Are Robo-Advisors Actually AI? Mostly No, and the Name Predates the Word
Last updated August 2026
Short answer
Mostly no. A robo-advisor is a rules engine: a questionnaire produces a risk score, the score selects one of about ten pre-built portfolios, and threshold rules handle rebalancing and harvesting. The allocation logic descends from portfolio theory published in 1952. The robo in the name arrived around 2010 and referred to automating the plumbing, not to intelligence. Machine learning does appear in these products, in onboarding, support chat and behavioural nudges, and almost never in choosing what you hold. Walnut is informational and is not an investment adviser.
Nearly every page answering this is written by someone selling one of these products, which is why the answers are vague. The actual answer is not embarrassing. It is that the category solved a real problem with appropriate engineering, and then a marketing word attached itself afterwards.
What is actually under the hood
1. A questionnaire mapped to a risk score
Age, timeline, income, and a few questions about how you would feel in a decline, resolved into a number. The mapping is a table someone wrote, and it is deliberately conservative and deliberately simple.
2. A model portfolio chosen from a small set
The risk score selects one of perhaps ten pre-built allocations. You are not receiving a portfolio computed for you, you are being assigned to a bucket, and thousands of other clients hold the same bucket.
3. Allocation logic from the 1950s
Modern portfolio theory, published in 1952, formalised the idea that the mix of assets matters more than the individual picks. It won a Nobel prize and it remains the intellectual core of every robo-advisor operating today.
4. Rules that fire on thresholds
Rebalance when an allocation drifts past a set band. Harvest when a position falls below its basis by some amount and no wash-sale window applies. These are conditions written by people, not judgements learned from data.
The second point is the one that surprises people most. You are assigned to a bucket rather than given a portfolio computed for you, and thousands of other clients hold exactly the same one. That is not a criticism of the product, since bucketing is how you deliver reasonable allocations cheaply at scale, but it is quite far from what personalised implies.
Where machine learning genuinely does appear
| Function | What is really happening |
|---|---|
| Onboarding and identity checks | Document reading and fraud detection, which is genuine machine learning and nothing to do with investing |
| Support chat | Increasingly a language model answering account questions, and this is what most recent AI announcements refer to |
| Cash-flow and spending features | Categorising transactions and projecting balances, which is pattern recognition on your data |
| Nudges about behaviour | Predicting who is likely to withdraw or stop contributing, then intervening. Real modelling, aimed at retention |
| Choosing what you hold | Almost never. This is the part people assume is AI and it is a lookup table |
The fourth row is the most interesting and the least advertised. Predicting which clients are about to withdraw or stop contributing, then intervening, is real modelling applied to real data, and it is aimed at retention. It also happens to help the client, since the people who withdraw during declines are the ones most harmed by doing so.
Get a recommendation for your situation
Walnut is the AI that knows your portfolio: ask anything in plain English, research any fund, and get an honest second opinion. On the broker you already use, read-only, and you approve every trade. Walnut is not a registered investment adviser.
Three reasons this is fine
Nothing here needs prediction
The value of a robo-advisor is diversification, low cost, discipline and automation, and not one of those improves with a better forecast. A rules engine is the correct engineering for the job it is doing.
Rules are auditable and predictable
You can read a rebalancing band and know exactly what will happen. A learned model that reallocated your retirement savings on patterns nobody can articulate would be a considerably worse product, and it would be harder to regulate.
Prediction is where things go wrong
Products that do claim predictive investing are making a much stronger claim, and the long-run evidence on beating a broad market after costs is unkind to it. The absence of prediction is a feature of the category rather than a gap in it.
It is worth sitting with the second one. A system that reallocated your retirement savings based on patterns nobody could articulate would be a worse product and a harder one to supervise. Rules you can read are a feature, and the reason nobody markets it that way is that auditability does not sell.
What the word is doing when a product does use it well
There is a real distinction between automating money movement and being able to explain a portfolio in language. A language model can read what you hold, tell you why a fund moves the way it does, compare two options and answer follow-up questions, none of which a rules engine attempts, and none of which requires predicting anything. That is a different product from allocation and it is where the word applies honestly.
The place to be sceptical is any claim to predict. Related: what AI-powered actually means on these products, and do robo-advisors beat the market.
FAQ
Are robo-advisors actually AI?
Mostly no. A robo-advisor is a rules engine: a questionnaire produces a risk score, the score selects one of a small number of pre-built portfolios, and threshold rules handle rebalancing and harvesting. The allocation logic descends from portfolio theory published in 1952, and the robo refers to automating the plumbing rather than to intelligence.
Do robo-advisors use machine learning?
In places, though rarely for choosing what you hold. It shows up in identity verification, fraud detection, support chat, transaction categorisation and predicting which clients are about to withdraw. Those are real applications, and none of them touches the investment decision.
Why is it called a robo-advisor then?
The name arrived around 2010 to describe advice delivered by software instead of a person, and it referred to automation rather than artificial intelligence. The word AI attached itself to the category later, largely through marketing, after the products were already built.
Is an AI investing app different from a robo-advisor?
Often yes, and the honest distinction is what the software does. A robo-advisor allocates and manages money by rules. Products described as AI investing usually either analyse and explain, which language models do well, or claim to predict, which is a much stronger claim requiring much stronger evidence.
Does it matter that robo-advisors are not AI?
Only if you were paying for intelligence you are not receiving. The things a robo-advisor is genuinely good at, diversification, low cost, discipline and automation, do not improve with a better forecast, so a rules engine is the right tool. The problem is the marketing rather than the product.
Are there AI-powered robo-advisors?
Several providers now describe themselves that way, and it is worth reading closely to see which part they mean. Almost always it refers to a chat interface, onboarding automation or behavioural nudges rather than to how your allocation is chosen, which remains a table.
Can AI pick better investments than a robo-advisor?
There is no good evidence that anything reliably picks investments well enough to beat a broad market after costs over long periods, and that applies to models as much as to people. What AI does well here is explanation, comparison and analysing what you already own.
What should I look for instead of AI claims?
Total cost including the underlying funds, which account types are supported, whether tax-loss harvesting applies to your situation, what the actual allocation is rather than its risk label, and whether human access is available. All five are knowable in advance and all five affect the outcome.
Related articles
Walnut is informational and is not an investment adviser, and nothing here is investment advice. Implementation varies by provider, and what any specific product automates is described in its own disclosures.