AI enablement for an advisory firm is not a tools class. It has four jobs: decide which client information may touch which tool, teach a verification habit that holds when the record has to be true, settle what counts as an advertisement once a model drafts the copy, and convert two or three repeated tasks on the firm’s own documents. Everything else a course catalog offers can wait, and most of it should.
The reason is narrow. The time savings advisors report land almost entirely in client communication, the one part of the practice that three separate rulebooks already govern. Buy a session that skips them and you have paid to accelerate the regulated surface.
The number the trade press read backwards
The most circulated adoption statistic in wealth management is being quoted upside down. The 2026 Morningstar Investor Perspectives Advisor Study surveyed 501 advisors between July 21 and August 16, 2026, and found that the share reporting no AI use fell from 33% in 2025 to 20% in 2026. Several trade outlets ran it as adoption dropping from 33% to 20%. It is the same survey, read in reverse: four advisors in five now use these tools.
The rest of the study is more useful than either headline. Nearly half (48%) said the technology improved day-to-day efficiency. Yet advisors still spend 53% of the week on client work against a 63% target they set themselves, and 56% named admin and operations load as the main barrier to the service they want to give. Efficiency went up. Client time did not move.
That gap is the argument for AI enablement rather than another license: a dashboard showing seats in use tells you nothing about whether the work changed, the case we make at length in our argument that buying licenses is not the same as building capability.
Where the gains land is where the rules are
Two more figures from the same study decide the agenda. Of the advisors surveyed, 57% saw the clearest benefit in client communications: summarizing meeting notes, drafting follow-ups. Only 18% rated these tools highly reliable for investment recommendations or portfolio decisions.
So the capability a firm is buying sits on drafting, correspondence, and meeting records, which is the surface the Investment Advisers Act regulates most closely. The SEC’s 2026 examination priorities, released November 17, 2025, say the Division will review registrant representations about AI capabilities for accuracy, expects policies and procedures to monitor and supervise AI use, and will look at the “training and security controls employed by firms to identify and mitigate new risks.” Training is itself an exam item now.
FINRA’s 2026 Annual Regulatory Oversight Report, out December 9, 2025, carries a generative-AI section for broker-dealers and dually registered firms. It names summarization as the most common use case, expects an enterprise-level supervisory process with controls for accuracy and bias, and holds that AI-drafted communications must still be fair, balanced, supervised, and retained.
There is no standalone SEC rule for AI, which is why the training question gets ducked. The existing rules apply, and they are specific about what a firm has to be able to show.
Block one: the client-information line
This is the hour that cannot be cut, and the one every catalog course reduces to a slide saying “protect client data.”
Start with the rule whose deadline has already passed. The amended Regulation S-P required smaller entities to comply by June 3, 2026 (larger ones by December 3, 2025). Covered firms, advisers and broker-dealers among them, must run an incident response program, notify affected clients no later than 30 days after an incident involving sensitive customer information, oversee service providers, and keep records proving it. A personal chat account holding client nonpublic information is a service-provider question the firm has not answered.
Writing on Kitces.com in November 2025, securities attorney Chris Stanley set out the practical test for applying existing rules to AI tools: know the tool’s terms of use and privacy policy, and do not put client nonpublic information into a prompt without knowing whether the provider trains on it or shares it. Business-tier accounts with terms someone at the firm has read are where most practices land. In the sessions we run, the tool nobody mentions until the second hour is usually the one somebody signed up for personally.
The output of the hour is one page, posted where people work, in three columns: tools inside the line and what may go into them, tools outside it and therefore prose-only, and the information that never leaves the firm’s own systems. Our guide to the distance between an AI policy and safe practice covers why writing the policy and skipping the session leaves you with a document rather than a control.
Block two: verification, because the record has to be true
Advisers Act Rule 204-2 obliges an adviser to keep records that are true and accurate. A fabricated figure in a client email is not a quality problem; it is a records problem. Stanley draws a line worth teaching directly: a notetaker’s recording kept inside the vendor’s own environment may not be a written communication subject to retention, but the moment someone emails the transcript or summary, it is a record like any other.
Teach the failure mode by producing it. Ask for a summary of a plan document holding a number the model is likely to round, then have the room check it against the source. People who have watched a clean paragraph arrive with an invented figure stop treating polish as a signal of accuracy. People who were only warned carry on.
Then make the standard concrete: every number traced to the document it came from, every piece of client-facing language read by the person whose name is on the relationship, and a written rule for when someone stops and asks a human. Three triggers is enough. The output contradicts the source. The question involves a recommendation the firm has not made before. A client asks whether AI touched their plan.
