More insights.
Subscribe to our newsletter.
Deep dives into design thinking, creative process, and the intersection of business and aesthetics.
Claude Just Entered Wealth Management. Here's What RIAs Should Actually Change.
Claude for Financial Advisors moves generative AI closer to the center of the RIA technology stack. But RIAs should not respond by giving every employee a new chatbot and hoping efficiency appears. Leadership should redesign one governed workflow first: pre-meeting preparation.
That workflow is repetitive enough to produce measurable savings, important enough to improve the client experience, and controlled enough to preserve human judgment. It also exposes the questions every larger AI initiative eventually has to answer: Which systems contain the truth? Who may access which data? What must be reviewed? What gets retained? Who is accountable when an output is wrong?
The firms that answer those questions will gain more than faster summaries. They will build an operating model for practical AI adoption. The firms that skip them may simply add a sophisticated new layer to an already fragmented business.
What Anthropic actually launched
On September 14, 2026, Anthropic introduced Claude for Financial Advisors, a suite of connectors and workflow skills designed for advisory work. It is not a separate model that suddenly became a financial adviser. It is an industry workflow layer that lets Claude work with information from systems financial professionals already use.
The announced connections include platforms and data from Charles Schwab Advisor Services, Addepar, BlackRock Advisor Center, Envestnet, Tamarac, MoneyGuide, iCapital, Orion, Redtail, SS&C Black Diamond, Wealthbox, Wealth.com, Vanguard, and Zocks, along with existing connections such as Microsoft 365, Salesforce, DocuSign, Box, FactSet, S&P Global, and Morningstar.
The packaged skills cover work including:
Pre-meeting preparation
Post-meeting notes and follow-up
Prospect intake
Portfolio-rebalance review
Estate and tax briefs
Alternative-investment briefs
Adviser onboarding
Compliance review support and AI-policy documentation
The language around control matters. Anthropic states that investment recommendations, client communications, compliance determinations, and other regulated activity remain subject to human review and approval. It also recommends its Enterprise plan for RIAs because it includes audit logs that can support recordkeeping.
This is not “AI replaces the adviser.” It is “AI can now gather, structure, and stage more of the work surrounding the adviser.”
Why this release matters more than another financial chatbot
Until now, many firms have treated AI as an isolated writing tool. An employee opens a chat window, enters a prompt, copies the result, and decides what to do with it. That may save a few minutes, but it leaves the employee responsible for finding the source material, moving information between systems, checking permissions, and creating the record.
Claude for Financial Advisors points toward a different model: AI operating across connected systems through defined skills and approval points.
That changes the leadership question.
The question is no longer only, “Which AI tool should we allow?” It is also:
Which workflows are worth redesigning?
Which system is authoritative when records conflict?
What data may the workflow retrieve?
What action may it stage or complete?
Where must a person review, edit, approve, or stop it?
Which inputs, outputs, approvals, and actions must be retained?
In other words, AI adoption is becoming an operations and governance decision—not an individual productivity experiment.
The first workflow RIAs should redesign: pre-meeting preparation
For most RIAs, the best starting point is the client pre-meeting brief.
Meeting preparation often requires someone to reconstruct a household from several places: the CRM, custodian, portfolio reporting system, financial plan, recent email, prior meeting notes, open service tasks, signed documents, and calendar. The work is valuable, but much of the gathering and formatting is repetitive.
Anthropic's release specifically positions connected pre-meeting preparation as a core use case. More importantly, this workflow has the right implementation profile for a first controlled pilot.
Why meeting prep is the right starting point
It happens frequently. A recurring workflow produces enough repetitions to find defects and calculate whether the change is worthwhile.
The current process can be measured. A firm can record preparation time, missing information, last-minute corrections, and rework before changing anything.
The output remains internal. The brief can stay with the adviser and service team. It does not need to send a message, place a trade, alter a record, or make a recommendation on its own.
The adviser already has a natural review point. A professional must read the brief before the meeting. Human review is part of the work rather than a compliance step awkwardly added later.
Clients can feel the improvement. Better preparation can produce more specific questions, faster resolution of open items, and fewer moments when the client has to repeat information the firm should already know.
It creates a foundation for later workflows. Once the firm has reliable sources, permissions, templates, and approvals for meeting prep, it can consider extending the design into post-meeting notes, CRM tasks, or follow-up drafts.
By contrast, autonomous recommendations, unsupervised client messages, money movement, and trade execution create much higher stakes. They are poor first experiments.
