The problem nobody talks about out loud
Ask any CDO or head of analytics at a bank in Chile, Peru, Colombia, or Mexico what keeps them up at night, and most will give you the same answer — eventually. Not regulation. Not competition. Not even bad data. It's the team.
Finding a senior data scientist who understands credit risk modeling, knows how to work with alternative data, can communicate results to a regulator, and is willing to stay at a mid-sized bank in Santiago, Lima, Bogotá, or Monterrey instead of moving to a tech company in São Paulo or joining a US-based fintech remotely — that person is extraordinarily rare. And expensive. And if you find them, keeping them is a different problem entirely.
This is the data science talent gap. And across Chile, Peru, Argentina, Brazil, Mexico, and Colombia, it's not a footnote — it's one of the central constraints on how fast financial institutions can grow, how well they can manage risk, and how effectively they can serve the customers who need them most.
Why these markets have it worse than they should
The talent gap exists everywhere. But in developed markets, institutions have three advantages that partially compensate for it: deeper talent pools, higher salaries that can compete with tech, and simpler data environments where standard models trained on bureau data perform reasonably well.
Across Latin America's six largest economies, none of those advantages apply cleanly. Talent pools are shallower than the size of these economies would suggest. Brazil and Mexico have strong university systems producing more data scientists than a decade ago — but the pipeline of practitioners with real-world experience in credit risk, fraud modeling, and financial analytics is still thin relative to demand. In Chile, Peru, Argentina, and Colombia, the gap is even more pronounced. The best practitioners get recruited by fintechs, global banks, or technology companies — and increasingly, they work remotely for US or European employers while living in Santiago or Medellín, effectively removing themselves from the local talent market.
Salaries are constrained by local economics. A regional bank in Peru or a cooperative lender in Colombia cannot match what a US tech company offers a strong data scientist, even adjusting for purchasing power. The talent that stays is often junior, requires heavy supervision, and takes years to develop the judgment that only comes from experience.
The data environment is harder. Unlike developed markets where bureau data is rich and relatively standardized, data scientists across LATAM work with fragmented internal systems, inconsistent data quality, and the constant need to integrate alternative sources. In Argentina, economic volatility makes historical data less reliable. In Peru and Colombia, bureau penetration outside major cities drops sharply. In Mexico, the informal economy means that a large share of creditworthy customers simply don't appear in traditional data sources.
The result is a painful paradox: the markets that most need sophisticated predictive models are the ones with the least capacity to build them.
What this costs in practice
Slow time to market. When a model takes three to six months to build, validate, and deploy — because the team is small, overloaded, or lacks specific expertise — the business suffers. Credit products launch late. Fraud detection lags behind new attack patterns. Marketing campaigns run on stale propensity scores.
Model quality ceiling. A small team under pressure doesn't have the bandwidth to run exhaustive experiments, test multiple algorithms, or continuously monitor performance after deployment. They build one model, validate it minimally, deploy it, and move on. The model works — but it's rarely the best possible model. In credit risk, the difference between a good model and a great one translates directly into approval rates, default rates, and profitability.
Key person dependency. In many institutions across Chile, Peru, Argentina, Colombia, and Mexico, the entire predictive modeling capability lives in one or two people. When one of them leaves — and they will leave — the institution faces a crisis. Models go unmaintained, drift goes undetected, and institutional knowledge disappears. Recovery time is measured in months.
How AI is changing the equation
The emergence of AI-native modeling platforms is beginning to shift this dynamic — but not in the way most vendors describe it. The common pitch is automation: AI does the work faster, so you need fewer people. That's true but incomplete, and it misses the more important change.
The real shift is capability amplification. A junior data scientist using the right AI-powered platform can now do work that previously required a senior practitioner with years of experience. Not because the AI replaces judgment — but because it encodes best practices, automates the repetitive parts, flags anomalies, and guides the analyst through the process in a way that raises the floor of what any team member can produce.
Concretely, this means a team of two or three analysts — which is the reality for most mid-sized banks and lenders across LATAM — can now run parallel experiments across dozens of algorithms, automatically select the best performer, generate regulatory documentation for the CMF, SBS, BCRA, BCB, CNBV, or Superintendencia Financiera, monitor model drift, and integrate alternative data sources — all without hiring five additional senior data science resources they couldn't find or afford anyway.
The data dimension
There's a second layer to this problem that's often missed: even when talent exists, the data often doesn't — at least not in the form needed to build good models. In Colombia, an estimated 40% of adults have no formal credit history. In Peru, that number is closer to 55%. In Mexico, the informal economy means millions of creditworthy people are effectively invisible to traditional scoring models.
Behavioral signals — app usage patterns, digital payment frequency, device characteristics, transactional data — contain meaningful information about financial behavior that traditional data sources miss entirely. Combining these signals with internal data can improve predictive accuracy by 20% to 150% compared to models trained on internal data alone.
Platforms that arrive with alternative data already integrated, validated, and ready to use for the specific realities of these markets eliminate that burden entirely. The team focuses on the model, not the plumbing.
What world-class looks like — without a world-class team
The best data science teams in the region — at Nubank in Brazil, Mercado Libre across LATAM, or the major global banks operating in Mexico and Colombia — have something that most institutions don't: scale, resources, and the accumulated experience of hundreds of models across millions of customers. That gap used to be insurmountable.
That's changing. The right combination of automated modeling, embedded best practices, pre-integrated alternative data calibrated for LATAM realities, and AI-guided decision support compresses the gap significantly. A small, capable team with the right platform can now produce models that perform at a level that would have required a team three times its size just three years ago.
The talent gap across Chile, Peru, Argentina, Brazil, Mexico, and Colombia is real, structural, and slow to close on its own. But its consequences are no longer inevitable. The institutions that recognize this earliest will have a meaningful and durable advantage.