CFP Board’s Generative AI Ethics Guide, published February 2025 as a checklist against the Code and Standards, lands in the same place: safeguard confidentiality, verify what the model produced, keep inside privacy law, and watch for conflicts. A free baseline, not a capability program.
Block three: what counts as an advertisement
This is the block almost no one runs, and the one with a price tag attached.
The Marketing Rule, Advisers Act Rule 206(4)-1, bars an adviser from making a material statement of fact it cannot substantiate on demand, along with untrue or misleading claims and anything not fair and balanced. The rule does not mention AI and does not need to. A bio, a website page, a newsletter, or a prospect email is an advertisement whether a person or a model wrote it.
On March 18, 2024, the SEC settled charges against two advisers over claims about their own AI use. Delphia (USA) Inc. paid a $225,000 penalty and Global Predictions Inc. $175,000, under negligence-based antifraud provisions and the Marketing Rule. Global Predictions had called itself the “first regulated AI financial advisor.” The exposure was in the marketing copy, not in the model.
A session should produce two things: a short list of claims nobody makes about the firm’s process without compliance sign-off, “AI-driven” among them, and a review path for model-assisted copy that matches the one a seminar deck already goes through.
Block four: three workflows, converted in the room
Generic agendas list use cases. A working session converts a few end to end, on the firm’s own material, and leaves the prompts where colleagues can find them.
| Candidate workflow | Why it fits | What to produce in the room |
|---|---|---|
| Meeting notes to follow-up letter | Highest reported gain, reviewed anyway | A template plus the review step before it sends |
| First-pass research and plan narrative | High volume, text-based, no filed number | A prompt that cites the source page for every claim |
| Recurring client correspondence | Repetitive, low judgment, measurable load | Three drafts for the firm’s most common emails |
| Review and annual-meeting prep | Cuts reading load without touching the portfolio | A worked example from a real file, scrubbed |
Three is the number. Two is thin, four stops fitting in the practice time. Note what is absent: anything that generates a recommendation or an allocation. Fiduciary duty does not thin out because a model drafted the output.
Defer the rest: model architecture, tool bake-offs, agentic workflows, service-line redesign. The test for any agenda item is whether it changes what someone does in the first week back at their desk.
One firm, five curricula
An advisory firm is not one seat, and selling it one curriculum is how vendors keep the day cheap. The 2026 Investment Adviser Industry Snapshot from the IAA and Comply counts 16,544 SEC-registered advisers, 92.8% of them employing 100 or fewer people. At that size nobody has a Chief AI Officer; the work lands on an operations lead or a chief compliance officer as a second job.
| Seat | What the session has to teach |
|---|---|
| Advisor | Drafting and summarizing with the relationship owner’s review duty attached |
| Client service associate | Correspondence, scheduling, data requests, and the tool line |
| Paraplanner | Research and plan narrative with sourcing, no generated numbers |
| Operations | Repeated internal work, and where a transcript becomes a record |
| Compliance and marketing | Review paths, substantiation, supervision evidence |
Splitting a day by seat costs nothing extra and changes how much of it survives the week. Buying one session for the whole team and calling it done is the cheapest-looking option and the one that transfers least. Our role-based enablement playbook sets out the full version, and the same logic drives the privilege and confidentiality limits an in-house legal team works under.
Credit and capability are two purchases
Settle this before the invoice, because it is where advisory firms waste the most money on AI enablement.
Credit depends on the sponsor, not on the quality of the teaching. Under NASAA’s model rule, an investment adviser representative needs 12 CE credits a year, split 6 Products and Practice and 6 Ethics and Professional Responsibility, and both the provider and the course must be NASAA-approved. Completions report to the CRD record; miss them and the IAR goes CE Inactive, with two consecutive inactive years blocking renewal. Quest CE’s tracker, updated March 18, 2026, counts 25 jurisdictions adopted, Illinois effective 2026 and Indiana in 2027.
CFP Board runs its own: 30 credit hours per two-year reporting period, 28 general plus one 2-hour Code of Ethics program, from registered sponsors, with non-registered programs self-reported at $60 each.
A hands-on session from a consultancy that is not an approved sponsor earns no credit at all. So split the purchase: satisfy CE through approved providers built for exactly that, and buy the capability separately. Vendor-run advisor programs sit in between. Horsesmouth’s four-week cohort runs four 90-minute live sessions plus small-group clinics and publishes no price (checked October 2026).
