McKinsey 2026: Almost Everyone Uses AI, Few See It in Their Profits

Almost every organization now uses artificial intelligence. Very few can show that it makes a difference to their profits. That, in one sentence, is the most uncomfortable takeaway from McKinsey’s annual report on the state of AI, published on August 25, 2026.

For business and IT decision makers in Latin America, this matters for a simple reason: the question is no longer whether to adopt AI, but how to avoid ending up in the group that invests, experiments, and never moves the financial needle.

What McKinsey measured and what it found

The 2026 “The state of AI” study is based on 1,719 responses from 97 countries, collected between May 4 and June 8, 2026. These are the figures worth keeping at hand:

Indicator	Result
Use AI regularly	88%
Are scaling AI	44%
Attribute at least some EBIT impact to AI use	37% (roughly flat vs. 2025)
Qualify as “high performers”	6%
Companies with more than US$1 billion in revenue scaling AI agents	40% (up from 27%)
Decided against buying at least one software product because they built it in-house with agentic coding tools	32%
Expect AI to reduce headcount next year	39%
Saw a headcount decline last year	14%
AI users reporting better personal productivity	80%

Read together, these numbers tell a clear story: adoption is settled; financial value is not.

The gap: massive usage, stalled financial impact

Going from 88% regular use to 44% scaling already means stopping halfway. But the most revealing figure is another one: only 37% attribute at least some EBIT impact to AI, and that share was roughly flat compared with 2025. Another year of investment and pilots did not translate into more organizations able to see the effect on operating results.

At the same time, 80% of AI users say they are more productive personally. That distance between individual productivity and company-level impact is the heart of the problem.

Someone drafting an email or a report faster does not, on its own, change a company’s margin. For savings to reach the income statement, someone has to redesign the process, reallocate the freed-up capacity, and measure the effect.

What sets the few who capture value apart

Only 6% of respondents qualify as “high performers.” Beyond the report’s precise definition, there are common-sense practices that separate AI projects that move business metrics from those that stay stuck in demos:

• They start with a process, not a tool. They pick a concrete workflow — quotes, reconciliations, request handling — whose cost or cycle time can be measured today.
• They define the metric before the pilot. Without a baseline, there will be no way to attribute EBIT impact later.
• They redesign the work, not just speed it up. Individual productivity becomes value when roles, steps, and responsibilities change.
• They solve data and integration. An AI that can’t read from or write to core systems — ERP, transactional core, legacy databases — stays isolated.
• They have a business owner. The project answers to a unit with a budget and targets, not just to the technology team.

Build or buy: a shift worth watching

One of the most interesting findings for IT teams: 32% of respondents decided against buying at least one software product because they built it in-house with agentic coding tools.

It signals that, in some cases, the cost of building has dropped enough to compete with a license. But it’s worth not romanticizing it. Every in-house product brings maintenance, security, upgrades, and knowledge that must be preserved when the team changes. Building with AI doesn’t eliminate technical debt; it can accelerate it if there’s no architecture and no testing.

A practical rule of thumb:

• Build what differentiates your business or what no product handles well for your local operation.
• Buy what is commodity and what requires certifications or support you don’t want to sustain internally.
• Revisit the decision periodically: the balance between both options is shifting.

AI agents and jobs: expectations ahead of reality

Among companies with more than US$1 billion in revenue, 40% are already scaling AI agents, up from 27%. Agents — systems that carry out multi-step tasks with some autonomy — are leaving the lab in the largest organizations.

On jobs, expectations are running ahead of what has been observed: 39% expect AI to reduce headcount next year, while 14% saw a decline last year. For an executive, that suggests caution with plans that take cuts for granted: savings only materialize if the process has actually changed.

A signal from customer service

As reported by Customer Experience Dive, a Gartner survey of roughly 200 customer service and support leaders, conducted in April and May 2026 and released on August 26, points in the same direction. These leaders’ AI spending is up 38% and technology spending 15%, while overall service budgets rise just 2%. On average, AI already accounts for 13% of those budgets.

The implication is obvious: money for AI is coming out of other lines. Gartner cautions that organizations shouldn’t deploy AI to the detriment of the human agents. It’s a useful reminder for any team that thinks of AI only as a replacement.

Checklist for companies in Latin America

1. Take an honest inventory: list active AI initiatives and how many have a business metric with a baseline.
2. Prioritize two or three processes with measurable impact on cost, revenue, or cycle time.
3. Check integration: can the AI access data from your core systems, including legacy ones?
4. Define governance: who approves, which data can be used, and how results are reviewed.
5. Assess build vs. buy case by case, including the cost of maintaining what you build.
6. Plan for people: training and role redesign before assuming reductions.
7. Measure at 90 days and decide: scale, adjust, or shut down.

Frequently asked questions

If 88% already use AI, am I late?

Not necessarily. The relevant figure is that only 37% attribute some EBIT impact to it. Arriving with a method at a few well-measured use cases can be worth more than piling up pilots.

Should we build software in-house with agentic tools?

For some products, yes: 32% of respondents have already skipped buying at least one that way. But you need to budget for maintenance, security, and knowledge continuity.

Will AI reduce my headcount?

39% of organizations expect it to, but only 14% saw a decline last year. The outcome depends on how you redesign your processes, not on the tool.

Conclusion

McKinsey’s report confirms that AI is already part of daily work at almost every organization, but also that profit impact remains the exception. The difference isn’t access to the technology; it’s connecting it to real processes, data, and metrics.

If you want to find where AI can move the numbers in your operation, explore our AI for business approach or let’s talk. And if your data lives in older systems, legacy system modernization is often the first step.