Using AI has become normal. Making money from it, not yet. That is the honest summary of the latest numbers, and it helps decide where to put your first dollar.
What the survey shows
In August 2026 McKinsey published the yearly edition of The state of AI, with 1,719 participants from 97 countries surveyed between May and June 2026. A few numbers that matter if you run a smaller company:
- Nearly nine in ten respondents say their organization uses AI regularly in at least one function.
- 44% say AI is scaling across the enterprise, up from 38% a year earlier.
- Only 37% attribute any EBIT (operating profit) impact to AI, about the same as the year before.
- At smaller organizations (under $1 billion in revenue), the share scaling AI agents stayed flat at 22%, while at large ones it rose to 40%.
- About 20% say AI operating costs, tokens included, limited their use.
In Brazil, Sebrae’s survey on digital transformation in small businesses, released in June 2025, found 44% of owners saying, unprompted, that they had used some kind of AI. When the question named specific tools, numbers went up: 51% use text tools like ChatGPT, Gemini or Copilot, and 41% use WhatsApp chatbots (Agência Sebrae, in Portuguese).
Why so much adoption and so little result
The data does not give a single answer, but there is a pattern. Individual use (someone on the team opening ChatGPT to write an email) is easy and widespread. Results that show up on the books need AI inside a process, with a start, middle and end, measured before and after.
That is the difference between “my team uses AI” and “the order that comes in through the website gets read, sorted and entered into the system with nobody typing”.
A roadmap for your first automation
1. Pick the most repetitive task of the week
Good candidates share three traits: they happen often, follow a pattern and today someone copies and pastes between systems. Email and request triage, reading standard documents, hand-built reports and the same questions on WhatsApp are common examples.
2. Measure before you change anything
How many hours does the task take per week? How many mistakes happen? Without that number you cannot tell whether the automation paid off, and you land right in the group that uses AI without seeing impact.
To estimate what those hours cost, enter your team’s figures in the manual work calculator.
3. Use AI only where it is needed
Language models are great at reading, summarizing and sorting free text. For the rest of the flow (moving data, sending notifications, applying fixed rules), plain automation is usually cheaper and more predictable, because it does not call a model at every step. Remember that token costs held back one in five McKinsey respondents.
4. Keep a person in the loop where mistakes are expensive
AI makes mistakes. Put human review on sensitive steps (amounts, deadlines, customer data) and let AI run alone where errors are easy to spot and fix.
5. Start small and grow from results
One automation that is well built, measured and working beats a big project stuck in pilot. With the first result in hand, the second choice is much easier to justify.
Sources
- The state of AI in 2026: On the road to ROI. McKinsey & Company (Dan Tinkoff, Lieven Van der Veken, Michael Chui), 2026-08-25. accessed October 2026.
- De GPS a Chat GPT: maioria dos pequenos negócios abraça a inteligência artificial para aumentar resultados. Agência Sebrae, 2025-06-25. accessed October 2026.