How businesses are coping with AI in 2026: plenty of use, few returns yet
Most companies now use AI somewhere. The latest surveys show far fewer can point to profit from it, and what the companies that do get returns have in common.
Most companies now use AI in some form. The harder question is what it has done for the P&L. The 2026 surveys give a mixed answer: people feel faster, but very few firms can show the money.
Most firms use AI. Few can show the profit yet
McKinsey's State of AI in 2026, “On the road to ROI” reports that 80% of respondents say AI has improved their own productivity. Yet only 37% attribute at least some EBIT impact to AI, about the same share as last year, and high performers (those who credit AI with 5% or more of EBIT) stay flat at about 6% of respondents.
PwC's 2026 Global CEO Survey is blunter. 56% of CEOs say AI has brought no significant financial benefit so far. Only 12% report both cost and revenue gains, and 33% report gains in one of the two.
Agents are the next step, and most firms are still testing
McKinsey finds 62% of organisations at least experimenting with AI agents, and 23% scaling an agentic system in at least one function. Large companies scaling agents rose from 27% to 40% in a year. Smaller organisations stayed flat at 22%, which is a gap a small team can use.
What the firms that get returns do differently
PwC's CEOs who report both cost and revenue gains are two to three times more likely to have embedded AI widely across products, demand generation and strategic decisions. Those with strong foundations (a responsible-AI framework and technology that lets AI spread across the company) are three times more likely to report meaningful financial returns.
Our reading
The pattern points to workflow, not licences. A tool bought for a whole company does little until one process is rebuilt around it: a named owner, a number that should move, and time for people to learn the new way. Small teams can often do this faster than large ones, because there are fewer people to bring along.
A practical test before you buy
Pick one process and write down how many hours, errors or sales it produces today. Add the tool for that process only. Check the same numbers after 90 days. If they have not moved, the problem is usually the process, not the model.
Small team, one process to automate?
See our products