AI returns depend on processes

, AI and governance

According to McKinsey & Company 80% of people say AI makes them more productive. Only 37% of companies see any EBIT impact, the same share as a year ago. That is the central finding of McKinsey’s State of AI in 2026.

So where does the value get stuck? Between the individual and the workflow.

The 6% of companies that do see real results (the report calls them high performers) do not spend their way there. They work differently:

  1. They redesign the workflow, they don’t add a tool to it. About three quarters of high performers rebuilt processes end to end. Among the others, it’s about one in four.
  2. They aim beyond cost. 74% target growth, against 47% of the rest. Efficiency alone rarely changes the P&L.
  3. They decide where the human stays in the loop. Which actions an agent can take alone, which need validation. High performers manage unauthorized action and explainability far more than their peers.
  4. They measure. They define upfront which number proves an initiative worked, and who tracks it.
  5. Leaders go first. Senior sponsorship and impact tracking are twice as common among high performers.
  6. They treat token cost as a design constraint. Usage grows faster than the price per token falls.

Fair enough. Yet, according to me, there are at least two more main item to value even before these pointed out. One is competence. There are no chances that any agent would replace a domain competence as the experts you have in your company. Another factor is time. Consider AI as an actor of your company and leave the organization enough time to get used to it.