Banks and financial advisers put to the test by artificial intelligence

Artificial intelligence promises to save banks time. The question is how they will use that time: to get to know clients better and support them in the most difficult decisions, or to make product placement more efficient? Technology can serve both purposes. The difference depends on what banks choose to reward and on how they decide to compensate advisers for their work.

For those who manage wealth, the issue goes beyond the ability to build a portfolio. A system can compare costs and risks in a few seconds; what is harder is understanding what a family can afford to lose, what commitments it must meet and when the most useful advice is to make no move at all. And if a recommendation turns out to be unsuitable, one practical question remains: who is accountable to the client?

We discuss this with Francesco Megna, whom we asked what will change for banks, advisers and savers. The key issue is whether artificial intelligence will make wealth advisory accessible even to those who receive little of it today, and what choices will be needed for this possibility to become a real service.

Artificial intelligence is changing the banking system. Are we facing a simple technological evolution or a structural transformation of the banks’ business model?

It depends on where the bank decides to use it. If artificial intelligence is used to summarise a document, route requests or prepare a case more quickly, we are dealing with a technological evolution: the bank is still doing the same work, but in less time. The transformation becomes structural when it changes how the bank decides which clients to serve, with which people and at what price.

Think of a family with a mortgage, a cash reserve and some investments. Today it may speak to different people depending on the need. If the bank can bring this information together and use it, within the permitted limits, to identify a problem in advance, the service can change. For example, it may notice that the available cash is shrinking just as an important expense is approaching. At that point it can offer a review of the situation before the client asks for help. The same technology, however, can be used to identify the moment when that family is most likely to subscribe to a product. In one case it improves the ability to assist; in the other it improves the ability to sell.

That is the dividing line. Artificial intelligence can make the current model far more efficient, or it can help the bank build a different service. The bank makes that choice when it decides what to measure and reward: the quality of decisions for the client, the number of cases closed, or the volume of products placed.

How much of the investment in AI is really aimed at improving client services, and how much instead at cutting operating costs and staff?

I would not put a percentage on it. The surveys by the European Banking Authority show uses across several areas, including customer support and segmentation, but knowing how many banks adopt a technology for a certain function does not tell us how they allocate spending, nor what the effect on employment is. Moreover, the same application can reduce costs and improve service: if it takes less time to review a document, the bank saves money and the client may receive a reply sooner.

The assessment has to be made on results. A bank may announce that its virtual assistant handles thousands of conversations. What matters to the client is how many requests are resolved at first contact, how many errors occur and how easy it is to speak to a person when the case is out of the ordinary. If, to contest a payment, the client has to repeat the same story three times, the bank may have lowered the cost of the first contact and worsened the service overall.

As for staff, the most useful question is what happens to the hours saved. They may translate into fewer employees, or into more time to handle complex cases, contact neglected clients and train staff. To distinguish between the two, I would look at a few concrete indicators: response times, complaints, access to an operator and the amount of time advisers actually devote to clients. That is where the investment’s balance sheet becomes visible.

If AI can analyse wealth and build personalised portfolios, what space will remain for the financial adviser?

The part of the job most exposed is the one that can be turned into a repeatable process: reconstructing a portfolio’s exposures, comparing costs, identifying concentrations, simulating scenarios. It is useful work, and technology can make it faster. But building a portfolio consistent with a set of data is one thing; understanding whether those data truly describe the client’s situation is another.

Take an entrepreneur who has €1 million available to invest. If we look only at the securities account, we might consider him able to withstand significant swings. But if almost all the rest of his wealth is tied up in the business, part of the debt is personally guaranteed and he will need to buy out a partner in the coming years, the picture changes. He may need more liquidity and less exposure to the risks already present in the company. The point is not whether the machine can do the calculation: it is who brings out the missing information, verifies it and decides what conclusions to draw from it.

The adviser will therefore have fewer reasons to present himself as the person who knows which fund performed best. He will have more reasons to show that he can connect financial assets, property, business and family; that he knows when it is appropriate to involve another professional; that he can explain to the client why, at certain moments, the best decision is to do nothing. The quality of his work will also be seen in the ability to revise the plan when circumstances change and to be present when the client has to make a difficult decision.

Could artificial intelligence speed up the shift from product-based remuneration to independent advice, paid directly by the client?

It can speed it up, but it can also reinforce the existing model. If it makes it easier to compare costs, risks and alternatives, it becomes harder to justify a recommendation simply because a product is in the network’s catalogue. If, on the other hand, the system is built to identify the product the client is most likely to subscribe to, the technology makes distribution more efficient without changing incentives.

The economic issue is this: how do you pay for good advice when that advice does not produce a sale? If the work consists of reconstructing a family’s wealth, ordering priorities, correcting a concentration and deciding not to subscribe to anything for the time being, the client has received a service. A fee for the plan, for the periodic review or for a defined mandate makes more visible what the client is paying for. Naturally, that fee too must be judged: it has to be understandable and proportionate to the work done.

A fee must then be distinguished from independence in the regulatory sense. Being paid directly by the client is not enough, on its own, to qualify advice as independent. MiFID II requires, among other things, the assessment of a sufficiently broad and diversified range of instruments and sets specific limits on the possibility of retaining third-party incentives. Even a non-independent service must still comply with conduct rules and act in the client’s best interest. The question for networks is whether they want the value of the service to be recognisable and remunerable even when the result of the advice does not coincide with the placement of a product.

Can an algorithm really understand the emotional and psychological components that influence investment decisions?

