Quick verdict: If you want an AI system to place trades on your behalf without asking first, you need discretionary AI investment advice — and that requires a written grant of trading authority, a fiduciary relationship, and a qualified custodian holding your assets. If you want an AI tool that only suggests moves while you click the final button, you’re using a non-discretionary tool, which usually carries a lighter regulatory duty, a lower price tag, and zero ability to act while you’re asleep, at work, or simply not paying attention. Neither is “better” in the abstract. The right pick depends on how much you trust an algorithm to act on your account without a second opinion, and how much you’re willing to pay for that convenience.
The one-sentence answer
Discretionary AI advice trades your account without asking each time; non-discretionary AI advice only recommends and waits for you to click “confirm” — the difference is a legal one, not a technology one, and it changes who’s liable when the trade goes wrong.
Every robo-advisor pitch and every AI chat assistant bolted onto a brokerage app in 2026 sounds roughly the same: personalized, data-driven, always watching your portfolio. What almost none of them say plainly, in the first screen of onboarding, is which authority model you’re agreeing to. That’s not an accident. Discretionary authority is a legal grant, and grants require disclosure, which slows down a signup flow. Non-discretionary tools skip that step because, legally, they don’t need it — they’re not making decisions for you, only decorating your decision with data.
This matters more than it sounds like it should, because the two models sit on opposite sides of a regulatory line that predates AI by roughly eighty-five years. The Investment Advisers Act of 1940 already distinguished between advisers who exercise discretion over client accounts and those who merely furnish advice. FINRA’s guidance on artificial intelligence, issued as Regulatory Notice 24-09 in June 2024, reaffirmed that this distinction doesn’t dissolve just because a large language model is doing the talking. An AI wrapper doesn’t get a new rulebook. It inherits the old one, and the old one asks a single blunt question: who pulled the trigger on the trade?
Discretionary AI Advice and Non-Discretionary AI Advice, Side by Side
Picture two people opening AI-driven investing accounts on the same afternoon with the same $40,000 and roughly the same goals. One signs a discretionary investment management agreement with a robo-advisor. The other opens a self-directed brokerage account and turns on an AI research assistant that flags opportunities and risks.
The first investor’s account will rebalance itself the moment it drifts from target, harvest tax losses the day the IRS wash-sale window allows it, and shift allocations if the platform’s risk model calls for a change — all without a phone call, a push notification requiring approval, or a signature. The second investor gets a running stream of suggestions: “Your tech weighting is 9 points above your stated target,” or “This fund carries a 0.68% expense ratio; a comparable index fund charges 0.03%.” Nothing moves until that investor logs in and executes the trade personally.
Both experiences are marketed as “AI investment advice.” Only one of them is legally an act of discretion. The other is closer to a smart calculator with excellent timing.
What “Discretionary” Actually Means in Writing
Discretionary authority isn’t implied by convenience or by how confident the platform’s marketing copy sounds. It’s a specific legal grant, typically documented as a limited power of attorney embedded in the advisory agreement, that lets the manager (human, algorithmic, or some blend) buy and sell securities in the client’s account without seeking consent transaction by transaction. Once that grant exists, the manager — and by extension, the AI system executing on its behalf — owes the client a fiduciary duty under the Investment Advisers Act: a continuing obligation to act in the client’s best interest, monitor the account, and disclose conflicts.
What “Non-Discretionary” Actually Means in Writing
A non-discretionary relationship keeps the final decision with the account holder. The AI can rank, forecast, flag, and nudge, but every order still needs a human click. This is the model behind most in-app “AI insights” panels bolted onto brokerages like Merrill, Morgan Stanley’s client-facing assistant tools, or the growing wave of AI chat features layered onto retail platforms. It’s also the model behind general-purpose AI chatbots that answer finance questions — those tools are explicitly not registered investment advisers and carry no fiduciary duty at all, a distinction many users don’t clock until something goes wrong.
Who Holds the Trading Authority — the Legal Line That Decides Everything Else
Trading authority is the single fact that determines almost every other difference on this page: fees, custody obligations, liability, and even how fast the account can react to a falling market. Get this piece right and the rest of the comparison mostly falls into place on its own.
