For two years, AI adoption was measured using a question that has already outgrown its usefulness: are you using it or not? On October 6, the World Bank posed a different one. In its Latin America and the Caribbean Economic Review, it concluded that while businesses across the region are adopting AI broadly, they rarely integrate it into core process activities. Furthermore, the principal barriers to productive adoption are neither cost nor access, but rather managerial capabilities, workforce skills, and organizational capacity to restructure around these tools.

Two weeks later, on October 20, ECLAC and CENIA published the fourth edition of the Latin American Artificial Intelligence Index (ILIA), analyzing 126 sub-indicators across 20 countries. One of its ten key findings carries a headline that captures the essence of the week: "Agents are ready; organizations, not so much."

What both assessments describe has carried a specific name in management literature for 36 years.

In 1990, economists Wesley Cohen and Daniel Levinthal published a paper in Administrative Science Quarterly introducing the concept of absorptive capacity: a firm's ability to recognize the value of new external information, assimilate it, and apply it to commercial ends. Their central thesis presents an uncomfortable truth for any technology vendor: this capacity depends heavily on the organization's prior knowledge. A firm lacking foundational understanding cannot learn to effectively deploy what it buys.

In 2002, Shaker Zahra and Gerard George refined the framework into four distinct dimensions: acquiring, assimilating, transforming, and exploiting knowledge. The first two constitute potential capacity, while the latter two form realized capacity—the only dimension that actually drives business outcomes.

When applied to AI, the framework is straightforward. Purchasing the tool represents acquisition—the easiest step. Value materializes only when organizations transform their business processes and exploit what they have learned. This is precisely the gap currently being measured by both the World Bank and ECLAC.

The ILIA estimates that 27.6% of large companies in the region use AI, compared to 12.8% of medium-sized enterprises and 6.8% of small businesses. Across the five benchmark countries (Germany, Spain, Estonia, Portugal, and Singapore), those figures reach 66%, 41%, and 24%, respectively. ECLAC’s presentation placed the regional average overall at 9.1%. The World Bank adds another metric: a median of 17% of working-age adults report using generative AI—approximately half the rate observed in the United States and Canada.

It is critical to distinguish between two different realities. Those who build foundation models face a genuine computational bottleneck: the infrastructure concentration highlighted by the ILIA is unmistakable. However, the World Bank itself asserts that Latin America does not need to build frontier models to capture the vast majority of AI's economic benefits, advocating instead for "small-scale AI"—low-cost applications tailored to localized operational needs. The ILIA, in turn, redefines AI sovereignty not as complete self-sufficiency, but as the capacity to act strategically within global interdependence.

For a bank, an insurance firm, or a healthcare provider—entities that are consumers of AI rather than foundation model developers—the strategic takeaway is entirely practical. The primary constraint is not acquiring the technology itself; it is how much of that technology the organization can absorb, govern, and convert into bottom-line business outcomes.

Regulators are already translating this concept into formal mandates. On October 1, the Superintendent of Banks of Panama, Milton Ayón, stated that it is no longer enough to prove that an automated tool yields accurate results: banks must be able to demonstrate how it reached a given conclusion, what data it relied upon, and what operational oversight governs its execution.

In Colombia, the bill introduced in September to mitigate systemic risks arising from critical AI dependencies across banking, securities, and insurance sectors requires a similar level of rigor: maintaining an active inventory of system dependencies, stress-testing failure modes and corrupted output scenarios, and maintaining fully validated business continuity and exit strategies. While still a legislative proposal rather than enacted law, it outlines, article by article, precisely what an organization with mature absorptive capacity must be equipped to answer.

If absorptive capacity is the prerequisite, the core question for a Chief Technology Officer becomes: which capabilities build it? At Q-Vision—a company with over 22 years of experience driving the intelligent transformation of organizations—we group these capabilities into three core pillars:

• Digital Product Engineering: Converts AI into functional products through agentic software lifecycle development, quality intelligence to rigorously test what is built, and traceable data architectures that an organization can reliably leverage. This directly addresses the prerequisites of strategic clarity, data readiness, and performance measurement.

• Cloud: Modernizes and optimizes the underlying infrastructure hosting AI models—focusing on cloud migration, security by design, and cost optimization. This fulfills the infrastructure prerequisite, which, for enterprise AI consumers, means effectively selecting and governing tech dependencies rather than owning massive compute power.

• Human-Agent Talent: Integrates specialized human talent with AI agents into high-performing hybrid teams governed by defined roles and oversight structures. This addresses the human capital and governance requirements.

None of these three capabilities is sufficient on its own. A digital product built without governed infrastructure cannot scale; an infrastructure lacking human oversight cannot be sustained; and a hybrid team operating without structured product frameworks or impact measurement cannot demonstrate business value. Absorptive capacity is the thread that binds them, and verifiable performance is the proof that it works.

The World Bank and ECLAC have fundamentally shifted the question. The benchmark is no longer how many companies have adopted AI, but how many have absorbed it: how many can explicitly demonstrate what models they run, what data fuels them, who supervises their behavior, and what bottom-line outcomes they deliver.

In the words of the World Bank's acting Chief Economist for the region, Carlos Rodríguez-Castelán: "The question is whether the region will be able to use it productively." Organizations capable of answering that question with empirical evidence will not need to justify, further down the line, why their AI adoption failed to translate into measurable business results.