AI trading vs robo-advisor vs index fund: cost, protection and evidence

These three are usually compared on headline fee, which is the least informative of the three things that actually separate them. The others are what protection you get when something goes wrong, and what evidence supports the expected outcome. On cost, the arithmetic is unforgiving: over 25 years at a 7% gross return, a 1.00% annual fee removes 20.9% of the final balance. On protection, only two of the three involve a regulated firm that owes you anything. On evidence, one of the three has fifteen years of index data behind it and one has fifteen years of index data against it.

Published 27 August 2026 · Updated 28 August 2026 · AI Trading Book Editorial · Reading time about 12 minutes

What each one actually is

The three are often presented as competing products. They are not the same category of thing, and the differences in category matter more than the differences in price.

The last row is the one people skip. An index fund has one failure mode. A robo-advisor has two. An automated trading system has at least three, and the third — operational failure — is the one that produces losses unrelated to any market view, as our API key security page documents.

The cost stack

Comparing headline fees compares one layer of three or four. What matters is the total annual cost as a percentage of what you actually invest, and the shape of that cost differs fundamentally between percentage-based and fixed-fee approaches.

What fees cost over time

The following is our own calculation with stated assumptions, not a citation. A starting balance of 100,000 currency units, a 7% gross annual return, a 25-year horizon, and the annual fee deducted from the return each year. The 7% figure is an assumption for illustration and is not a forecast.

Two observations follow. The difference between a 0.25% and a 1.00% annual cost is not a rounding detail; over 25 years it exceeds the amount originally invested. And a fee is a certain cost paid against an uncertain return, which is why it is the one variable in the comparison worth optimising hard.

Why fixed costs change the picture for small accounts

Percentage fees scale down with the balance. Subscriptions do not, and that asymmetry decides whether an AI tool is even arithmetically viable.

Take a subscription of roughly 720 units a year — around the middle of the retail tier pricing documented on our platforms page, before data add-ons. On a 100,000 account that is 0.72% a year, comparable to a managed service. On a 10,000 account it is 7.2% a year, before a single trade. On a 5,000 account it is 14.4%.

A 7.2% hurdle has to be cleared before the strategy breaks even against holding cash, and it must be cleared consistently. Set that against the accuracy evidence on our accuracy page — a few points of edge over a well-chosen baseline, before trading costs — and the arithmetic answers itself for small accounts.

Why there is no provider fee table on this page

A comparison page would normally list named providers with their fees. We are not publishing one, and the reason is worth stating because it is the same reason the rest of the site is built the way it is.

Checking secondary sources for UK robo-advisor fees produced conflicting figures for the same provider. One well-known provider's headline annual fee appeared as 0.25%, 0.62%, 0.65%, 0.72% and 0.75% across five comparison sites — largely because each was quoting a different portfolio tier, or including or excluding the underlying fund charge, without saying which. Australian and New Zealand coverage has the same problem with fewer sources.

Our editorial policy does not permit publishing a figure we cannot trace to the provider's own page with a capture date. Publishing a table assembled from conflicting affiliate reviews would look more useful and be less true. What we can say with confidence is the structural point: robo-advisor management fees layer on top of the underlying fund charges, so the all-in cost is always higher than the tracker inside it.

If you are choosing between providers, take the fee from each provider's own pricing page on the day you decide, confirm whether it includes the fund charge, and confirm which portfolio tier it applies to. Those three questions resolve most of the discrepancy above.

Protection: what you get when something goes wrong

This is the dimension that headline-fee comparisons omit entirely, and it is the one that matters most in the situations where anything matters.

The asymmetry in the last two rows is the practical core of it. Buy a regulated service and a firm carries duties toward you and answers to a complaints scheme. Buy software and operate it yourself, and every outcome is yours — including the ones caused by a bug, a bad data feed or a compromised API key. Detail by market is on the UK, Australian and New Zealand regulation pages.

Evidence behind each approach

The three are not equally supported, and the gap is not close.

• Index tracking has the strongest evidence, and much of it comes from the failure of the alternatives. The Eurekahedge AI Hedge Fund Index returned 9.8% annualised against 13.7% for the S&P 500 from December 2009 to July 2024, and 115% cumulative against 210% for the S&P over January 2011 to January 2020.
• Robo-advice is a delivery mechanism rather than a strategy. Its portfolios are typically index funds, so its expected outcome is the index outcome minus the additional management layer. That is a legitimate trade — you are buying discipline and rebalancing, not excess return.
• AI trading has the weakest evidence for outperformance. A peer-reviewed review of 27 machine-learning equity experiments found no conclusive evidence of machine-learning funds delivering returns at scale, and careful directional accuracy sits a few points above a well-chosen baseline before costs.

What AI does demonstrably well is different work. The Bank of England and FCA survey of 118 firms found the leading use cases were process optimisation at 41%, cybersecurity at 37% and fraud detection at 33% — cost, control and detection rather than return generation. That is covered on our evidence page.

Most robo-advisors are not AI

Marketing has blurred this, so it is worth stating plainly. A typical robo-advisor asks you a risk questionnaire, maps the answers to one of a handful of model portfolios built from index funds, and rebalances on a schedule or when drift exceeds a threshold. There is no price prediction and usually no machine learning.

That is automation, and it works, because the thing being automated — consistent allocation and unemotional rebalancing — is a task where consistency genuinely helps. It is a different proposition from a system attempting to forecast prices, and it should not be evaluated as though it were.

Who each one suits

Combining them is common and reasonable: a passive core with a small active satellite. If you do that, size the satellite as money you can lose entirely, and measure it against the passive core rather than against zero. Measuring against zero makes any gain look like success even when the core would have done better with no effort.