A platform obtains discretionary authority through an explicit, signed grant — usually a checkbox-and-signature step buried in account opening, formalized as a limited power of attorney. Once granted, the firm (and its algorithm) can place trades without asking. State securities law and the Advisers Act both treat this authority as a serious responsibility: firms with discretionary authority must maintain books and records showing every trade was suitable, must supervise the algorithm’s behavior, and must be ready to explain, trade by trade, why the system did what it did.
A non-discretionary setup never crosses that threshold. The AI can generate an order ticket, pre-fill it, even attach a countdown timer urging you to act before a price moves — but it cannot submit that order itself. This is why so many AI investing chat tools stop just short of a “buy” button and instead deep-link you into your brokerage’s order screen. That extra click isn’t a UX choice. It’s the legal firewall between “advice” and “discretionary management.”
Who presses “execute”?
The 6% sliver reflects order-routing convenience features (one-click confirm) — the decision itself still sits with the account holder.
Fiduciary Duty Versus Suitability: The Standard Each Model Answers To
Discretionary AI managers are held to the fiduciary standard, the highest duty of care in U.S. financial regulation. That means the platform must act in the client’s best interest at all times, not merely recommend something “suitable” once and walk away. Fiduciary duty is ongoing: it covers the initial allocation, every subsequent rebalance, every tax-loss harvest, and every moment the account sits untouched while market conditions change around it. If a discretionary AI model quietly drifts a retiree’s portfolio into an aggressive allocation because its optimization function chased returns, that’s a breach the firm has to answer for — the algorithm’s autonomy doesn’t shield the firm from responsibility.
Non-discretionary AI tools generally sit under a lighter standard, and sometimes under no advisory standard at all. If the tool is offered by a broker-dealer, it may fall under Regulation Best Interest, which requires recommendations to be in the customer’s best interest at the time they’re made but doesn’t impose the same continuous monitoring duty a fiduciary carries. If the tool is a general AI chatbot with a finance skin, it may not be regulated as investment advice in any formal sense — the platform’s terms of service will usually say so directly, in language most users scroll past.
This gap explains a pattern that trips up a lot of otherwise careful investors: they assume that because a chatbot sounds authoritative and cites real numbers, it must carry the same accountability as a licensed advisor. It doesn’t, unless the entity behind it is a registered investment adviser or broker-dealer operating under one of those frameworks — and even then, the level of duty depends entirely on whether discretion was granted.
Where the SEC’s Attention Has Landed
Regulators have been circling this exact seam. The SEC proposed a rule in 2023 aimed at conflicts arising from predictive data analytics used in investor interactions — the idea being that an AI system optimizing for engagement or platform revenue could subtly steer non-discretionary “suggestions” in the firm’s favor rather than the client’s, without ever crossing into formal discretionary territory. That proposal drew heavy industry pushback over its breadth and has not moved forward as adopted rule text, but the underlying concern hasn’t gone away: a tool that never touches your trade button can still shape which trades you’re likely to click.
Cost Structure: Asset-Based Fees Versus Subscriptions
Discretionary AI management is priced the way discretionary human management always has been: as a percentage of assets under management. Typical figures across mainstream robo-advisors run from roughly 0.25% to 0.50% annually, sometimes bundled with a flat monthly fee for a “premium” human-plus-AI hybrid tier. The fee scales with your balance, which means the dollar cost rises even though the service — daily rebalancing, tax-loss harvesting — doesn’t get meaningfully more expensive to deliver as your account grows.
Non-discretionary AI tools tend to price like software, not like asset management, because legally that’s closer to what they are. Expect flat monthly subscriptions in the range of commonly seen consumer fintech pricing, or bundling into a brokerage’s existing commission-free structure at no separate charge, since the platform’s incentive is order flow and account retention rather than an AUM cut. This creates a crossover point worth knowing before you pick a lane.