What the redesigned workflow should produce
The goal is not a long AI-generated biography of the client. It is a concise, source-grounded brief an adviser can trust enough to review quickly.
A useful template might include:
Meeting purpose and attendees
Client-stated goals and recent life events
Portfolio or plan changes since the last meeting
Recent deposits, withdrawals, alerts, or unusual activity for review
Open service items and the person responsible
Prior commitments made by the firm or client
Documents or information still missing
Topics that require adviser judgment
Source and date for every material fact
A visible “reviewed by” field before use
Every item does not belong in every firm's brief. The template should reflect the firm's service model, client segments, systems, policies, and meeting standards.
How to redesign the workflow before installing the tool
The technology will not repair an undefined process. If two advisers prepare differently, the CRM is incomplete, and no one owns open tasks, AI can assemble inconsistency faster—but it cannot decide what the firm's standard should be.
1. Define the approved output
Start with the final brief, not the software demonstration.
Ask advisers, operations, client service, compliance, and technology leaders what information should appear, what should never appear, and which items require a citation to the underlying record. Decide how current each data point must be and how uncertainty should be displayed.
A good output definition includes format, required fields, prohibited content, source rules, review status, and escalation conditions.
2. Map sources, permissions, and ownership
For every field in the brief, identify:
The system of record
The data owner
Who may access it
Whether the AI provider may process it under the firm's agreement
How long inputs and outputs are retained
What happens when two systems disagree
Which vendor or internal team supports the connection
Do not confuse a connector's availability with the firm's approval to use it. Technical access, contractual permission, data governance, cybersecurity review, and regulatory responsibility are separate questions.
Data quality will become visible quickly. If the same household has different names, incomplete fields, inconsistent task statuses, or outdated objectives across systems, the pilot may reveal that the first project is partly data cleanup.
3. Put human approval at the decision points
“Human in the loop” is too vague to be a control by itself. Name the person, the decision, and the evidence of approval.
For a pre-meeting workflow, a controlled design could allow Claude to retrieve approved data, assemble a draft, flag gaps, and stage suggested questions. The adviser then checks material facts, corrects the draft, and marks it reviewed. Anything involving advice, a client communication, a transaction, or a compliance conclusion moves to the firm's established approval process.
Permissions should follow the smallest practical scope. A workflow that prepares one adviser's meetings may not need access to the entire firm's client base or the authority to modify source records.
4. Test against completed meetings
Before using the workflow for a live client meeting, run it against a controlled set of past meetings. Compare the generated brief with the records and with what an experienced adviser would have prepared.
Test for:
Incorrect facts
Missing facts
Stale information
Duplicate or conflicting records
Unsupported conclusions
Excessive access
Inconsistent performance across client types
Failure when a connector or source is unavailable
Document the failures. A pilot succeeds when the firm understands the workflow's limits—not when the demonstration looks impressive.
5. Measure the workflow, not the novelty
Set a baseline before the pilot, then compare:
Average preparation time per meeting
Percentage of briefs completed on time
Material corrections required during review
Missing open tasks or documents found
Adviser confidence in the brief
Compliance or security exceptions
Client follow-up time after the meeting
Adoption by the intended team
Time saved matters, but it is not the only result. A faster brief that introduces more errors is not an improvement. A technically accurate brief that advisers ignore is not an adopted workflow.
The practical compliance guardrails
Claude's compliance skill may help flag language, document review activity, and support AI-policy work. It does not transfer regulatory accountability to Anthropic. The firm remains responsible for the way the tool is selected, configured, supervised, and used.
The correct obligations depend on the firm's registration, business model, jurisdiction, data, and use case. The following is a leadership checklist, not a substitute for legal or compliance advice.
