What separates companies that get results from AI
AI use has become widespread, but the share of companies that see its effect on operating results did not move in a year. The evidence points to management decisions, such as redesigning whole processes, training executives and assigning owners.
Executive summary
Nearly nine in ten organisations use AI regularly in at least one function, according to McKinsey's global survey published in August 2026. In one year the share applying it across several functions and the share scaling it across the enterprise both grew.
Financial results did not follow that expansion. The share attributing some EBIT effect to AI stayed about the same as in 2025, and the high-performer group remains small and stable.
That group stands out for redesigning work. Almost three in four of its members fundamentally transformed their workflows, against one in four among the remaining organisations.
In a randomised experiment with executives from 90 Chilean companies, firms in the group trained in AI applied to business adopted on average 1.37 more tools in production. Those with a formal AI role used two more tools.
Context
For three years, boardroom conversations about AI revolved around adoption, measured in functions, users and pilots. Today almost every organisation meets that indicator, so it does little to tell them apart.
The measure that separates them is slower and more demanding, and it asks whether usage shows up in operating results. For a mid-sized company in the region, that means choosing a few processes, measuring before and after and making someone responsible for the result.
Figure 1 · Organisations that fundamentally redesigned their workflows
% of organisations, 2026Table 1 · AI tools in production after training, GerencIA experiment
| Tool | Control group | Applied training | Difference |
|---|---|---|---|
| Business process automation | 50.0% | 83.3% | +33.3 pts |
| Recommendation systems | 12.5% | 50.0% | +37.5 pts |
| Predictive analytics for sales or demand | 50.0% | 66.7% | +16.7 pts |
| Computer vision | 25.0% | 33.3% | +8.3 pts |
| Natural language processing | 50.0% | 50.0% | no difference |
Business implications
Redesign one complete workflow before adding functions. It is one of the traits that set the high-performer group apart in McKinsey's survey. Choose a process with volume, a measurable cost and a clear owner.
Train executives and appoint an AI owner. In the Chilean experiment, targeted training translated into more tools in production, and companies with a formal AI role used more tools than the rest.
Measure on the income statement. Cost per transaction, cycle time or margin are verifiable and comparable over time. Active users help track adoption; investment is justified with outcome metrics.
Methodology & data
McKinsey & Company online survey, 4 May to 8 June 2026, with 1,719 participants from 97 countries. The GerencIA experiment, by CENIA and the IDB, randomly assigned executives to specialised or introductory training; its preliminary results are published in the ILIA 2025.
McKinsey's survey collects executives' answers and does not break out Latin America; the experiment works with a small sample.
References
McKinsey & Company (25 August 2026). The state of AI in 2026: On the road to ROI. · ECLAC and CENIA (2025). Índice Latinoamericano de Inteligencia Artificial (ILIA) 2025, box 5: impact analysis of GerencIA.