A system can pick up useful signals. It can detect that a client says he can tolerate risk but, at every downturn, asks to liquidate investments. It can help ask more concrete questions: 'If the portfolio fell by 15% just when your son has to start university, what expenses could you postpone?' A question like that often says more than a box ticked on a scale from one to five.

The limit is that the same behaviour can have different causes. The request to sell during a downturn may stem from fear, but also from an unexpected expense, a family conflict or the fact that the client has finally understood a risk that had not been properly explained before. Behaviour is observable; its cause has to be reconstructed. That is where a conversation can change the quality of the decision, also because the client can correct the interpretation that is being proposed.

I would always distinguish between the psychological willingness to see wealth fluctuate and the financial ability to absorb a loss. A person may be calm in the face of volatility and still not be able to afford to lose money that will be needed in two years. Another may have ample resources but lose sleep over modest swings. Suitability rules already distinguish the ability to bear losses from risk tolerance. AI can help bring inconsistencies to light; the value of advice lies in clarifying them before they become a problem.

Can the technological advantage narrow the gap between large fortunes and small savers?

Yes, especially if it lowers the cost of basic planning. A family with a modest financial wealth can benefit greatly from a clear inventory of assets and debts, a plan for future expenses and a periodic review of costs. It may need to understand how much to keep readily available and what can be committed over the long term. If done entirely by hand, that is work that can become too expensive relative to the fee that family is willing to pay.

The advantage can come from a multi-level service: digital tools to collect and update data, a professional to interpret the steps that require judgement and specialists when the issue warrants it. A family with €150,000 in savings, a home of its own and an inheritance to be divided may have a more complex situation than its bank balance suggests. It does not necessarily need a dedicated structure, but it may need a serious discussion before selling a property, making an early gift or tying up liquidity.

The risk is that access to a very convincing interface will be mistaken for access to good advice. The gap only truly narrows if the client receives a clear plan, transparent costs and the possibility of speaking to a person when the case requires it. Otherwise, access to recommendations broadens without necessarily improving the quality of decisions.

What risks does the use of AI entail in terms of privacy, data protection, transparency and conflicts of interest?

The first concerns what can be inferred by cross-referencing data. On their own, an account movement or a change in spending says little. Together they can reveal financial difficulties, family changes or habits that the client did not think he had disclosed. The bank must know what information is necessary for each use, on what basis it processes it, who can access it and what happens when the data are passed to a supplier. The fact that information is already in the bank’s systems does not authorise any subsequent use.

The second risk is opacity. For an investment recommendation, useful transparency does not mean handing the client the model code. It means being able to explain which pieces of information had an impact, why the proposal is consistent with the client’s situation and what alternatives were considered. For decisions based solely on automated processing that produce legal effects or similarly significant effects on the person, the GDPR provides specific limits, exceptions and safeguards. Not every use of AI falls within this category, but a merely formal human check is not a credible solution.

The third risk is the conflict of interest built into the objective. A system can profile the client with great precision and then rank proposals according to the likelihood of sale or the intermediary’s profitability. The client would see a highly personalised recommendation without knowing for which outcome it was personalised. That is why controls must also concern the products the system tends to propose, the ones it leaves out and the treatment reserved for clients with similar needs.

If AI suggests an unsuitable investment that causes a loss, who is concretely responsible?

The first point of reference is the intermediary that provided the advisory service. European rules expressly state that, if advice or portfolio management is provided wholly or partly through an automated system, responsibility for the suitability assessment remains with the firm providing the service. The bank or investment firm cannot therefore reply to the client: 'The programme made the proposal'.

This does not exclude the role of the individual adviser. If he merely confirmed a recommendation while knowing that the profile was outdated, or ignored relevant information communicated by the client, his conduct must be assessed. In the Italian case of off-premises solicitation, Article 31 of the Consolidated Law on Finance provides for joint liability of the party that gave the assignment for damage caused to third parties by the authorised adviser. Any defect attributable to the technology provider opens a separate issue, to be assessed on the basis of the facts and the applicable relationships; it does not cancel the intermediary’s obligations toward the client.

Finally, a loss alone does not prove that the investment was unsuitable. It is necessary to reconstruct what was known at the time of the recommendation, what risk the client could bear, what information was provided and what link there is between the possible breach and the damage. For example, if a client had stated that he would need that sum within a year and the system recommended an illiquid instrument, it must be possible to verify whether the information had been collected, whether the model had taken it into account and who had approved the proposal. Without this trail, the ability to correct the system and prevent the next mistake is also reduced.

Will AI reduce the number of advisers or reward those with more advanced wealth, tax and succession expertise?

Both things can happen at once. A network may need fewer people to carry out standard activities, but more professionals capable of handling complex cases. Or it may use the hours saved to offer a service to clients who currently receive little advice. The effect on headcount depends on what the network wants to do with the time freed up. That is why I would be cautious in predicting how many jobs will disappear: it is easier to identify the tasks that will change.

Selection will mainly concern the ability to understand when a financial decision is, in reality, a family or business decision. Before proposing how to invest the proceeds from the sale of a company, for example, one must understand what commitments remain with the seller, how much liquidity is needed and what the family intends to do with that wealth in the coming years. The adviser must know how to frame the questions and coordinate with the accountant, notary or lawyer for the aspects within their remit. His contribution also lies in ensuring that correct specialist answers, taken separately, do not together produce an incoherent plan.

The test will be to see how networks evaluate the work of their advisers. If the dominant criterion remains the volume placed, artificial intelligence will make that model more productive. If what counts is the quality of the plans, the ability to follow them over time and the clarity with which the client understands his or her choices, technology can encourage more genuinely wealth-based advice. The future of the profession will depend on this organisational decision too, as well as on individual professionals’ ability to use the new tools.