Where do most analyzed organizations stand?

80.9% remain at individual use and 9.3% report isolated pilots. Only 3.4% show evidence of operational adoption or institutional scale.

Does employee use of ChatGPT mean the company has adopted AI?

No. Use becomes adoption when it is part of a recurring workflow with an owner, real data, a business metric, and review rules.

What separates a pilot from operational adoption?

A pilot proves something can work. Operational adoption proves it can run repeatedly with real users, integration, ownership, measurement, and exception handling.

Why does regional data show high interest while adoption remains low?

They measure different things. Regional traffic captures access and tool consumption; the first-party benchmark looks for evidence that AI changed an organizational process.

Which problems appear most often?

Manual operations, coordination costs that grow with the business, fragmented information, unclear prioritization, and individual use without shared standards.

Is this benchmark representative of every company in Latin America?

No. It is a self-selected sample of people and organizations interested in AI. It is useful for identifying patterns and asking better questions, not for estimating the entire regional economy.

What is the first step to move up one level?

Choose one concrete workflow, measure its current state, and assign an owner before selecting technology or expanding a pilot.

Original benchmark · updated July 2026

Using AI is not the same as adopting it

By Alex Torrenegra and Alan Arguello

Torrenegra & Co is part of the OpenAI Partner Network; this report also reflects learnings from our certifications.

The central finding

The region is not blocked by access. It is blocked between trying tools and redesigning work: choosing a workflow, assigning an owner, measuring a baseline, and integrating the solution with real data and systems.

80.9%

remain at individual use

9.3%

report isolated pilots

3.4%

show operational adoption or scale

Where organizations stand today

We classified declared evidence into five levels. The level is not defined by how many tools are used, but by how deeply AI is integrated into a repeatable way of working.

Aggregate distribution by level

80.9%

Individual use

People use assistants, copilots, or automations for their own tasks.

Observable evidence: There is personal productivity, but no shared workflow, owner, or common metric.
Typical bottleneck: Too many possibilities and little clarity about which workflow to change first.
Useful next move: Choose one workflow and document its current time, cost, errors, or conversion.

The sample is self-selected and directional. It describes people and organizations that already expressed interest in AI; it is not a regional census.

Latin America shows interest before operational depth

Public data explains why adoption can look more advanced. The region consumes AI intensely, but usage concentrates on ready-made solutions while the layers required to build proprietary workflows remain smaller.

Explore regional signals

Global visits to AI solutions from LAC14%

Global internet users in LAC11%

The region generates a larger share of traffic to AI solutions than its share of global internet users. Curiosity and demand for access are real.

ILIA 2025 · ECLAC and CENIA

This is not a contradiction

High attention, user growth, and AI content can coexist with few companies changing operations. Traffic measures access and consumption; this benchmark looks for evidence of organizational integration.

The gap appears inside the organization

An additional comparison with software companies in Colombia shows three frequently missing conditions. These are independent gaps, not maturity levels, and they can coexist.

35%

do not have a clear adoption strategy

33%

lack clarity on use cases

31%

do not measure the impact of current experiments

Tools change quickly. Without priority, ownership, and measurement, organizations accumulate experiments without knowing which ones to sustain.

The questions change, but the blockers repeat

Across open responses, sessions, and cross-industry conversations, the same patterns sit behind apparently different questions.

Explore a pattern

How it appears: Manual follow-up, tasks moving across people, information copied between systems, and reporting that consumes much of the week.
What it usually means: The process exists in practice but is not defined well enough for end-to-end automation.
First useful metric: Cycle time, rework, or cost per transaction.

Four conclusions behind the benchmark

Access is not adoption

Tools have reached individuals. The pending step is turning that use into shared workflows with owners, data, and quality criteria.

A visible process comes before good automation

When inputs, decisions, exceptions, and outcomes are unclear, AI usually accelerates confusion or automates rework.

Economic pain appears as coordination cost

The best opportunities keep growth from requiring more people to chase information, prepare reports, or move tasks between systems.

Sequence matters more than the tool

The strongest pattern is process, owner, baseline, integration, and control. Changing the order creates attractive demos with little operational impact.

A quick test of real adoption

Mark what exists around one concrete use case. Do not rate the company in the abstract; think about a workflow the team is already trying to improve.

This tool does not send or store your answers.

A recurring workflow is clearly described.One person owns the outcome of the workflow.The baseline and target metric are known.The solution uses real data and systems rather than demo-ready information.Review rules, exceptions, and enough documentation are in place to sustain it.

Signals present

0/5

Use or exploration

There is a tool or idea, but not yet an operating system around it.

The path to move up one level

It does not begin by buying another platform. It begins by reducing one operating uncertainty at a time.

01
Process
Choose a frequent, expensive, or slow workflow.
02
Owner
Assign who is accountable for the outcome.
03
Baseline
Measure current time, cost, errors, or conversion.
04
Integration
Connect real data, systems, and users.
05
Control
Define review, exceptions, and criteria to scale.

Methodology and limitations

This analysis anonymously consolidates and deduplicates forms, open questions, events, sessions, and executive conversations from Torrenegra & Co between March and July 2026. Declared evidence was mapped to one maturity rubric, and ambiguous responses were excluded from the benchmark.

Qualitative findings were triangulated across sectors and compared with current public research from ECLAC, CENIA, and the Digital Development Observatory. Public and first-party data are presented separately because they measure different phenomena.

The sample is self-selected and directional, not representative of all Latin America. No names, companies, emails, identifiable quotes, or private source counts are published.

Public comparison sources

Suggested citation

Torrenegra & Co (2026). AI adoption benchmark in Latin America. July 2026 update. https://www.torrenegra.com/en/insights/ai-adoption-latin-america