure.
2. Decisions versus discussion: a summary can turn "we talked about" into "the client agreed". Mark which items were agreed and which were only raised.
3. Omissions: a client's stated concern, vulnerability or change in circumstances that never reached the summary. Skim the transcript for words like "worried", "health", "divorce", "retire".
4. Suggested tasks: keep only those with a real owner and date; delete the rest.
5. The follow-up email: apply the three-line draft check from earlier.

Example: the summary that sounded decided

Illustrative transcript line: "We could look at moving some of it into the cash account, but I'd want to see the tax position first." Illustrative summary line: "Client agreed to move funds to cash." The adviser's edit changes it to "Client raised moving some funds to cash; adviser to review the tax position first." One keeps the record true, and the other would have created a suitability problem. This is the kind of change a quick skim misses, so build a rule: any sentence containing "agreed", "decided" or "instructed" is checked against the transcript every time.

AI for Reports: check what the filter left out

Wealthbox introduced AI for Reports in November 2024 for its Premier and Enterprise plans, and the pricing page still shows it as absent from Basic and Pro. You start a new report, click "Generate with AI" and type the criteria; it pre-fills the report's name, filters and display fields, which you can change before saving. Those pre-filled filters are what you are reviewing. The risk in any prompt-driven report is a silent filter: "clients with no review in 12 months" may exclude households with no review recorded at all, or those where the date sits in a custom field.

A reconciliation habit

For any AI-built report, run one control: count the rows, then count the same population with a plain filter. Illustrative figures: the AI report lists 42 households overdue for review. A plain filter on "last review date before 1 November 2025" returns 51. Nine households differ. Looking at those nine usually reveals the cause, for example a missing date or a different field name, and fixes the prompt for next time. Save the corrected prompt so the correction is not lost.

Set the review rules before you switch anything on

A firm of two to ten advisers can write its rules on one page. Here is a filled-in version for illustration:

AI use rules (illustrative, v1)
1. AI output is a draft. Nobody sends, files or acts on it unread.
2. Client-facing text: the author of the message reads it in full and owns it.
3. Meeting records: the adviser who ran the meeting reviews the summary within 2 working days.
4. Agents: no agent edits client records or sends messages until the log has been reviewed weekly for a month.
5. Reports: any AI report used for decisions is reconciled against a plain filter.
6. Errors found are logged (what, where, how caught). Review the log monthly.
7. Owner of this page: compliance lead. Review every quarter.

Rule 6 is the one most firms skip, and it is the one that tells you whether the tool is earning its place. If after a month you have logged few errors and saved real time, widen the use. If the log is long, narrow it.

Permissions: who sees what the AI sees

Wealthbox's Notetaker page says meeting records and reports follow your existing roles and permissions, and its June 2026 post on the Assistant says the Assistant works within the permissions and access controls already set in Wealthbox. Test that rather than assume it. Pick a household that only one adviser should see, log in as a colleague with a different role, and ask the Assistant about that household. If it answers, you have found a permission gap before a client does. Repeat the test for the Notetaker's stored summaries and for any report that spans households.

Illustrative case: an administrator with access to scheduling but not to financial notes asks the Assistant for "everything on Household B". The correct result is a refusal or a reply limited to what that role may see. Write the result of the test into your pilot log with the date, because it is the evidence you will want if a regulator or an auditor asks how you controlled access.

Turning the findings into a go or no-go

After the pilot, score each feature on three questions: did it save time, did the review step catch real errors, and could a colleague run the review without you? Illustrative scoring for a small firm: Assistant, yes, yes, yes, keep and widen; Notetaker, yes, yes, no (only the lead adviser knew what to check), keep but train the reviewers first; Agents, not yet available in the workspace, revisit in a quarter. Put the decision, the date and the name of the person who made it in the same place as the use rules.

Measuring a one-month pilot

Pick one adviser and one client list. For four weeks, record three figures: minutes saved per task against your normal way, share of AI outputs that needed a correction, and share of corrections that were material (a wrong fact, an unsupported promise, a missing concern). Illustrative result: 60 outputs, 14 corrected (23%), 3 material (5%). A 5% material-error rate on client-facing text means the review step is doing real work; it would be hard to defend skipping it. If the material rate falls to nothing across several hundred outputs, you could sample rather than read everything, with your compliance lead's agreement.

For wider context, Jump with Wealthbox shows what a third-party notetaker syncing into the same record looks like. If you would rather map the jobs and reviewers with someone, that is what my AI implementation consultation covers, at the contact page.

What this does not settle

Wealthbox has not published, on the pages I read, how Agents are priced, what approval options exist, or what the retention rules are for Notetaker records. The vendor's pricing page lists AI Agents as a row without stating an extra charge, and I could not verify one either way. Ask three questions in writing before a rollout: is there an extra cost; can an agent's actions require approval; and how long are AI-generated records and logs kept, and can they be exported for your archive? Until those are answered, run the features in read-and-draft mode only.

Questions advisers ask before switching Wealthbox AI on

Is the Wealthbox AI agent the same as the AI Assistant?

No. The Assistant answers when you ask, in a chat-style box, and can search records or make changes you request after you confirm them. An Agent is set up once and then works in the background on a schedule or trigger without being prompted. Wealthbox's own product page describes Agents as the next step after the Assistant, so check which one your plan currently shows.

Does the AI agent replace a Wealthbox workflow?

Not for fixed, rule-based steps. A workflow that creates the same five tasks when a client signs is predictable and needs no AI. An agent suits jobs that need judgement across notes and history, such as ranking who needs contact. Keep rules in workflows and use AI where the input is untidy.

Who is responsible when an AI-drafted note is wrong?

The adviser and the firm. The record carries your name and your compliance obligations whether a person or software drafted it. That is why each AI output in this tutorial has a named review step, and why you should write down who does the review.

Should I pilot with one adviser first?

Yes. Switch the features on for one adviser and one household list for a month, count how many AI outputs needed correction, then decide whether to widen it. A pilot also gives you real figures to show your compliance lead.

Further reads

• Best AI App for Wealth Management: Jump, Zocks and Wealthbox — Sets built-in features beside the specialist adviser apps.
• Jump vs Zocks: Which AI Notetaker Fits Your Compliance Set-Up? — Compliance questions for the two main notetakers.
• Can Financial Advisers Use ChatGPT? What Compliance Allows — What compliance allows when AI sits outside the CRM.
• How Advisers Use AI to Prepare for Annual Client Reviews — Uses CRM data to prepare annual reviews.
• HubSpot for Financial Advisers: Pipeline and Where Breeze Stops — The same questions asked of a different CRM.
• Zapier for Financial Advisers: Five Automations for Onboarding — Five onboarding Zaps for a small advisory firm, each counted in monthly tasks, with the records and compliance checks to settle first.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Wealthbox AI Assistant, AI Notetaker and pricing pages; Wealthbox's March 2026 early-access announcement, its June 2026 Assistant post and November 2024 AI for Reports post; Kitces AdvisorTech roundup (April 2026), reprinted by InvestmentNews.