Khắc phục thành kiến giới tính thuật toán trong tài chính cá nhân dựa trên AI

Artificial intelligence is transforming our world and financial services are no exception. AI is reshaping the personal banking sector but where does it currently stand on gender parity, transparency and fairness?

When someone applies for a loan today, there is a growing chance that no human ever reads their application. A data–driven algorithm decides whether they qualify, how much they can borrow, and how risky they are considered, often in a matter of seconds and without explanation, quietly shaping financial opportunities in ways most people never see but feel in their everyday lives.

These systems are usually presented as neutral tools: faster than people, more consistent, less prone to prejudice.

In a sector long criticised for opacity and bias, that promise is appealing and frequently echoed in industry and policy debates. But that promise rests on a fragile assumption, rarely made explicit, that the data these systems learn from reflects everyone’s lives equally.

A recent report by the EU Agency for Fundamental Rights, based on fieldwork in five member states, examined how high–risk AI systems are governed under the EU AI Act in areas such as employment, public benefits and law enforcement. It found a striking gap between legal ambition and practice: while risks of discrimination are broadly acknowledged, providers and deployers often lack the tools, expertise and guidance to assess them systematically. Self–assessments tend to be inconsistent, and oversight remains thin.

This is an important issue. When the data feeding these systems fails to capture the reality of women’s financial lives with the same depth and accuracy as men’s, the result is not just a technical shortcoming but a structural distortion, one that shapes who gets access to credit, on what terms, and with what long–term consequences. For AI-driven finance to be fair, women must first be “visible” in the data on which these systems rely.

Algorithms do not judge fairness or ask whether an outcome makes sense, but estimate what is most likely to be correct based on the data they are given, drawing patterns and projecting them forward. When data is incomplete or distorted, the system’s conclusions rest on shaky assumptions from the start.

If women are underrepresented, poorly measured, or never analysed separately from men, the system cannot see unequal outcomes, and what it cannot see, it cannot correct. Bias is simply carried forward and made routine.

This dynamic is easy to miss when discussions stay at the level of models and regulation, but its effects become clear as soon as automated systems are observed in practice. Across different countries, evidence shows how quickly inequality can be embedded in algorithmic decisions, not because systems are designed to discriminate, but because they faithfully reproduce the distortions already present in the data they learn from.

Kenya offers a telling illustration. According to published studies, a widely used digital lending algorithm consistently offered women smaller loans than men, in some cases by more than a third, despite stronger repayment performance. The system did not single women out deliberately: it simply learned from data shaped by long–standing social and economic disparities, and then applied those patterns at scale.

What matters in this example is not Kenya itself, but what the case makes visible. The algorithm did exactly what it was designed to do, learning from past behaviour and applying those patterns consistently, yet without the ability to distinguish between women’s and men’s outcomes, there was no way to detect that inequality was being reproduced in real time. The problem was not automation, but blindness.

How can finance overcome the gender blind spot?

That is where sex-disaggregated data becomes essential. By sorting financial data by gender, regulators, financial institutions, and technology designers can uncover the impacts of automated systems, identify who has access to finance, and pinpoint areas where outcomes begin to diverge. Without that visibility, gender gaps remain hidden, and hidden gaps have a habit of becoming permanent. In digital finance, data is “a girl’s best friend”, not as a slogan, but as a practical condition for accountability.

Most financial institutions already record a customer’s gender as part of basic identification. On paper, the information is there, embedded in routine reporting and basic customer records. In practice, however, recording a variable is not the same as using it. In many countries, the sex of the customer appears in databases but is never analysed, reported, or monitored by supervisors, including in core supervisory frameworks such as prudential reporting. Too often, the data already exists, but it is collected, filed away, and then quietly ignored. The problem lies not in what can be done, but in what is done.

Fairer finance: developing countries are leading the way

The picture looks very different in countries often assumed to have fewer resources. In parts of Latin America and Africa, regulators have required sex-disaggregated reporting for years and regularly publish data on gender gaps in finance.

In Chile, financial authorities have tracked gender differences in loans and deposits for more than two decades, publishing regular sex-disaggregated financial statistics.

In Mexico, regulators combine bank data with national household surveys to understand how women and men use financial services and how they perform as borrowers.

That visibility has had practical consequences. In Mexico, supervisory data showed that women’s loans were smaller but less risky, evidence that fed into changes in loan loss provisioning rules.

In Chile, the data revealed that equal access to accounts did not translate into equal outcomes in savings or insurance, prompting more targeted policy responses. Once these gaps became visible, they became far harder to ignore.

Seen from this perspective, the situation in many high–income economies looks less like a technical lag and more like an institutional hesitation. In much of Europe, gender data remains voluntary or fragmented despite advanced data infrastructures, a failure not of technical capacity but of institutional choice. My upcoming policy paper “Data Are a Girl’s Best Friends: Tackling Digital Financial Inequality Through Sex‑Disaggregated Data”, due to be published in May explores this.

As artificial intelligence becomes more deeply embedded in financial decision–making, that choice becomes harder to defend. At a time when Europe is implementing the EU AI Act and debating how to regulate algorithmic decision–making in finance, the absence of systematic gender data raises a basic question: how can fairness be monitored if the data needed to detect inequality is never analysed?

Making women visible in the data is not symbolic. Without it, fair finance is little more than a claim.

A weekly e-mail in English featuring expertise from scholars and researchers. It provides an introduction to the diversity of research coming out of the continent and considers some of the key issues facing European countries. Get the newsletter!

Eliana Canavesio est membre de Volt Europa.