Handle Client-Facing Use
This is the seventh step in The Adoption Path. Everything before this step could go wrong internally and only cost the firm time to fix. This step covers a narrower, higher-stakes case: AI-assisted output that a client sees, or that shapes what a client is told. The workflows from Build the Workflows already name a review point for these uses; this step is about what that review has to catch specifically, and what the firm tells clients about how the output was produced.
Why it matters
A review point that works fine for an internal draft can miss exactly what matters once a client is on the other end of it. An internal summary with a wrong date costs someone a few minutes to fix; the same wrong date in a letter a client receives is a different kind of problem; a plausible-sounding recommendation nobody actually made is a different kind of problem again. This step exists because client-facing use deserves its own, stricter pass, not just the same review checklist with higher stakes attached.
What actually happens
Distinguish communication from advice
Client-facing use splits into two categories that need different handling. Communication covers letters, emails, summaries, and marketing material — anything that conveys information without itself being a recommendation. Advice covers anything that becomes or supports a recommendation to a client. This reuses the assist, analyze, decide framing from Decide What AI Is For rather than inventing new terms: advice-category uses are exactly where the decide bucket’s rule — AI does not make the recommendation — actually gets enforced, since this is the point where an unreviewed draft would otherwise reach the client as if it were the adviser’s own conclusion.
Know what the review has to specifically catch
The review point named in each workflow has to check three things here that a general internal review does not always cover. Factual accuracy against a source the reviewer actually checked, not just a sentence that sounds plausible. That individualized, client-specific facts are correct — a wrong number in a client-facing document is a different order of problem than the same wrong number in an internal draft, because a client may act on it directly. And that nothing in the output implies a recommendation the adviser has not actually made, which is easy to miss when a draft is well-written and confident-sounding throughout.
Decide what the firm tells clients about AI involvement
Whether, and how, the firm discloses that AI was involved in producing something a client receives is a marketing and advertising question — see the relevant row in Map the Rules That Apply — as much as it is a question about AI itself. This page does not take a position on what a specific disclosure requires or how it should be worded; that is a fact-specific compliance question for the firm’s own marketing materials and disclosure brochure. What this step asks is that the firm actually answer the question and say where the answer lives, rather than treating it as covered by the general AI policy without checking.
Watch for inconsistency across similar clients
Two clients asking a similar question at different times getting inconsistently framed answers is a suitability problem, not just an AI quirk — the duty of care means treating similarly situated clients consistently, and a tool that answers the same question differently on different days can undermine that without any single answer looking wrong on its own. This is easiest to catch at the point output goes out the door, by the reviewer who has seen what similar clients were told before, not after the fact when two clients compare notes.
Set a harder line for the advice category
Regardless of how the firm’s workflows are otherwise structured, set one line that does not bend: nothing in the advice category leaves the firm without the adviser’s own review and sign-off, full stop. No volume-based exception — a busy week is not a reason to skip the review that makes the recommendation the adviser’s own rather than a tool’s.
Questions to ask yourself
- For a given piece of client-facing output, is it communication or advice — and does the review match which one it is?
- Has the reviewer actually checked the source for every fact in the output, or just read it and found it plausible?
- Does the firm have an answer to what it tells clients about AI involvement, and where does that answer live?
- If two similar clients asked the same question a month apart, would they get consistently framed answers?
- Is there any volume-based exception, formal or informal, to the review requirement for advice-category output?
Common mistakes
- Applying the same review standard to client-facing output as internal drafts, when the stakes are different.
- Treating disclosure as solved because the firm has a general AI policy, without checking whether marketing rules require something specific.
- Letting volume pressure erode the review step for advice-category output.
- Assuming consistency across clients is automatically fine because each individual answer looked reasonable.
- Conflating the communication and advice categories when they need different handling.
Which Firm Functions apply
- Advice and Planning Work: The decide-bucket rule that AI does not make the recommendation is enforced right here, at the point advice reaches a client.
- Client Communication and Marketing: The disclosure question and the marketing-rule angle on client-facing output both live in this Firm Function.
- Compliance and Supervision: The stricter review line for client-facing use is a supervision question — who actually enforces it, and how it gets checked.
At a larger firm, client-facing review usually already has a named owner and a required sign-off step built into the firm’s existing process for client communications and advice. The job is to route AI-assisted output through that existing checkpoint rather than treating it as a new category that needs its own review — the checkpoint that already catches a human-drafted letter with a wrong number should be the same one that catches an AI-assisted one.
Next step
Review and Keep Records: Check that the workflows, including this stricter client-facing review, are actually working as intended.