Area | Why it matters | Practical design question |
|---|---|---|
Client information and privacy | The SEC's amended Regulation S-P requires covered institutions, including SEC-registered investment advisers, to maintain written safeguards and incident-response policies addressing unauthorized access to or use of customer information. | What client information enters the workflow, who can access it, how is the vendor assessed, and what is the incident-response path? |
Fiduciary duty | The SEC's interpretation states that an investment adviser's fiduciary duty includes duties of care and loyalty across the advisory relationship. | Where could an inaccurate, incomplete, or biased output affect advice, and who checks it before it influences a client decision? |
Marketing and communications | The SEC Marketing Rule applies to advertisements disseminated directly or indirectly by covered advisers and includes related recordkeeping duties. FINRA Rule 2210 standards can apply to public communications by member firms whether content is produced by a person or technology. | Is AI-generated language internal, one-to-one, or promotional—and which review and retention path applies before publication or delivery? |
Books and records | SEC-registered advisers have recordkeeping duties under Rule 204-2. Anthropic recommends Enterprise access for RIAs partly because of audit logs, but a product feature is not a complete records program. | Which prompts, source records, drafts, edits, approvals, and final communications must the firm preserve, where, and for how long? |
Supervision | FINRA Regulatory Notice 24-09 says its technology-neutral rules apply when member firms use GenAI and points to supervision, model risk, privacy, integrity, reliability, and accuracy. | For a FINRA member or hybrid firm, which written supervisory procedures and review responsibilities apply to this specific workflow? |
Testing and monitoring | FINRA's 2026 GenAI regulatory report discusses formal approval, testing, monitoring, prompt and output logs, model-version tracking, and human review. | Who tests the workflow before launch, monitors it afterward, and pauses it when performance changes? |
Third-party and agent risk | A connected agent may retrieve data and stage actions across systems. That creates different exposure from a standalone drafting tool. | What is the tool allowed to read, write, send, or trigger—and how is each action limited and traced? |
Standalone RIAs should not treat FINRA material as if FINRA directly regulates them. The guidance is directly relevant to FINRA member firms and hybrid organizations and can still be useful as a governance reference. State-registered advisers also need to consider the rules and guidance of their state securities regulators.
Useful primary sources include the SEC's Regulation S-P amendments, its investment-adviser fiduciary interpretation, and its Investment Adviser Marketing compliance guide.
What RIAs should not change
The arrival of a wealth-management-specific Claude product does not make every part of advisory work an automation target.
RIAs should not:
Replace adviser judgment with an unreviewed model output
Allow autonomous client communications because the first drafts sound polished
Put confidential client information into consumer AI accounts outside firm policy
Grant broad system access when a narrow permission will do
Treat the vendor's compliance feature as the firm's compliance program
Automate a broken process before deciding what good work looks like
Add “AI-powered” to the brand simply because the firm installed a plugin
Measure success only through hours saved or headcount avoided
Trust is not an inefficiency to remove. The adviser-client relationship, professional accountability, judgment, and difficult conversations remain central to the service.
What this changes about the client experience
Most clients will not care which model prepared an internal brief. They will care whether the firm remembers what matters, follows through, responds promptly, and brings a coordinated view of their financial life to the meeting.
That is where AI infrastructure becomes a brand issue.
A brand is not limited to the logo, website, or marketing campaign. For a service business, the brand is also the expectation created by the firm and the experience that either confirms or contradicts it.
If a firm promises proactive, personal advice but advisers spend meetings searching for context, the operating model weakens the promise. If connected workflows help the team arrive prepared, surface open items, and follow through consistently, the system supports the position the firm wants to own.
The strategic opportunity is not to make every RIA look technologically advanced. As connected AI becomes more common, faster administrative work may become the baseline. Differentiation will come from what the firm does with the recovered capacity: more thoughtful meetings, better coordination, clearer communication, deeper planning, or access for households the firm previously lacked capacity to serve.
The technology can prepare context. Leadership still has to decide what a better client experience should feel like.
What leadership should do in the next 30 days
Week 1: Select the workflow. Choose one repeatable pre-meeting process, one team, and a limited client set. Record the current time, errors, and handoffs.
Week 2: Design the control map. Define the brief, systems of record, permissions, data owners, approval points, retention needs, exception process, and accountable leader. Involve compliance, security, legal, and technology professionals appropriate to the firm.
Week 3: Run a historical test. Use completed meetings and controlled data. Compare results with source records and experienced-adviser expectations. Record every material failure.
Week 4: Decide whether to pilot live. Approve, revise, or stop based on evidence. If approved, launch narrowly, keep the output internal, require named adviser review, and monitor the agreed measures.
Do not begin with a firmwide license announcement. Begin with a workflow that can earn the right to expand.
The bottom line
Claude did not enter wealth management to become the relationship. It entered the layer around the relationship: the research, preparation, documentation, coordination, and follow-up that consume an adviser's finite time.
That is significant. It means AI is moving beyond isolated prompts and into the systems where advisory work happens.
The correct response is not panic, a rushed purchase, or a promise to become “AI-first.” It is disciplined redesign. Start with pre-meeting preparation. Define the output. Control the data. Name the reviewer. Preserve the record. Test the failures. Measure the client and operational result.
