Skip to Content
Firm FunctionsAdvice and Planning

Advice and Planning Work

This is the work clients actually pay for: gathering facts about their situation, analyzing options, and recommending a course of action. AI can speed up pieces of that work, but the advice itself stays the adviser’s, in a way that shapes what AI can and cannot safely be used for here.

Key ideas

Where AI genuinely helps

AI tools are good at the parts of advice work that are mechanical once the facts are in: pulling a first-draft summary out of a long document, organizing scattered notes into a plan outline, running a scenario a client asked about (“what if I retire two years earlier”), or turning an internal analysis into plain language a client can read. Used this way, AI is doing what a capable assistant would do — producing a draft that a person still checks.

Where it should not be the final word

The recommendation itself — what this client should actually do — is a judgment call that draws on the adviser’s read of the client’s whole situation, much of which never makes it into a prompt. An AI tool has no knowledge of the conversation last month where the client mentioned a job change they had not decided on, or the tone in which they described their risk tolerance. Treating AI output as a finished recommendation, rather than an input the adviser still weighs, is the central risk in this area.

Fiduciary duty does not delegate

An adviser’s fiduciary duty to act in the client’s best interest belongs to the adviser, not to any tool used along the way. Using AI to help produce a plan does not shift or share that responsibility — if the advice is wrong, “the AI produced it” is not a defense, in the same way that a spreadsheet error would not be. This is the main reason a person has to review AI-assisted analysis before it becomes a recommendation, not an optional nicety.

Hallucinations show up as confident, specific mistakes

A hallucination in advice work rarely looks like an obvious error. It looks like a specific, plausible number — a contribution limit, a tax rate, a historical return — that is simply wrong, attached to a plan that otherwise reads well. Because these mistakes are confident and specific, they are easy to miss on a quick read and easy to pass along to a client without noticing. Anything an AI tool states as a fact needs to be checked against a real source before it appears in front of a client.

Consistency across similar clients

Two clients in similar situations should get advice that reflects the same reasoning, even if the words differ. AI tools can introduce inconsistency here — the same question, asked slightly differently or on a different day, can produce a materially different answer. That is a reasonable way to get a first draft and a poor way to make the actual recommendation, since it can quietly produce different outcomes for clients who should be treated alike.

Scope creep from planning into decisions

A tool that starts by summarizing a client’s documents can drift, over time, into effectively deciding the recommendation, especially if the adviser gets used to accepting its draft language with only light edits. Firms that watch for this drift tend to build a habit of asking, for each use, whether the tool is still producing a draft the adviser meaningfully reviews or has quietly become the source of the advice itself. This tends to happen gradually rather than as one clear decision — an adviser who started by asking a tool to reword a paragraph ends up, months later, asking it to work out what the paragraph should say, without any single moment where that shift was decided on. Naming the line in advance, before the tool is in daily use, makes it easier to notice when a workflow has crossed it.

Where it shows up in the process

  • Decide What AI Is For: Advice and planning tasks are usually the first candidates a firm considers, and the ones that most need the who-sees-the-output sort described there.
  • Vet and Choose Tools: A tool used in analysis or planning needs to be evaluated for accuracy and traceability, not just convenience.
  • Handle Client-Facing Use: Once a draft produced with AI assistance is heading to a client, it falls under the extra scrutiny that step describes.
  • Review and Keep Records: Ongoing spot-checks of AI-assisted advice belong in the same review cycle as any other part of the advice process.

Common misconceptions

  • “The AI checked the numbers” is not the same as a person checking the numbers; an AI tool can state a wrong figure with exactly the same confidence as a right one.
  • Better output does not mean less review. A more fluent, well-organized draft is often harder to fact-check carefully, not less in need of it, because it reads as already finished.
  • This is not only a risk for firms that let AI write directly to clients. A recommendation shaped by an unreviewed AI-generated analysis carries the same risk even if a person retypes it in their own words before sending it.
  • Disclaiming AI use to the client does not substitute for reviewing the advice; a disclosure explains what happened, it does not fix a wrong recommendation.

At a larger firm, advice and planning work is more likely to run through a shared model, template, or planning platform rather than an individual adviser’s own prompts, which changes where review has to happen. Instead of each adviser checking their own AI-assisted drafts, the firm typically needs a documented review step built into the workflow itself — someone specifically responsible for confirming that AI-assisted output meets the same bar as work produced without it, especially where many advisers rely on the same underlying tool or template.

Go deeper