Effective annual cost as a share of portfolio value
$10,000 portfolio
$50,000 portfolio
$150,000 portfolio
$500,000 portfolio
Illustrative figures using a flat 0.30% discretionary AUM fee and a flat $180/year non-discretionary subscription. Actual pricing varies by platform and account size.
Read the chart plainly: on a small account, a flat subscription can cost more, proportionally, than an AUM fee. On a large account, the AUM fee compounds into real money — 0.30% on $500,000 is $1,500 every year, forever, regardless of how much actual work the algorithm does that year. Below roughly $60,000 in this example, the subscription model is cheaper as a percentage of assets; above it, the AUM model quietly becomes the pricier choice unless the discretionary features (automatic harvesting, drift correction) generate enough extra after-tax return to offset the difference.
Custody, Execution Speed, and What Happens During a Sharp Drop
Discretionary authority triggers a custody obligation. Under the SEC’s Custody Rule, a firm with discretionary trading power and access to client assets generally must hold those assets with a qualified custodian and undergo periodic surprise examinations or audits, precisely because the firm can move money without asking. This is regulatory infrastructure built for a world of human portfolio managers, and it applies whether the trading decisions come from a person or from a model running on a server.
Non-discretionary tools mostly sidestep this obligation because they never touch the assets. The AI layer sits on top of a brokerage relationship that already has its own custody arrangement; the tool itself doesn’t need one, because it can’t move a dollar without you.
The practical difference shows up hardest in a fast market. When an index drops 6% in a single session, a discretionary AI account can rebalance, harvest losses, or reduce equity exposure within the trading day, following rules set in advance, with no human bottleneck. A non-discretionary account with the same drop generates an alert — “Your portfolio is down 6.1%; consider rebalancing” — and then waits. If the account holder is asleep, in a meeting, or simply frozen by the number on the screen, nothing happens until they act. That gap between signal and action is the entire value proposition of discretionary management, and it’s also exactly the point where things can go wrong if the underlying model is poorly calibrated, because there’s no human pause between the algorithm’s decision and the money moving.
Head-to-Head Comparison Table
| Criterion | Discretionary AI Advice | Non-Discretionary AI Advice |
|---|---|---|
| Who places the trade | The platform’s algorithm, under signed trading authority | The account holder, always |
| Governing legal standard | Fiduciary duty (Investment Advisers Act of 1940) | Suitability / Reg BI, or no formal advisory standard at all |
| Custody requirement | Qualified custodian, subject to SEC Custody Rule oversight | None — tool never holds or moves assets directly |
| Typical fee model | 0.25%–0.50% of assets under management, annually | Flat subscription, or bundled free with a brokerage account |
| Reaction speed to a market move | Same trading session, no human step required | Depends entirely on when the human reads the alert and acts |
| Automatic tax-loss harvesting | Yes, executed directly when conditions are met | Flagged as a suggestion; human must execute the swap |
| Liability if the model errs badly | Firm bears fiduciary liability for the trade itself | Firm’s exposure is narrower; the human made the final call |
| Best suited to | Hands-off investors, behavioral panic-sellers, drift-prone accounts | Control-focused investors, complex outside holdings, cost-sensitive accounts |
| Common examples | Standard robo-advisor managed accounts (Betterment-style, Wealthfront-style) | In-app AI insight panels, general finance chatbots, screening tools |
Worked Example: Two $75,000 Accounts, One Bad Week in the Market
Say Priya and Daniel each open an account with $75,000 in January. Priya signs up for a discretionary AI-managed portfolio charging 0.30% annually; her target allocation is 70% equities, 30% bonds, and she never touches the app after setup. Daniel opens a self-directed brokerage account and turns on a non-discretionary AI insight tool that costs $15 a month; he sets the same 70/30 target manually and checks in when he gets a notification.
In March, equities drop sharply over four trading days, and Priya’s allocation drifts to roughly 63% equities as bond prices hold steadier. Her discretionary system detects the drift past its rebalancing threshold on day three and sells a slice of bonds to buy equities at depressed prices, restoring the 70/30 target without her doing anything. It also identifies a security trading below its cost basis and executes a tax-loss swap the same afternoon, banking a harvestable loss for her year-end tax bill.