Then decide what deserves to come next.
Claude Just Entered Wealth Management. Here's What RIAs Should Actually Change.
Claude for Financial Advisors moves generative AI closer to the center of the RIA technology stack. But RIAs should not respond by giving every employee a new chatbot and hoping efficiency appears. Leadership should redesign one governed workflow first: pre-meeting preparation.
That workflow is repetitive enough to produce measurable savings, important enough to improve the client experience, and controlled enough to preserve human judgment. It also exposes the questions every larger AI initiative eventually has to answer: Which systems contain the truth? Who may access which data? What must be reviewed? What gets retained? Who is accountable when an output is wrong?
The firms that answer those questions will gain more than faster summaries. They will build an operating model for practical AI adoption. The firms that skip them may simply add a sophisticated new layer to an already fragmented business.
What Anthropic actually launched
On September 14, 2026, Anthropic introduced Claude for Financial Advisors, a suite of connectors and workflow skills designed for advisory work. It is not a separate model that suddenly became a financial adviser. It is an industry workflow layer that lets Claude work with information from systems financial professionals already use.
The announced connections include platforms and data from Charles Schwab Advisor Services, Addepar, BlackRock Advisor Center, Envestnet, Tamarac, MoneyGuide, iCapital, Orion, Redtail, SS&C Black Diamond, Wealthbox, Wealth.com, Vanguard, and Zocks, along with existing connections such as Microsoft 365, Salesforce, DocuSign, Box, FactSet, S&P Global, and Morningstar.
The packaged skills cover work including:
Pre-meeting preparation
Post-meeting notes and follow-up
Prospect intake
Portfolio-rebalance review
Estate and tax briefs
Alternative-investment briefs
Adviser onboarding
Compliance review support and AI-policy documentation
The language around control matters. Anthropic states that investment recommendations, client communications, compliance determinations, and other regulated activity remain subject to human review and approval. It also recommends its Enterprise plan for RIAs because it includes audit logs that can support recordkeeping.
This is not “AI replaces the adviser.” It is “AI can now gather, structure, and stage more of the work surrounding the adviser.”
Why this release matters more than another financial chatbot
Until now, many firms have treated AI as an isolated writing tool. An employee opens a chat window, enters a prompt, copies the result, and decides what to do with it. That may save a few minutes, but it leaves the employee responsible for finding the source material, moving information between systems, checking permissions, and creating the record.
Claude for Financial Advisors points toward a different model: AI operating across connected systems through defined skills and approval points.
That changes the leadership question.
The question is no longer only, “Which AI tool should we allow?” It is also:
Which workflows are worth redesigning?
Which system is authoritative when records conflict?
What data may the workflow retrieve?
What action may it stage or complete?
Where must a person review, edit, approve, or stop it?
Which inputs, outputs, approvals, and actions must be retained?
In other words, AI adoption is becoming an operations and governance decision—not an individual productivity experiment.
The first workflow RIAs should redesign: pre-meeting preparation
For most RIAs, the best starting point is the client pre-meeting brief.
Meeting preparation often requires someone to reconstruct a household from several places: the CRM, custodian, portfolio reporting system, financial plan, recent email, prior meeting notes, open service tasks, signed documents, and calendar. The work is valuable, but much of the gathering and formatting is repetitive.
Anthropic's release specifically positions connected pre-meeting preparation as a core use case. More importantly, this workflow has the right implementation profile for a first controlled pilot.
Why meeting prep is the right starting point
It happens frequently. A recurring workflow produces enough repetitions to find defects and calculate whether the change is worthwhile.
The current process can be measured. A firm can record preparation time, missing information, last-minute corrections, and rework before changing anything.
The output remains internal. The brief can stay with the adviser and service team. It does not need to send a message, place a trade, alter a record, or make a recommendation on its own.
The adviser already has a natural review point. A professional must read the brief before the meeting. Human review is part of the work rather than a compliance step awkwardly added later.
Clients can feel the improvement. Better preparation can produce more specific questions, faster resolution of open items, and fewer moments when the client has to repeat information the firm should already know.
It creates a foundation for later workflows. Once the firm has reliable sources, permissions, templates, and approvals for meeting prep, it can consider extending the design into post-meeting notes, CRM tasks, or follow-up drafts.
By contrast, autonomous recommendations, unsupervised client messages, money movement, and trade execution create much higher stakes. They are poor first experiments.
