Financial Advisor Magazine's chatbot test finds plans close to human advisors for mass-affluent clients

For two years the reassurance from conference stages has been that artificial intelligence will thin the middle office, sharpen the research and leave the advisor's chair exactly where it sits. Writing in Financial Advisor Magazine this month, Steve Garmhausen turns that sentence over and asks what happens if the software leaves the advisor in the chair and takes the clients instead.

Garmhausen, who has spent two decades interviewing advisors about investing, planning and the rest of their businesses, did not leave the question as a thought experiment: earlier this year he fed hypothetical client profiles to ChatGPT, Claude and Google Gemini, then asked the follow-up questions a real household would ask once a first answer arrived. He is not an advisor himself; his conclusion is that the chatbot planners sound very much like their human counterparts and that each produced a solid plan for the mass-affluent family he had sketched. The excerpt available to this publication ends mid-sentence on his one apparent reservation, describing the chatbots as 'a bit malleabl', so whatever caveat follows is not in the material here.

The shift has been quiet rather than dramatic: when ChatGPT arrived in 2022 as the first widely available generative chatbot, the industry waved off the threat by pointing at hallucinations and inaccuracies, but by the magazine's account the same tools now produce work that a veteran of thousands of advisor interviews describes as close to the professional version. Aditi Kapadia's own use is the more revealing data point because it comes from someone who ran a software platform built to automate advice and still leans on the new tools for her daily work.

Kapadia led Bank of Montreal's robo-advisor platform and is now chief wealth strategist at the Chicago advisory firm Wealth IQ. She told the magazine she uses the tools for much of her work and double-checks her calculations against them, and her assessment is that the technology will disrupt every white-collar business, though it will be up to the community to adapt.

The résumé line carries weight. Robo advice was the industry's last sustained effort to reach smaller households with software instead of a person, likely the same households the chatbots handle best today, and what changed between the two waves is less the target than the output, which on Garmhausen's reading and Kapadia's is now closer to the professional version than a discount copy of it.

Garmhausen drew a boundary worth reading twice: he built the test around an upper-middle-class household on the reasoning that the wealthiest Americans are not in play, because their money buys private investments, concierge service and an exclusivity no model produces. The pressure therefore collects where the advice is most standardized, among households paying 1% of assets under management, a charge the chatbots do not levy.

The difficulty shows up first as a pricing question rather than a hiring one. A fee stated as a percentage of assets is straightforward to defend while the alternative is nothing; it grows harder to defend when the alternative returns a written plan overnight and the client can lay the two documents side by side. Whether a plan is the product or merely the opening of a relationship is now something a household can decide without asking an advisor's opinion.

Schwab's response has been to put a model inside the advisor's kit: the custodian recently added Claude for Financial Advisors, a tool from Anthropic, and the framing around it positions the software as part of the advisor's weaponry and a help with workflows. As this publication argued in September, the exclusive buys queue position, not a better model, and what that seat is worth depends on who writes the plumbing beneath it.

The durable half of the job sits behind the plan: the governed client record, the permissioned and auditable version of what a household owns and what it has been told, and the confirmation that follows it, meaning the signature, the suitability file and the person who answers when markets gap. A model can draft the first document, but someone still has to stand behind it. The model layer is commoditizing while value collects with whoever owns a governed copy of the client record, and the money in this category accrues to whoever owns the connector rather than the model.

None of this sustains the strongest version of the bear case: Garmhausen does not argue that advisors lose their seats, and by his own account the households with the most complicated balance sheets sit outside the chatbots' reach, while his method is one journalist's judgment rather than a controlled study and a single round of prompts cannot show how a chatbot handles a client panicking through a drawdown. Kapadia's call for the community to adapt leaves the shape of that adaptation open, and the logic points upmarket, toward the clients Garmhausen set aside.

Platforms, custodians and fund managers are building the on-ramp to private funds for exactly that household, and the migration carries its own arithmetic: the further a firm moves upmarket, the more of its economics rest on access and service no model generates.

What adaptation means in practice is where the argument gets thin, because the clients most exposed are the ones Garmhausen built his test around, paying a percentage of assets for a document a chatbot now drafts for nothing and holding a relationship whose contents they can finally compare against a free alternative. The industry has a ready answer about the parts of the job software cannot do, but whether any firm puts a separate price on those parts, rather than bundling them into a percentage of assets, is the thing to watch.