Define The Suitability Evidence Problem
AI can help financial advisers prepare meetings, summarize portfolios, draft follow-ups, and compare planning options. The risk is that a useful assistant begins to look like a recommendation engine before the firm has defined what evidence must support each client-specific suggestion.
A suitability evidence file is the record that connects an AI-assisted output to the client facts, source systems, assumptions, adviser review, and limits behind it. The file does not make the AI the adviser. It makes clear what the AI used, what it did not know, and who accepted or rejected the resulting work.
Why Financial Context Gets Fragile
Anthropic's Claude for Financial Advisors announcement emphasizes connectors to custodians, portfolio platforms, CRMs, planning tools, and meeting intelligence. That breadth is valuable because adviser work is scattered across many systems, but it also means an AI answer can combine data with different freshness, permission, and reliability levels.
Financial advice depends on facts that are easy to miss: household goals, time horizon, tax situation, risk tolerance, concentration, liquidity needs, account restrictions, fees, and prior communications. A polished AI summary can hide whether those facts were current, complete, or appropriate for the question being answered. The evidence file makes those dependencies visible before the output leaves the adviser workstation.
Count The Cost Of Weak Evidence
The immediate cost of weak evidence is adviser rework. A human must chase missing records, correct stale holdings, rewrite a communication, or remove advice-like language from a draft. The deeper cost is supervisory risk when a firm cannot reconstruct why an AI-assisted note or recommendation was reasonable at the time.
A simple estimate can start with AI-assisted client outputs per month, multiplied by the percentage that require compliance review and the average minutes needed to reconstruct support. If 400 outputs require review and 15 percent need twenty minutes of reconstruction, the firm spends twenty hours each month rebuilding evidence that should have been captured at creation.
Diagnose The Evidence Gap
Start by selecting one AI-assisted advisory workflow, such as meeting preparation or portfolio drift explanation. Ask whether the firm can identify every source the AI used, the timestamp of each source, the client facts considered, the assumptions made, the adviser who reviewed the output, and the final version sent or stored.
Warning signs include copied AI text with no citations, summaries that mix client records and general market commentary, no distinction between draft and recommendation, and no field showing why the adviser approved the final message. Those gaps are not just documentation issues. They weaken the firm's ability to prove care.
Choose A File Scope
The evidence file should be scoped to client-impacting outputs, not every internal note. A market research summary may need citations and disclaimers. A client-specific tax-loss harvesting suggestion, rollover discussion, estate summary, or portfolio-change recommendation needs a stronger record because it can influence financial decisions.
The firm can use tiers. Low-risk drafts require source links and adviser review. Medium-risk client summaries require client context, assumptions, and approved language. High-risk recommendations require compliance routing, conflict checks, fee impact, alternatives considered, and a clear statement that the adviser, not the model, made the recommendation. This tiering keeps routine preparation fast while preserving stronger evidence where client reliance is likely.
Build The Suitability Evidence File
Each file should capture the client or household, the requested task, connected systems used, source timestamps, data excluded, assumptions, recommended action if any, alternatives considered, risk and cost factors, reviewer, approval status, and final communication. It should also record whether the AI output was used, edited, rejected, or escalated.
The most important field is the decision boundary. The file should say whether the AI prepared information, suggested language, surfaced an issue, or proposed an action. Without that boundary, a later reviewer may mistake an AI draft for an adviser recommendation or a general note for a client-specific analysis.
Worked Example: Portfolio Drift
Imagine Claude prepares a meeting note showing that a client's portfolio drifted away from the model allocation. The evidence file records holdings from the custodian, model portfolio data, client risk profile, cash needs noted in the CRM, and the date each record was retrieved.
The AI drafts three options: rebalance now, wait until a scheduled review, or use new cash flows to reduce the drift. The adviser edits the note after checking tax exposure and client preferences, then approves a client-ready explanation. The file shows that the final communication came from human judgment supported by documented AI assistance.
Measure Evidence Quality
Useful measures include the share of client-impacting outputs with complete files, average reconstruction time during review, number of outputs rejected for missing evidence, and number of adviser edits caused by stale or incomplete data. The goal is not more paperwork. The goal is less reconstruction and clearer accountability.
Firms should also track whether the file improves client service. If advisers spend less time gathering routine facts and more time discussing tradeoffs, the AI workflow is doing useful work. If compliance exceptions rise, the firm may need narrower tasks, better connectors, or stronger review rules. The point is an adviser workflow that becomes easier to supervise as it becomes faster.
Start With One Client Workflow
Choose one advisory workflow that already has a compliance or supervisory touchpoint. Build the file template around that workflow's real evidence needs instead of trying to define a universal AI policy first.
Run ten past cases through the template and mark which fields are hard to complete. Those missing fields identify the real integration problem. The first practical win is a file that makes one AI-assisted workflow reviewable, teachable, and defensible.
Sources And Methodology
This article uses Anthropic's Claude for Financial Advisors announcement as the news trigger, including its connector and workflow descriptions. It also references the SEC's investment adviser fiduciary-duty interpretation and FINRA's Regulation Best Interest overview for the general importance of care, loyalty, and recommendation support.
The suitability evidence file is SynHy analysis for AI-assisted adviser operations. It is not legal advice, a compliance program, or an assessment of Anthropic's product controls. Registered firms should adapt the file with qualified legal, compliance, supervisory, and technology leaders before relying on it in regulated workflows.