Your Banking App Calls It Coaching. Regulators Call It Something Else.

Your app has started telling you things. Move this to a savings pot. You are on track to overspend this month. This subscription looks unused. The rate looks good, consider converting now.

Every one of those sentences was drafted by someone who thought carefully about where a legal line sits, and made very sure to stay on the inexpensive side of it.

That is not a scandal, and it is not a reason to ignore the app. It is the single most useful thing to understand about AI money coaching, because it tells you precisely how much weight each suggestion was designed to carry — and what happens to you if it turns out to be wrong. Once you have chosen a platform from our comparison of AI banking apps, this is the question that decides how you should actually use it.

The line, in the regulator's own words

In the European Union, the relevant definition sits in MiFID II. Article 4(1)(4) defines investment advice as:

the provision of personal recommendations to a client, either upon its request or at the initiative of the investment firm, in respect of one or more transactions relating to financial instruments

Definition checked in ESMA's Interactive Single Rulebook, MiFID II Article 4, on 5 September 2026.

Four elements have to be present at once: it must be a recommendation, it must be personal, it must be made to a client, and it must concern transactions in financial instruments. Remove any one of them and the activity falls outside that definition — and a firm that is not giving investment advice is not carrying the duties that come with giving it.

That single sentence explains an enormous amount about how your app is written.

Notice in particular the last element. A plain savings account, a payment account and a currency balance are not financial instruments under that definition. This is one honest reason so much in-app coaching concerns exactly those products: the firm can be genuinely helpful about them without stepping into a regulated advice perimeter. It is also why the language of the nudges is so carefully impersonal — "customers like you", "you could", "consider" — rather than "you should".

Why every app wants to be on the guidance side

The bottom two rows are the whole story. Advice is expensive because someone has to take responsibility for it. Guidance is cheap because nobody does. An AI assistant is an extraordinarily efficient way to deliver the cheap thing at scale, and the incentive to keep it on that side of the line is overwhelming — which is exactly why a well-run firm is careful about the wording, and a badly-run one is careless with it.

Five questions that reveal which side you are on

Put these to the assistant, or to a human at the same firm, before acting on anything meaningful:

1. "Is this a personal recommendation?" A firm that is giving guidance will say no, usually quickly and clearly, because the answer is load-bearing for them.
2. "Have you assessed whether this is suitable for me?" Suitability is the technical word. If the answer is no, the suggestion is a general observation wearing your name.
3. "What about my situation did you not take into account?" The honest answer is usually long: other accounts, debts, dependants, tax position, job security, everything held elsewhere.
4. "Does anyone earn a fee, commission or margin if I do this?" An in-app suggestion to convert currency, upgrade a tier or move into a partner product may be perfectly sound and still be a revenue line.
5. "Can I have this in writing, with a complaint reference if I need one?" Anything that cannot survive being written down should not survive being acted on either.

The arithmetic a nudge does not show you

The most common coaching prompt in this category is some version of move money here and you will be better off. It is testable in about four minutes, using numbers from the two product pages rather than from the prompt.

Take A, the amount the app proposes to move. Take D, the difference in annual rate between where the money is now and where it would go — read from both product pages, because a nudge quoting only the destination rate is not a comparison. The gross annual gain is A multiplied by D.

Then subtract, in this order:

• any account or plan fee charged for holding the destination product
• any conversion cost if the two sit in different currencies
• any tax on the interest or gain in your jurisdiction
• the cost of the money being less available, if the destination locks it up

Suppose — and these are numbers you substitute from your own statements, not rates we are quoting at you — that A is 5,000 in your currency and D works out at one percentage point. The gross gain is 50 a year. If surfacing the suggestion required a paid tier at anything over about four units a month, the feature has cost you money to be helpful. If the destination charges a conversion spread on the way in and the way out, the gain can disappear entirely on a round trip.

None of that arithmetic requires a view on the product. It requires the two rates and a calculator, and it is the reason our verdict on in-app FX-timing prompts in which AI features are worth paying for was so unenthusiastic. The related question of what those conversions actually cost is in what multi-currency accounts really cost and exchange rates explained.

What you give up on the guidance side

Three things, and none of them is visible in the interface:

• The suitability check. Nobody has asked whether this is right for you specifically, because nobody is required to. The model has your transactions. It does not have your circumstances.
• The redress route. If a regulated recommendation turns out to have been unsuitable, the complaint is about the recommendation. If a guidance nudge turns out badly, the complaint is about the service — a much narrower thing to argue.
• The paper trail. Regulated advice generates a documented rationale. A dismissed in-app card generates a log entry, at best, and you cannot see it.

There is a fourth, subtler loss: framing. An assistant that surfaces one option at the moment you open the app has already made a choice on your behalf, and it made it without the duty that would normally attach to making choices on your behalf.

Outside the EU, ask the same structural question

The names differ everywhere. The perimeter does not. Whatever your market, the questions that matter are the same three: does a regulator define a category of personal recommendation that carries extra duties, is your provider authorised inside that category, and does your country provide an independent complaints body once the firm's own process is exhausted. Look those up for your own jurisdiction and provider rather than assuming any single regime applies — and if you are choosing across borders, global banking regulation is the wider map.

The counter-argument, and it is a strong one

Regulated advice is expensive, and in most markets it is effectively unavailable below a certain amount of money. For a very large number of people the realistic choice is not between an AI coach and a human adviser. It is between an AI coach and nothing at all — and an app that reliably points out an unused subscription, a duplicated charge or an overdraft heading for you next Thursday is worth considerably more than a regulated recommendation nobody will ever offer them.

So the failure mode is not that apps give guidance. Guidance is good. The failure mode is guidance dressed in the clothing of advice: confident, personalised in tone, arriving at the moment of decision, and carrying none of the responsibility that tone implies. Use it for what it is genuinely good at — noticing things in your own data that you had not noticed — and do the arithmetic yourself before you move money. That habit is most of what learning personal finance amounts to in practice.

FAQ

Is an AI assistant allowed to give regulated advice at all? It depends on the firm's permissions and on your jurisdiction, not on the technology. Some firms are authorised to advise and choose to deploy automated advice under those permissions. Most in-app coaching is deliberately not that, and says so in the small print.

Why does the app keep saying "this is not financial advice"? Because that disclaimer is doing real legal work. It is not boilerplate politeness. Treat it as the most informative sentence on the screen.

If it is not advice, is it worthless? No. Pattern-spotting in your own transaction history is the thing these systems are genuinely good at, and it is where nearly all the value sits. Judgement about your circumstances is where they are weakest.

What about AI suggestions on crypto products? Apply the same test, and then more scepticism, because yield products carry risks that no coaching prompt has room to express. Start from crypto banking explained.

Sources

• MiFID II, Directive 2014/65/EU, Article 4(1)(4) — definition of investment advice, via ESMA's Interactive Single Rulebook: esma.europa.eu (checked 5 September 2026)

Disclaimer: BestAiGlobalBank is an independent comparison site published by NorwegianSpark SA. This article explains a regulatory distinction; it is general information and not financial, legal or investment advice. Rules differ by country — verify the position with your own regulator and provider.