Daniel’s tool sends him the same signal — “portfolio has drifted to 63/37; consider rebalancing,” plus a note about a tax-loss opportunity — but Daniel is traveling for work and doesn’t open the app for six days. By the time he logs in, the market has partly recovered, the rebalancing opportunity has narrowed, and the specific tax-loss window on that security has closed because its price bounced back above his cost basis. He still executes a rebalance, just a less advantageous one, at his own pace and his own judgment, paying no AUM fee for the privilege.
Over one year, Priya pays $225 in advisory fees on her $75,000 account and likely captures a modest tax-alpha benefit from harvesting she’d otherwise have missed. Daniel pays $180 in subscription fees and keeps full control, but he also personally absorbs every timing decision, including the ones he’s too busy or too anxious to make well. Neither outcome is automatically better — it depends whether Daniel’s personal discipline beats the algorithm’s consistency, which is really a question about Daniel, not about the technology.
When Discretionary AI Wins, and When Non-Discretionary Wins
Discretionary AI advice tends to win for:
- Investors who know they panic-sell during drops and want the decision taken out of their hands entirely
- Busy professionals who won’t reliably check an app within days of an alert
- Anyone chasing tax-loss harvesting gains that require same-day or next-day execution to matter
- Retirement accounts where steady, rules-based rebalancing outperforms emotional timing over a multi-decade horizon
- People who have already vetted the platform’s Form ADV, understand the fee, and are comfortable with a fiduciary relationship
Non-discretionary AI advice tends to win for:
- Investors who hold outside assets (real estate, employer stock, a separate brokerage account) that the AI can’t see and that materially change what “correct” allocation even means
- People under specific trading restrictions — corporate insiders, compliance-flagged employees — who are legally required to control every trade personally
- Cost-sensitive investors with mid-size balances where a flat subscription beats a percentage-of-assets fee
- DIY investors who want research and pattern-spotting but genuinely distrust handing over execution authority to any third party, human or algorithmic
- Anyone still building conviction about a platform, who wants a trial period of “advice only” before ever granting trading authority
Common Mistakes People Make Choosing Between the Two
The most frequent error is assuming every robo-advisor is automatically discretionary just because it sounds automated. Some platforms market themselves as “AI-powered guidance” while structurally operating as non-discretionary recommendation engines — the distinction is buried in the account agreement, not the marketing page, and it’s worth five minutes to confirm before you assume trades are happening on your behalf.
A second common mistake runs the other way: treating a general-purpose AI chatbot’s financial commentary as if it carries the same accountability as a licensed fiduciary. It doesn’t. If the entity answering your question isn’t a registered investment adviser or broker-dealer, there’s no suitability review, no fiduciary duty, and typically no regulatory recourse if the suggestion turns out badly. That’s not necessarily a reason to avoid the tool — it’s a reason to treat its output as one input among several, not as advice you’re entitled to rely on.
People also underweight custody and insurance mechanics. Discretionary accounts sit with a qualified custodian and typically carry SIPC coverage against firm failure (not against market losses); non-discretionary tools riding on top of an existing brokerage inherit whatever protections that brokerage already offers. Confusing “the AI manages my money” with “the AI holds my money” leads to real misunderstandings about what happens if a platform goes under.
Finally, a lot of investors never check whether a discretionary platform’s rebalancing thresholds and tax-loss harvesting rules match their actual tax situation — a high earner in a high tax bracket benefits far more from aggressive harvesting than someone with modest taxable income, and paying an AUM fee for a feature that barely moves your tax bill is a quiet way to underperform a cheaper, simpler option.