What the redesigned workflow should produce
The goal is not a long AI-generated biography of the client. It is a concise, source-grounded brief an adviser can trust enough to review quickly.
A useful template might include:
Meeting purpose and attendees
Client-stated goals and recent life events
Portfolio or plan changes since the last meeting
Recent deposits, withdrawals, alerts, or unusual activity for review
Open service items and the person responsible
Prior commitments made by the firm or client
Documents or information still missing
Topics that require adviser judgment
Source and date for every material fact
A visible “reviewed by” field before use
Every item does not belong in every firm's brief. The template should reflect the firm's service model, client segments, systems, policies, and meeting standards.
How to redesign the workflow before installing the tool
The technology will not repair an undefined process. If two advisers prepare differently, the CRM is incomplete, and no one owns open tasks, AI can assemble inconsistency faster—but it cannot decide what the firm's standard should be.
1. Define the approved output
Start with the final brief, not the software demonstration.
Ask advisers, operations, client service, compliance, and technology leaders what information should appear, what should never appear, and which items require a citation to the underlying record. Decide how current each data point must be and how uncertainty should be displayed.
A good output definition includes format, required fields, prohibited content, source rules, review status, and escalation conditions.
2. Map sources, permissions, and ownership
For every field in the brief, identify:
The system of record
The data owner
Who may access it
Whether the AI provider may process it under the firm's agreement
How long inputs and outputs are retained
What happens when two systems disagree
Which vendor or internal team supports the connection
Do not confuse a connector's availability with the firm's approval to use it. Technical access, contractual permission, data governance, cybersecurity review, and regulatory responsibility are separate questions.
Data quality will become visible quickly. If the same household has different names, incomplete fields, inconsistent task statuses, or outdated objectives across systems, the pilot may reveal that the first project is partly data cleanup.
3. Put human approval at the decision points
“Human in the loop” is too vague to be a control by itself. Name the person, the decision, and the evidence of approval.
For a pre-meeting workflow, a controlled design could allow Claude to retrieve approved data, assemble a draft, flag gaps, and stage suggested questions. The adviser then checks material facts, corrects the draft, and marks it reviewed. Anything involving advice, a client communication, a transaction, or a compliance conclusion moves to the firm's established approval process.
Permissions should follow the smallest practical scope. A workflow that prepares one adviser's meetings may not need access to the entire firm's client base or the authority to modify source records.
4. Test against completed meetings
Before using the workflow for a live client meeting, run it against a controlled set of past meetings. Compare the generated brief with the records and with what an experienced adviser would have prepared.
Test for:
Incorrect facts
Missing facts
Stale information
Duplicate or conflicting records
Unsupported conclusions
Excessive access
Inconsistent performance across client types
Failure when a connector or source is unavailable
Document the failures. A pilot succeeds when the firm understands the workflow's limits—not when the demonstration looks impressive.
5. Measure the workflow, not the novelty
Set a baseline before the pilot, then compare:
Average preparation time per meeting
Percentage of briefs completed on time
Material corrections required during review
Missing open tasks or documents found
Adviser confidence in the brief
Compliance or security exceptions
Client follow-up time after the meeting
Adoption by the intended team
Time saved matters, but it is not the only result. A faster brief that introduces more errors is not an improvement. A technically accurate brief that advisers ignore is not an adopted workflow.
The practical compliance guardrails
Claude's compliance skill may help flag language, document review activity, and support AI-policy work. It does not transfer regulatory accountability to Anthropic. The firm remains responsible for the way the tool is selected, configured, supervised, and used.
The correct obligations depend on the firm's registration, business model, jurisdiction, data, and use case. The following is a leadership checklist, not a substitute for legal or compliance advice.
