Gender stops being a factor that creates exclusion
En México, donde las mujeres ganan un 20% menos que los hombres y acceden en menor medida a productos financieros básicos, algunas empresas de tecnología financiera están reentrenando sus algoritmos de inteligencia artificial para que el género deje de ser una variable que excluye. La IA, como todo espejo, refleja los prejuicios del mundo que la formó, pero también puede ser corregida por quienes deciden qué valores deben guiarla. Lo que está en juego no es solo la equidad, sino la riqueza colectiva que se pierde cuando la mitad de la población queda al margen del sistema económico.
- Las mexicanas enfrentan una brecha salarial del 20% y tasas de acceso a productos financieros —crédito, seguros, pensiones— que son sistemáticamente menores a las de los hombres, dejándolas expuestas a una vejez sin red de protección.
- Los algoritmos de inteligencia artificial, entrenados con datos históricos cargados de sesgos, pueden reproducir y escalar esas desigualdades a velocidad industrial, convirtiendo la tecnología en un amplificador del problema en lugar de una solución.
- Fintechs como Kueski y GOW están respondiendo con modelos que excluyen el género como variable de decisión, evaluando el riesgo crediticio a través de cientos de otras señales y logrando, en el caso de Kueski, una paridad real entre clientes hombres y mujeres.
- El reto no termina con el entrenamiento inicial: los modelos deben revisarse continuamente para evitar que nuevos sesgos se cuelen por la puerta trasera de datos imperfectos y realidades cambiantes.
- La inclusión financiera de las mujeres no es solo un asunto de justicia: genera efectos intergeneracionales en educación y salud, amplía la base fiscal y trae al mercado la perspectiva de la mitad de los consumidores del mundo.
La inteligencia artificial aprende del mundo tal como es, y el mundo está lleno de prejuicios. Investigaciones recientes —desde Harvard hasta la Universidad de Los Andes en Colombia— han confirmado que los modelos de lenguaje y los algoritmos financieros reproducen los sesgos culturales de los datos con los que fueron entrenados. En México, eso tiene consecuencias concretas: las mujeres ganan en promedio un 20% menos que los hombres, solo el 34% tiene cuenta de retiro frente al 51% de ellos, y apenas el 18% cuenta con algún seguro. La brecha no es solo económica; es también cultural. Más de la mitad de los mexicanos cree que los hijos sufren cuando sus madres trabajan, una de las cifras más altas entre los países de la OCDE.
Sin embargo, la tecnología no es un destino inamovible. Kueski, una fintech mexicana, ha construido un algoritmo de evaluación crediticia con más de 400 variables en el que el género no figura. El resultado es inusual en el mercado actual: una cartera de clientes con paridad entre hombres y mujeres. Brigitte Brousset, directora de GOW, otra startup mexicana de IA orientada a facilitar créditos para pequeñas empresas, señala que los sesgos humanos son en su mayoría inconscientes, y que un algoritmo bien entrenado no los arrastra de la misma manera. La diferencia, dice, es que la IA no amanece cansada ni frustrada.
Pero entrenar un modelo una sola vez no es suficiente. Los sistemas de crédito basados en IA requieren revisión constante para asegurarse de que no estén amplificando nuevas formas de discriminación a través de datos imperfectos. La gobernanza y la supervisión son tan importantes como el algoritmo mismo. Y el argumento para hacerlo bien va más allá de la equidad: cuando las mujeres tienen ingresos estables, invierten más en la salud y educación de sus hijos, amplían el mercado de consumo y fortalecen la base tributaria. Un análisis de Taxdown reveló que las mujeres recibieron en promedio 2,549 pesos menos en devoluciones fiscales que los hombres, reflejo directo de que ganan menos. El 60.5% de los beneficios fiscales favoreció a los hombres. La brecha es estructural, pero por primera vez, algunas herramientas tecnológicas están siendo diseñadas para erosionarla.
Artificial intelligence mirrors the world it learns from, which means it inherits the world's prejudices. Ask an AI system whether it can be sexist, and it will tell you yes—that it reproduces the biases, stereotypes, and inequalities baked into its training data. The technology appears neutral on the surface, but recent research has exposed the illusion. A Harvard study found that language models reflect the perspectives of WEIRD populations: Western, educated, industrialized, rich, and democratic. Colombian researchers at the Universidad de Los Andes and the firm Quantil discovered the same problem in Spanish-language models, showing that cultural and linguistic inheritance shapes algorithmic responses.
