The robo-adviser learns to talk back
AI portfolio tools now manage $2.6 trillion. The fee saved often reappears as a subscription charge.
By The Ledger · · 5 min read
Morgan Stanley gave 16,000 financial advisers a generative AI assistant in 2024. By early 2025, Vanguard, Schwab and a wave of standalone apps had followed with tools that don't just rebalance a portfolio but explain, in plain sentences, why. Assets in AI-assisted advisory products passed $2.6 trillion globally in 2024, according to Cerulli Associates. The question the industry is betting nobody asks closely: whether the fee it stopped charging for advice has simply moved somewhere else on the statement.
What actually happened
Robo-advisers have existed since Betterment launched in 2010, charging roughly 0.25% of assets annually to allocate a portfolio by algorithm. What's new is the layer bolted on top: a chat interface, built on large language models, that answers questions a human adviser used to answer. "Should I pay down my mortgage or invest the bonus?" now gets a generated answer in seconds, footnoted with the client's actual account data.
Morgan Stanley's tool, built with OpenAI, drafts meeting notes and pulls research for advisers rather than replacing them. That's the enterprise version. The consumer version is more direct: apps like Cleo and Origin now let users query their entire financial picture — checking, retirement, debt — in natural language, no adviser in the loop at all.
The distinction matters because the fee structure differs sharply on each side of it.
Who pays, who gains
A human financial adviser typically charges 1% of assets under management annually — $1,000 a year on a $100,000 portfolio, per the CFP Board's 2024 fee survey. A traditional robo-adviser charges 0.25%, or $250. The new AI-chat layer on consumer apps runs a flat subscription instead: Cleo charges $5.99 to $14.99 a month, Origin runs $12.99 monthly for full access. That's $72 to $180 a year, regardless of account size.
The arithmetic favors small accounts and penalizes large ones. On a $10,000 portfolio, a flat $150 annual subscription costs 1.5% — more than the human adviser's percentage fee would. On a $500,000 portfolio, the same subscription costs 0.03%, a fraction of what any percentage-based adviser charges. The tools are cheapest exactly where the stakes are highest, and relatively priciest where a beginner can least absorb the cost.
Wealth managers gain differently. Morgan Stanley isn't selling AI to clients directly — it's cutting the time advisers spend on research and paperwork, the firm said in its 2024 disclosures, freeing them to manage more accounts each. Cerulli estimates advisers using AI tools handle 15% to 20% more client relationships without adding headcount. That's a labor cost saved by the firm, not necessarily a fee cut passed to the client.
The mechanism
The tools work by combining two systems that used to sit apart: a rules-based portfolio engine, which has run robo-advice since 2010, and a large language model, which generates the explanatory text. The engine still decides the actual allocation — bonds, equities, rebalancing triggers — using the same mean-variance optimization it always did. The LLM's job is narrower: translate that output into a sentence a non-specialist can read, and answer follow-up questions by retrieving the client's own account data.
This is why the tools can sound like judgment while remaining, underneath, arithmetic. When Origin's assistant says "you're on track for retirement at 62," it isn't reasoning about the client's life. It's running a stored projection formula and phrasing the result conversationally. The model doesn't decide the number. It decides the sentence.
That distinction is also where the risk sits. The Securities and Exchange Commission fined two firms in 2024 — Delphia and Global Predictions — a combined $400,000 for what it called "AI washing," advertising AI-driven advice capabilities the firms hadn't actually built. The SEC's complaint, filed March 2024, found the tools described in marketing didn't match what the underlying software did. The gap between what a chat interface implies and what a portfolio engine computes is exactly where regulators are now looking.
Fiduciary duty complicates the mechanism further. A human adviser managing a retirement account is legally bound, under the Employee Retirement Income Security Act, to act in the client's sole interest. Whether an AI-generated recommendation carries the same legal weight is unresolved. The SEC's 2023 proposed rule on "predictive data analytics" would require firms to eliminate conflicts of interest in algorithm design, but the rule hadn't been finalized as of mid-2025. Until it is, the liability for a bad AI-generated allocation sits in an untested space between the software vendor and the registered adviser who deployed it.
What happens next
Expect the flat-subscription model to face pressure within two years, once enough small accounts calculate the effective percentage they're paying. Apps will likely shift toward tiered pricing — a lower flat fee below some asset threshold, a percentage above it — because the current structure is arithmetically hostile to the beginners it's marketed toward.
Expect the SEC's predictive-analytics rule, or a version of it, to finalize by 2026, given two enforcement actions already on the books. That will force vendors to document exactly what their models optimize for, which will slow deployment but reduce the AI-washing gap.
Watch for a divergence between enterprise and consumer products. Morgan Stanley's model — AI assisting a licensed human who retains fiduciary responsibility — faces less regulatory ambiguity than a standalone app answering allocation questions with no adviser attached. The firms building the latter are the ones a 2026 rule would most directly constrain.
FAQ
Is an AI financial assistant a fiduciary? Not currently, in most consumer apps. Fiduciary status attaches to a registered investment adviser, human or algorithmic, that has formally accepted that duty under the Investment Advisers Act. Many chat-based tools operate as unregistered software, outside that framework — check the app's own disclosures, not its marketing.
Does the AI actually pick the investments? Usually no. The underlying allocation engine — often built years before the chat layer — runs standard portfolio optimization. The AI model generates the conversational explanation and answers follow-up questions, drawing on the engine's output rather than replacing it.
Is a flat subscription cheaper than a percentage fee? Depends entirely on account size. Below roughly $10,000, a $150 annual subscription costs a higher percentage than most robo-advisers charge. Above $200,000, the same subscription is close to free by comparison. Calculate the percentage on your own balance before assuming either model is the better deal.