Area | Why it matters | Practical design question |
|---|---|---|
Client information and privacy | The SEC's amended Regulation S-P requires covered institutions, including SEC-registered investment advisers, to maintain written safeguards and incident-response policies addressing unauthorized access to or use of customer information. | What client information enters the workflow, who can access it, how is the vendor assessed, and what is the incident-response path? |
Fiduciary duty | The SEC's interpretation states that an investment adviser's fiduciary duty includes duties of care and loyalty across the advisory relationship. | Where could an inaccurate, incomplete, or biased output affect advice, and who checks it before it influences a client decision? |
Marketing and communications | The SEC Marketing Rule applies to advertisements disseminated directly or indirectly by covered advisers and includes related recordkeeping duties. FINRA Rule 2210 standards can apply to public communications by member firms whether content is produced by a person or technology. | Is AI-generated language internal, one-to-one, or promotional—and which review and retention path applies before publication or delivery? |
Books and records | SEC-registered advisers have recordkeeping duties under Rule 204-2. Anthropic recommends Enterprise access for RIAs partly because of audit logs, but a product feature is not a complete records program. | Which prompts, source records, drafts, edits, approvals, and final communications must the firm preserve, where, and for how long? |
Supervision | FINRA Regulatory Notice 24-09 says its technology-neutral rules apply when member firms use GenAI and points to supervision, model risk, privacy, integrity, reliability, and accuracy. | For a FINRA member or hybrid firm, which written supervisory procedures and review responsibilities apply to this specific workflow? |
Testing and monitoring | FINRA's 2026 GenAI regulatory report discusses formal approval, testing, monitoring, prompt and output logs, model-version tracking, and human review. | Who tests the workflow before launch, monitors it afterward, and pauses it when performance changes? |
Third-party and agent risk | A connected agent may retrieve data and stage actions across systems. That creates different exposure from a standalone drafting tool. | What is the tool allowed to read, write, send, or trigger—and how is each action limited and traced? |
Standalone RIAs should not treat FINRA material as if FINRA directly regulates them. The guidance is directly relevant to FINRA member firms and hybrid organizations and can still be useful as a governance reference. State-registered advisers also need to consider the rules and guidance of their state securities regulators.
Useful primary sources include the SEC's Regulation S-P amendments, its investment-adviser fiduciary interpretation, and its Investment Adviser Marketing compliance guide.
What RIAs should not change
The arrival of a wealth-management-specific Claude product does not make every part of advisory work an automation target.
RIAs should not:
Replace adviser judgment with an unreviewed model output
Allow autonomous client communications because the first drafts sound polished
Put confidential client information into consumer AI accounts outside firm policy
Grant broad system access when a narrow permission will do
Treat the vendor's compliance feature as the firm's compliance program
Automate a broken process before deciding what good work looks like
Add “AI-powered” to the brand simply because the firm installed a plugin
Measure success only through hours saved or headcount avoided
Trust is not an inefficiency to remove. The adviser-client relationship, professional accountability, judgment, and difficult conversations remain central to the service.
What this changes about the client experience
Most clients will not care which model prepared an internal brief. They will care whether the firm remembers what matters, follows through, responds promptly, and brings a coordinated view of their financial life to the meeting.
That is where AI infrastructure becomes a brand issue.
A brand is not limited to the logo, website, or marketing campaign. For a service business, the brand is also the expectation created by the firm and the experience that either confirms or contradicts it.
If a firm promises proactive, personal advice but advisers spend meetings searching for context, the operating model weakens the promise. If connected workflows help the team arrive prepared, surface open items, and follow through consistently, the system supports the position the firm wants to own.
The strategic opportunity is not to make every RIA look technologically advanced. As connected AI becomes more common, faster administrative work may become the baseline. Differentiation will come from what the firm does with the recovered capacity: more thoughtful meetings, better coordination, clearer communication, deeper planning, or access for households the firm previously lacked capacity to serve.
The technology can prepare context. Leadership still has to decide what a better client experience should feel like.
What leadership should do in the next 30 days
Week 1: Select the workflow. Choose one repeatable pre-meeting process, one team, and a limited client set. Record the current time, errors, and handoffs.
Week 2: Design the control map. Define the brief, systems of record, permissions, data owners, approval points, retention needs, exception process, and accountable leader. Involve compliance, security, legal, and technology professionals appropriate to the firm.
Week 3: Run a historical test. Use completed meetings and controlled data. Compare results with source records and experienced-adviser expectations. Record every material failure.
Week 4: Decide whether to pilot live. Approve, revise, or stop based on evidence. If approved, launch narrowly, keep the output internal, require named adviser review, and monitor the agreed measures.
Do not begin with a firmwide license announcement. Begin with a workflow that can earn the right to expand.
The bottom line
Claude did not enter wealth management to become the relationship. It entered the layer around the relationship: the research, preparation, documentation, coordination, and follow-up that consume an adviser's finite time.
That is significant. It means AI is moving beyond isolated prompts and into the systems where advisory work happens.
The correct response is not panic, a rushed purchase, or a promise to become “AI-first.” It is disciplined redesign. Start with pre-meeting preparation. Define the output. Control the data. Name the reviewer. Preserve the record. Test the failures. Measure the client and operational result.
Then decide what deserves to come next.