Yet AI is not destiny. It can be corrected. It can be retrained. And in the hands of Mexican financial technology companies, it is becoming a tool to close one of the country's most persistent economic gaps: the one between men and women. Mexican women earn 20 percent less than men on average—a gap slightly wider than the Latin American average of 17 percent, according to data compiled by BBVA and the Economic Commission for Latin America and the Caribbean. The disparity extends into every corner of the financial system. Only 34 percent of women hold retirement accounts, compared to 51 percent of men. Just 18 percent of women have insurance, versus 28 percent of men. These are not small differences. They are the difference between planning for old age and facing it unprepared, between having protection against illness and having none.
The barriers are not only economic. They are cultural, invisible but tangible. More than half of Mexico's population—53.2 percent—believes that children suffer when their mothers work, one of the highest figures in the OECD, of which Mexico is a member. This belief shapes lending decisions. Only 30 percent of women access formal credit, despite evidence that they have better payment histories and lower default rates than men, according to Mexico's National Institute of Statistics and Geography. The financial industry, Lisset May of the fintech Kueski observes, still operates on an older assumption: that men are the ones who generate money.
Kueski has used AI to challenge that assumption. The company has achieved something that defies the market's current reality: parity between its male and female clients. The feat rests on a simple principle: the algorithm does not see gender. The company evaluates risk and fraud using more than 400 variables in its internal algorithm, but gender is not one of them. Neither is sexual preference. Decisions that once took days now take seconds, powered by machine learning. The same approach applies to its buy-now-pay-later loans, approved instantly at checkout on platforms like Amazon, where Kueski has a partnership.
Brigitte Brousset, CEO of GOW, a Mexican AI startup that uses intelligent agents and predictive models to help small and medium-sized businesses access loans from financial institutions, frames the challenge differently. "If we use AI in unbiased models, gender stops being a factor that creates exclusion," she says. "If we make decisions the traditional way, with people in the process, biases can exist—not intentionally, but because we are human beings living with biases, mostly unconscious ones." The difference is that AI, once trained correctly, does not drift. It does not wake up tired or frustrated. It does not carry the weight of a lifetime of assumptions.
But AI is not a system you train once and leave running. In lending, models must be reviewed regularly to ensure they remain accurate and are not amplifying biases in new ways. Governance matters. Supervision matters. The data itself matters. When AI learns from imperfect information—and all information is imperfect—it can only be as fair as the humans who built it decide to make it.
Why does this matter beyond fairness? Because women's economic inclusion generates wealth. When women have stable incomes, they invest more in their children's education, health, and nutrition, creating effects that ripple across generations. Their presence in business decision-making brings the perspective of half the world's potential consumers. Their integration into the labor market expands the tax base. A study by Taxdown, a platform for tax filing and refund optimization, found that women received an average of 2,549 pesos less in tax refunds than men—a direct reflection of earning less because they are paid less. Of all tax benefits, 60.5 percent favored men. The gap is not accidental. It is structural. And it is beginning to crack.
Notable Quotes
If we use AI in unbiased models, gender stops being a factor that creates exclusion. If we make decisions the traditional way, with people in the process, biases can exist—not intentionally, but because we are human beings living with biases, mostly unconscious ones.— Brigitte Brousset, CEO of GOW
There is far less financing for women. It is an industry that tends to speak more to the male gender, because this concept still exists that 'the man is the one who generates money.'— Lisset May, senior vice president of sales at Kueski
The Hearth Conversation Another angle on the story
Why does an AI system trained on human data end up reproducing human bias? Isn't the whole point of algorithms to be objective?
Because the data itself is not objective. If you train an AI on information created by a society where women earn less, have fewer loans, and are less represented in certain fields, the algorithm learns those patterns as normal. It's not thinking—it's pattern-matching. And patterns can be prejudices.
So when Kueski says gender isn't a variable in their lending algorithm, what does that actually prevent?
It prevents the algorithm from using gender as a proxy for risk. Historically, lenders assumed women were riskier borrowers. But the data shows the opposite—women actually default less. Without gender in the model, the algorithm has to evaluate each person on their actual financial behavior, not on assumptions about their gender.
But doesn't removing gender from the algorithm mean ignoring a real pattern in the data—that fewer women have credit histories?
Exactly. That's the hard part. The pattern is real, but it's not because women are worse borrowers. It's because they've been excluded. So if you let the algorithm see gender, it learns to exclude them more. If you don't, it has to find other signals of creditworthiness. That's the correction.
Brigitte Brousset said AI needs to be "governed correctly." What does that look like in practice?
It means humans stay in the loop. You train the model, you test it for bias, you monitor it over time to make sure it's not drifting, and you're willing to retrain it when the world changes. It's not a one-time fix. It's ongoing supervision.
If AI can be corrected, why hasn't this happened at scale already?
Because it requires intention. Most companies don't build fairness into their systems from the start. They optimize for profit or speed. These fintechs are choosing to do it differently, but that choice has to be deliberate.