What to compare before choosing

1. Total annual cost as a percentage of what you will actually invest, including fund charges, platform fees, subscriptions, data and expected trading costs.
2. Whether a regulated firm owes you duties, and which register you can check that in.
3. What happens if the provider fails — custody arrangements, compensation schemes and complaint routes.
4. Time required, honestly assessed. A system needing weekly maintenance has a real cost that no fee table shows.
5. What evidence supports the expected outcome, and whether the vendor cites any.
6. Whether you would stay invested through a 30% drawdown, which decides more real outcomes than any of the above.

What we could not establish

• Verified per-provider robo-advisor fees for AU, NZ and the UK. Secondary sources conflict, in one case reporting five different headline fees for the same provider. Data not found at the standard we publish to.
• Any study comparing outcomes for retail users across these three approaches. No like-for-like performance research was identified. Data not found.
• Robo-advisor assets under management by country for our three markets. Data not found.

Key takeaways

• Cost is the one variable you control, and it compounds: 1.00% a year removes 20.9% of a 25-year outcome.
• Account size decides whether an AI tool is arithmetically viable, because subscriptions do not scale down.
• Two of the three come with a regulated firm that owes you something. The third generally does not.
• A robo-advisor is an index fund with a management layer, so its expected outcome is the index minus that layer.
• The evidence favours the cheapest option — largely because the expensive alternatives have underperformed it.
• Take fees from the provider's own page on the day you decide. Comparison sites contradict each other and each other's tiers.

Frequently asked questions

What is the difference between AI trading, a robo-advisor and an index fund?

An index fund is a product that tracks a market. A robo-advisor is a licensed service that builds and rebalances a portfolio of such products for you. AI trading tools are software you operate yourself to make trading decisions. Only the first two are regulated financial services in the markets we cover.

Which is cheapest?

Index funds, usually by a wide margin, because the total cost is one ongoing charge plus a platform fee. Robo-advisors add a management layer on top of the same underlying funds. AI trading tools add subscription, data and trading costs that are not proportional to the portfolio and therefore hit small accounts hardest.

How much do fees actually cost over time?

On our own calculation — 100,000 units, 7% gross annual return, 25 years — a 0.25% annual fee reduces the final balance by 5.7%, a 0.60% fee by 13.1%, and a 1.00% fee by 20.9%. Fees compound in exactly the way returns do.

Why does this page not list robo-advisor fees by provider?

Because secondary sources conflict. In one search, a single UK provider's headline fee was reported as 0.25%, 0.62%, 0.65%, 0.72% and 0.75% across five comparison sites, largely because different portfolio tiers were being quoted as the headline. Our editorial policy does not permit publishing figures we cannot trace to the provider's own page.

Which has the most regulatory protection?

Robo-advisors and index funds bought through licensed providers, because both involve a regulated firm with duties to you and access to a complaints scheme. AI trading software you operate yourself generally involves no regulated relationship at all — the vendor sells you a tool, not a service.

Which has the best evidence behind it?

Index tracking, by a large margin. AI hedge fund indices returned 9.8% annualised against 13.7% for the S&P 500 over roughly fifteen years, and a review of 27 machine-learning equity experiments found no conclusive evidence of returns at scale.

Do robo-advisors use AI?

Mostly not in the sense implied by marketing. A typical robo-advisor uses a risk questionnaire, a rules-based allocation model and scheduled rebalancing across index funds. That is automation, and it works, but it is not machine learning predicting prices.

Is an AI trading tool ever the right choice?

For research, screening, execution consistency and enforcing risk rules mechanically — plausibly yes. As a route to returns above a passive benchmark, the published evidence does not support it. The two uses have very different expected outcomes.

What costs do people forget with AI trading tools?

Data add-ons, broker commissions, spread, slippage, overnight financing, currency conversion and the time spent maintaining the system. Subscription is usually the smallest component, and none of these scales down with a small account.

Why do fixed costs matter more for small accounts?

Because a subscription is the same amount whatever the balance. A tool costing roughly 720 units a year is about 7.2% of a 10,000-unit account before a single trade, which is a return hurdle far above anything the published accuracy evidence supports.

Can I combine these approaches?

Many people do, holding a passive core and running a small satellite allocation actively. If you do, size the satellite as money you can lose entirely and measure it against the passive core rather than against zero, so you can see whether the activity added anything.

What should I compare before choosing?

Total annual cost as a percentage of the amount you will actually invest, whether a regulated firm owes you duties, what happens if the provider fails, how much time it needs, and what evidence supports the expected outcome. Headline fee alone answers none of these.

Sources

• Eurekahedge AI Hedge Fund Index — 9.8% annualised against 13.7% for the S&P 500, December 2009 to July 2024; 115% cumulative against 210% for the S&P 500 and 133% for MSCI World, January 2011 to January 2020.
• Buczynski, Cuzzolin and Sahakian, International Journal of Data Science and Analytics 11(3), April 2021 — review of 27 machine-learning equity experiments; no conclusive evidence of returns at scale.
• Bank of England and FCA — "Artificial intelligence in UK financial services – 2024", 21 November 2024, n=118 — leading use cases: process optimisation 41%, cybersecurity 37%, fraud detection 33%.
• ASIC — REP 828, 20 January 2026 — 68% of retail CFD investors lost money in FY2024, cited for the risk context of active leveraged trading.
• Platform subscription levels — captured from vendor pages, August 2026; consolidated in the data hub and marked verify before use.
• AI Trading Book calculation — Table 2, the fee-compounding chart and the subscription hurdle chart. Assumptions stated in the text.

Informational research only. Nothing on this page is personal financial, legal, tax or investment advice, or a recommendation to use any product or approach. Illustrative return assumptions are not forecasts, and past performance does not indicate future results.