Former Santander executive Javier Rumbo Lorenzo argues that artificial intelligence calculates accurately, but cannot set the investor's goals or their tolerance for losses. The CNMV and ESMA warn that these tools can provide misleading recommendations.
Artificial intelligence has entered wealth management with a promise: to process in seconds more financial information than a person would read in a lifetime. Javier Rumbo Lorenzo, former executive of Banco Santander and independent financial advisor, proposes a division of tasks. The calculation belongs to the model; setting the goals and timelines belongs to the investor. His analysis, reported by rotaaldia.com, starts from a common scene: typing a query on a mobile phone and instantly receiving a portfolio with percentages, funds, and time horizons.
Rumbo Lorenzo distinguishes between calculating well and deciding well. In nearly two decades of banking, he found that the most costly mistakes of an individual do not come from a poorly calculated operation, but from selling out of fear during a downturn, buying late out of euphoria, or taking on a product whose maturity clashes with liquidity needs. Behavioural finance documents this pattern: the pain of a loss weighs more than the satisfaction of an equivalent gain.
An automated system has an advantage in this area because it lacks fear. It applies the same rules on a panic Monday as on a calm Friday and adjusts the weights of each asset without hesitation. In the time it takes a person to open a report, it has already reviewed thousands of data points on prices, fees, and correlations. Rumbo Lorenzo considers this advancement useful for the small investor, who gains access to a discipline previously reserved for high-net-worth individuals.
“An automated manager with a license is subject to the suitability rules of MiFID II, which require it to know the situation of each client before recommending a product. A general-use chatbot, on the other hand, responds to what it is asked and lacks that duty.”
The boundary appears when the decision depends on information that the software does not know. An assistant is unaware that the user is considering buying a home in three years, that they care for a dependent family member, or that they lost sleep the last time their savings fell by ten percent. None of this is known to it unless someone explains it in detail.
Supervisors have reached a similar conclusion. The European Securities and Markets Authority (ESMA) and the CNMV issued a joint communication in March 2025 regarding publicly available AI tools. The text warned that they can generate inaccurate or misleading recommendations and that they lack the obligation to act in the interest of those who use them. The Spanish supervisor revisited the issue in a subsequent report, which warned of mistakes and fabricated responses when small investors resort to such systems without expert review.
For the former banking executive, the regulatory warning does not invalidate the technology but clarifies where it ends. An automated manager with a license is subject to the suitability rules of MiFID II, which require it to know the situation of each client before recommending a product. A general-use chatbot, on the other hand, responds to what it is asked and does not assume that duty.
Before opening any application, an investor should be clear about what they want the capital for and when they will need it, as well as how much decline they are willing to endure without selling. With that data on the table, a language model can be used to compare funds, explain concepts, or simulate scenarios. Without them, any suggestion lacks a foundation, whether human or automatic.
The CNMV's public register allows checking whether the entity offering the service is authorised. Cross-referencing the suggestions of an assistant with other sources complements the practical recommendation of Rumbo Lorenzo and the regulators.
You may also like