What it costs, and against what
Three options, three different things bought.
| Option | Market range | What it buys | Credit |
|---|---|---|---|
| Approved CE courses | $50 to $400 per person | Credit hours and vocabulary; transfer to the firm’s work unmeasured | Yes |
| Hands-on firm session | $2,000 to $15,000 | Practiced capability on named workflows, plus a written tool line | Usually none |
| Custom automation project | $25,000 to $150,000 | A built process, once people are fluent enough to supervise it | None |
Run the arithmetic against what it replaces rather than against zero. A ten-person team recovering half an hour a day each is a simple calculation at the firm’s own revenue per employee, and the Morningstar finding sets the honest target: client time recovered, not minutes saved. We work the same question through for an in-house finance function, and the calendar logic for a public accounting practice heading into filing season is a cousin of the one an advisory firm faces before year-end reviews.
What to do in the next two weeks
- List every tool in use, personal accounts included, and mark each one in or out against your Reg S-P service-provider position and the terms you have read.
- Check what your notetaker retains and where, then decide which transcripts are records and who reviews them.
- Pull your bios, website, and newsletter and strike any AI claim nobody can substantiate on demand.
- Pick the three workflows from work people already repeat, and book CE and the working session as two separate line items.
Frequently asked questions
What should AI training for financial advisors cover?
Four blocks: which client information may touch which tool, a verification habit practiced rather than described, what counts as an advertisement once a model drafts copy, and two or three repeated workflows converted on the firm’s own documents. Model theory and tool comparisons belong in a later session.
Can an advisory firm put client information into ChatGPT or Claude?
Not until two questions are answered: do the account’s terms permit it and does the provider train on what you submit, and does the firm’s Reg S-P service-provider oversight cover that tool. Business-tier accounts with reviewed terms are where most firms land; a personal subscription nobody vetted is the common failure.
Does a hands-on AI workshop count for IAR CE or CFP Board CE?
Usually not. IAR CE requires both the provider and the course to be NASAA-approved, and CFP Board credit comes from registered sponsors or a $60-per-program self-report that still has to meet its standards. Buy credit from approved providers and buy the working session separately.
Is there an SEC rule for AI use by investment advisers?
No standalone rule. Existing ones apply: Section 206 antifraud, Rule 204-2 recordkeeping, the Marketing Rule 206(4)-1, Rule 206(4)-7 compliance programs, and Regulation S-P. The SEC’s 2026 examination priorities say the Division will check AI representations for accuracy and look at firms’ training and security controls.
What did FINRA say about generative AI for 2026?
Its Annual Regulatory Oversight Report, released December 9, 2025, added a generative-AI section expecting an enterprise-level supervisory process, controls for accuracy and bias, and full compliance for AI-drafted communications, which must be fair, balanced, supervised, and retained.
How much does AI training for a wealth management firm cost?
Approved CE courses run roughly $50 to $400 per person. A delivered hands-on session runs about $2,000 to $15,000 at market rates, and custom automation is a separate $25,000 to $150,000 decision. Several advisor-specific programs publish no price, so ask for the per-seat number before comparing.
Did advisor AI adoption fall in 2026?
No. The 2026 Morningstar Investor Perspectives Advisor Study found the share of advisors reporting no AI use fell from 33% to 20%. Some coverage inverted it. Adoption rose; the time advisors spend with clients did not, which is the finding worth acting on.
Should advisors and support staff attend the same session?
They need different material. Advisors are learning to draft and summarize while carrying the review duty. Client service associates and paraplanners are learning where the tool line sits and what sourcing a plan narrative requires. One combined session teaches drafting and skips review.
Can we use an AI notetaker in client meetings?
Most firms can, once three things are settled: the client is told and agrees, the vendor sits inside the firm’s service-provider oversight, and someone has decided which outputs are records. A summary kept in the vendor’s environment may not be a written communication; emailed to a client or a colleague, it is. Decide that before the tool is switched on, not after.
Key takeaways
- The gains concentrate in client communication (57% of surveyed advisors) and are distrusted for portfolio work (18% call them reliable), so the capability a firm buys sits on the regulated surface.
- Non-use fell from 33% to 20% in the 2026 Morningstar study while client time held at 53% against a 63% target. Efficiency moved; the thing the firm wanted did not.
- Amended Reg S-P has bound smaller entities since June 3, 2026, with service-provider oversight and 30-day client notice. Personal accounts are an open item.
- The Marketing Rule carries the clearest price tag: $225,000 and $175,000 penalties in the SEC’s March 2024 AI-claim settlements, both about marketing copy.
- IAR CE and CFP Board CE both require approved sponsors, so a working session earns no credit. Buy credit and capability as two line items.
If you want a view on which of your tools sit inside the client-information line and which three workflows are worth converting before year-end reviews, book a free AI readiness call. We will tell you what a session should contain for your seats.
Arthur Wandzel
