THE DATA
The data on AI transformation and adoption
Independent research keeps arriving at the same conclusion: AI spreads through organizations quickly and pays off rarely. What separates the two groups is not which model they picked, but whether the work itself was redesigned. This page collects the three numbers that describe that picture, with their sources and with what each one actually means.
In short
- 88% of organizations use AI in at least one function; only 39% can attribute any EBIT impact to it.
- Only 21% of organizations using generative AI have redesigned a single workflow around it.
- Gartner expects more than 40% of agentic AI projects to be cancelled by the end of 2027.
- The common thread: the value comes from redesigning the process, not from adopting the tool.
Organizational AI use has more than quadrupled in eight years. In 2017 one organization in five used it; today nearly nine in ten do. The question is no longer whether it is in use, but whether it returns anything.
88%
Organizations using AI in at least one business function
39%
Organizations able to attribute an EBIT impact to it
21%
Organizations that redesigned a workflow around it
40%+
Agentic AI projects expected to be cancelled by end of 2027
Share of organizations reporting AI use in at least one business function, 2017–2025.
McKinsey, The State of AI (2025)Three numbers that describe the picture
All three come from independent research; every figure names the study it was taken from.
- 88% → 39%
- 88% of organizations now use AI in at least one function, but only 39% can attribute any EBIT impact to it. Adoption is not the bottleneck; measurable change is. When the tool is live but the way the work is done stays the same, the minutes saved never reach the books. McKinsey, The State of AI (2025)
- 21%
- Only 21% of organizations using generative AI have redesigned a single workflow around it — and of every organizational change studied, workflow redesign correlates most strongly with profit impact. This is what distinguishes the companies that can point to a return. McKinsey, The State of AI (2025)
- 40%+
- Gartner expects more than 40% of agentic AI projects to be cancelled by the end of 2027, driven by escalating costs, unclear business value, and inadequate risk controls. None of the three is a technology problem; all three are scope and measurement problems. Gartner, June 2025
Adoption is wide, profit impact is not
a 49-point gap
The minority that redesigned the work
How projects that start wrong end
by the end of 2027
What the data is saying
Read side by side, the three numbers tell one story: adoption is not a technology problem.
Three ratios on one scale
of all organizations
of all organizations
of organizations using generative AI
There is a 49-point gap between using AI and gaining from it
Forty-nine points separate the organizations that have deployed AI from the ones that can show a return on it. That gap does not open between pilot and production — it opens inside the process. If the same approval chain, the same handoffs and the same reporting survive, the minutes saved never add up anywhere.
What the winning group shares is redesign, not tooling
Of every organizational change studied, redesigning the workflow correlates most strongly with profit impact. The question is not which model to use, but what this work would look like if it were built from scratch today. Few organizations ask it — and that is where the returns concentrate.
Cancelled projects rarely fail technically
The three causes Gartner names — escalating cost, unclear business value, inadequate risk controls — are consequences of decisions made at the start. When nobody defines which process is in scope, how success will be measured, and where a human stays in the loop, the project gets shut down even when the system works.
How we answer it
Every finding maps to a concrete step in our method. This picture is why we run a two-week diagnosis before a line of code is written.
- AI is in use, impact is not measurable
- We measure the current state first: how long each step takes, how often work changes hands, what each error costs. What gets discussed afterwards is not a claim of improvement but the same measurement, taken again.
- A tool gets added, the process is left alone
- We redesign the process first, then decide which step needs automation or a model. Choosing the tool is one of the last decisions, not the first.
- Scope and risk controls are never defined
- Before anything ships, we put the scope, the points where human approval stays, and the success criterion in writing. Because the team that sets the strategy is the team that builds the system, that definition survives into the build.
The common failure is starting in the wrong place, not building badly. Finding the right starting point decides more than the system you end up building.
Read our methodCONTACT
Let's talk about your process
Tell us where you stand today and what you're trying to fix — we'll get back to you to find the right starting point.