Practical Checklist Before You Choose
- Read the account agreement’s authority section and confirm, in writing, whether the platform holds discretionary trading power over your account
- Check whether the firm is a registered investment adviser (fiduciary duty) or operates purely as a broker-dealer feature (Reg BI or lighter)
- Confirm where your assets are actually custodied, and whether that custodian is separate from the AI platform itself
- Calculate your all-in annual cost at your actual balance, not the marketing example — run the AUM percentage against a flat subscription for your specific number
- Ask how fast the system can react to a sharp market move, and whether that speed matches how much control you actually want to give up
- Verify whether the platform can see your full financial picture, including outside accounts, before trusting its rebalancing logic completely
- Decide in advance how you’ll handle a scenario where the algorithm’s decision looks wrong in the moment — discretionary accounts don’t pause for your gut feeling
If you’re weighing this decision inside a broader robo-advisor search, our guide to AI-powered wealth management platforms breaks down how several mainstream discretionary providers structure their fees and features, which is a useful next stop once you’ve settled the authority question here.
Key Takeaways
- The discretionary/non-discretionary split is a legal distinction about who holds trading authority, not a description of how smart or advanced the underlying AI is
- Discretionary AI advisors owe you a continuous fiduciary duty and typically charge an asset-based fee; non-discretionary tools usually answer to a lighter standard, or none, and typically charge a flat subscription or nothing at all
- Speed of execution is the practical payoff of discretion — a discretionary account can react to a market move the same day, while a non-discretionary tool waits on you
- Cost crossover depends on your balance: subscriptions tend to win on smaller accounts, AUM fees tend to cost more in absolute dollars as balances grow
- General AI chatbots offering financial commentary are, in most cases, neither discretionary nor fiduciary — treat their output as one data point, not as regulated advice
- Whichever model you choose, confirm the custody arrangement and the actual written grant of authority before assuming either is in place
Frequently Asked Questions
What is the main difference between discretionary and non-discretionary AI investment advice?
The main difference is who executes the trade. Discretionary AI advice means the platform has signed authority to buy and sell in your account without asking each time. Non-discretionary AI advice means the system can only recommend — you have to place every trade yourself.
Is a robo-advisor automatically discretionary?
Not automatically. Most mainstream robo-advisors that manage a diversified account and rebalance it for you are structured as discretionary, but some platforms market themselves as AI-powered while only offering recommendations that you must execute manually. Check the account agreement rather than assuming based on the marketing language.
Does a non-discretionary AI tool still owe me a fiduciary duty?
It depends on who is behind the tool. If it’s offered by a registered investment adviser, some duty of care generally still applies to the recommendations themselves, though it’s usually lighter than the continuous fiduciary duty tied to discretionary management. If the tool is a general-purpose AI chatbot with no advisory registration, no fiduciary duty applies at all.
Which option costs less?
It depends on your account size. Discretionary AI advisors typically charge 0.25% to 0.50% of assets annually, which scales up in dollar terms as your balance grows. Non-discretionary tools often charge a flat subscription fee, which can be cheaper in percentage terms on larger balances but costlier, proportionally, on smaller ones.
Can I switch from non-discretionary to discretionary later?
Generally yes. Most platforms that offer both models let you upgrade by signing the additional trading authorization required for discretionary management. Read the new agreement carefully, since granting discretion changes your legal relationship with the firm, not just the app’s behavior.
Who is liable if a discretionary AI system makes a bad trade?
The firm holding discretionary authority generally bears fiduciary liability for trades made under that authority, since it made the decision on your behalf. In a non-discretionary relationship, liability is narrower for the platform because the final decision to execute rested with you.
References
- U.S. Securities and Exchange Commission — Investment Advisers Act of 1940, provisions on fiduciary duty and discretionary authority.
- U.S. Securities and Exchange Commission — Custody Rule, Rule 206(4)-2 under the Investment Advisers Act.
- FINRA — Regulatory Notice 24-09, Artificial Intelligence in the Securities Industry (June 2024).
- U.S. Securities and Exchange Commission — Regulation Best Interest (Reg BI), standards of conduct for broker-dealers.
- U.S. Securities and Exchange Commission — Proposed rule on conflicts of interest associated with the use of predictive data analytics by broker-dealers and investment advisers (July 2023).
- U.S. Securities and Exchange Commission — Form ADV and Form CRS disclosure requirements for registered investment advisers.
