Designing AI based on local and representative data is becoming a key strategic advantage for organizations.

The starting point of artificial intelligence is data, its essential fuel. The raw material that must be curated, organized, and structured so AI can improve business processes, operations, and decision-making.

But once AI begins operating within a company, institution, society, or country, it also changes how that data—this “moment zero”—is created. This continuous feedback loop, this circular nature of innovation, is especially evident in Latin America, a region where cultural diversity, structural inequalities, and business heterogeneity shape how technology is adopted and scaled.

In this context, inclusion in AI can’t be limited to detecting or correcting algorithmic bias. The challenge is far more structural: what data feeds the models, from which geographies, in which languages, and under what cultural assumptions. Designing AI with context in mind is no longer just an ethical important, it's a strategic decision that directly affects organizational competitiveness.

From “Translated” AI to rooted AI

Latin America is home to more than 470 million Spanish speakers and 220 million Portuguese speakers, along with multiple dialectal variations, and Indigenous communities with their own languages. Yet, according to Stanford’s AI Index 2024, only 3% of the datasets used to train AI language models originate from Latin America.

This results in what can be described as “translated AI” solutions that function technically, but interpret reality through external frameworks. Much of the AI currently used in the region wasn't built for its unique contexts; instead, it was adapted from different ways of speaking, consuming, producing, and engaging with governments, businesses, and technology.

As a result, models may fail to fully understand local expressions, informal economies, hybrid social dynamics, or fragmented regulatory environments, leading to outputs that are less useful, less reliable, and, often, less equitable.

Data as strategic infrastructure

At the beginning of this article, I defined data as AI’s “moment zero.” But it's important to go further: data isn't neutral. Its quality, representativeness, and governance directly determine the kind of AI that can be built.

In the global AI conversation, the focus is gradually shifting from model performance to the real-world conditions that make AI work in practice, as highlighted in Stanford’s AI Index Report 2025. Data quality, governance, transparency, and alignment with regulatory and social contexts are becoming critical enablers of sustainable adoption.

For companies across the region, this means that training models on local data requires solving how data is captured, integrated, updated, and protected over time often within fragmented system landscapes and under specific regulatory frameworks.

Managing models with contextual variants

Effectively managing AI models in Latin America requires moving beyond the idea of a single “optimal” version and toward architectures capable of contextual adaptation.

The key question becomes: how do organizations design local benchmarks that allow them to adjust language, rules, and signals based on specific environments?

Recent research in language models shows that incorporating regional variation not only improves inclusivity but also enhances system performance and interpretability. In other words, contextualization isn't an added cost, it's a functional upgrade. The value of AI is no longer measured solely by technical sophistication. It also depends on its ability to operate with trusted, governed, and representative data.

Every interaction with customers, suppliers, and communities generates data that when properly managed can help build more context-aware models. Well-governed data enables personalized services that reflect real user diversity, instead of forcing behaviors into global averages that don’t exist in practice.

The contextual advantage

Thinking about inclusion through the lens of AI means to rethink the relationship between business and society. Latin America is at a unique inflection point. Despite structural challenges, it combines talent, creativity, and quick technological adoption. If organizations can strengthen data sovereignty and governance, the region has a real chance to use AI as a driver of competitiveness, productivity, and gap reduction.

Contextual data can enable more relevant AI—AI that, in turn, improves how data is used, creating a good cycle that benefits businesses, institutions, and society.

Designing AI with context in mind doesn't slow down innovation, it makes it sustainable in the end. The true competitive advantage lies in building intelligence that can adapt, learn, and evolve alongside businesses and the societies they serve.

How to scale

For companies in Latin America, the challenge isn't defining AI inclusion principles, it's operationalizing them within real-world data architectures, often fragmented and legacy-driven. This starts with a clear decision: treat data as a strategic business infrastructure, not as a byproduct of systems.

In practice, truly contextual AI is built long before the model itself. It begins with representative data at the source, data that can capture local signals without diluting them into global averages. It continues with strong governance frameworks that define what data is used, under which rules, and with what accountability. And it's reinforced by architectures that allow intelligence to adapt by market, language, or operational context without disrupting core business logic. That's why AI inclusion isn't solved by tuning algorithms—it's solved by making the right upstream decisions about data integration, quality, and governance. When these foundations are in place, contextualization stops being an added effort and becomes a natural outcome.