The short answer: Most organizations now use AI, but few capture meaningful financial return. McKinsey's August 2026 survey finds 44% scaling AI across the enterprise and 37% reporting some profit impact, while only about 6% qualify as high performers. What sets that 6% apart, per McKinsey, is that they redesign how work gets done.
Key takeaways
- Use is nearly universal; scale and profit impact are not. McKinsey reports 44% scaling across the enterprise and 37% seeing at least some EBIT impact.
- The studies measure different things, so their percentages cannot be compared directly. Read each as a direction, not a benchmark.
- The one finding that repeats across the research: value follows changes to how work is done, not the choice of tool.
- Gartner forecasts that over 40% of agentic AI projects will be scrapped by the end of 2027, citing escalating costs and unclear business value.
What four sources say
| Source | Published | What it found | Sample and definition |
|---|---|---|---|
| McKinsey, The state of AI | Aug 25, 2026 | 44% say AI is scaling across their enterprise (38% a year earlier). 37% attribute at least some EBIT impact (essentially unchanged). About 6% are high performers. | 1,719 participants, 97 countries, fieldwork May 4 to Jun 8, 2026. High performer = attributes 5% or more of EBIT to AI and calls the impact significant. |
| MIT NANDA, The GenAI Divide (via Fortune) | Aug 18, 2025 | About 5% of pilots achieved rapid revenue acceleration; the rest showed little to no measurable P&L impact. Vendor purchases and partnerships succeeded about 67% of the time; internal builds about one-third as often. | 150 leader interviews, a 350-employee survey and 300 public deployments. |
| Gartner (via Reuters) | Jun 25, 2025 | Over 40% of agentic AI projects will be scrapped by the end of 2027. Only about 130 of the thousands of vendors claiming agentic AI are real. | A forecast, not a measurement. Cited causes: escalating costs and unclear business value. |
| Cognizant, via Constellation Research | Feb 5, 2026 | CEO Ravi Kumar describes an "AI velocity gap": the disconnect between money spent on AI infrastructure and the business value it produces. | A leader's framing after Q4 2025 earnings, not a survey. |
How to read these numbers without fooling yourself
Each study defines success differently. McKinsey counts scaling and any EBIT impact. MIT NANDA, as reported, counts rapid revenue acceleration from pilots. Gartner forecasts cancellations. A 95% in one report and a 37% in another are not contradicting each other; they are answering different questions on different samples.
What survives that caution is a pattern. Adoption is broad, scaled deployment is partial, and financial payoff is concentrated in a small group. The useful question for a leader is not "which percentage is right" but "what do the few who capture value do differently".
What the high performers do differently
McKinsey's answer is specific. Nearly three-quarters of high performers report fundamentally redesigning workflows because of AI, up from 55% a year earlier. About a quarter of other respondents say the same. The high-performer share itself has not moved: still about 6%.
MIT NANDA's reporting points the same direction from another angle: empowering line managers rather than only central AI labs, and choosing tools that integrate deeply and adapt over time.
Reading the data through the Velocity Gap lens
This section is interpretation, not a finding of the studies above.
The Velocity Gap Framework argues that organizations keep rituals built for a time when execution was expensive: check before you do, get everyone aligned, plan before you build. When workflows have to change for AI to pay off, those same rituals are what slow the change. The eight habits it names, the Friction Defaults, map onto the redesign problem directly:
- Permission Loop and Structured Waiting stretch the time between an idea and a working version.
- Meeting Dependency, Consensus Lock and Deck Over Demo replace evidence with discussion.
- Planning Inversion, Polish Paralysis and Readiness Hoarding delay contact with real users and real data.
If redesigning work is what separates the 6% from everyone else, the habits that block redesign are the first place to look. For definitions of all eight, see the glossary; for the full argument, read The Velocity Gap Framework.
What to do in the next 30 days
- Pick one workflow, not one tool. Name the process that should change and who owns it. The research rewards redesigning work, not adding software.
- Define success before scaling. Gartner's cancellation forecast names unclear business value as a cause. Write the metric and the cost per outcome down first. The AI Agent P&L guide is built for this.
- Time your own approval chain. Compare how long it takes to get permission to try something with how long it takes to build it. If approval is longer, that is a Permission Loop.
- Set a promote-or-stop date. Pilots with no end date drift. Choose the date and the evidence that would make you scale or stop.
- Score where you stand. The free Velocity Gap Diagnostic rates your organization on the eight habits in about five minutes, and the pilot-to-production checklist turns each into a question your team can answer.
Frequently asked questions
What percentage of companies are scaling AI in 2026?
In McKinsey's survey of 1,719 participants in 97 countries (fieldwork May 4 to June 8, 2026), 44% said AI is scaling across their enterprise, up from 38% a year earlier. Only about 6% qualified as AI high performers.
Is it true that 95% of AI pilots fail?
The figure comes from MIT NANDA's 2025 report, as covered by Fortune: about 5% of pilots achieved rapid revenue acceleration while the rest showed little to no measurable P&L impact. It measures one specific outcome in one dataset, so treat it as a direction rather than a universal failure rate.
How is the Velocity Gap different from Cognizant's "AI velocity gap"?
Cognizant's CEO uses the phrase for the disconnect between AI spending and business value. The Velocity Gap Framework by Edward Chenard describes the gap between where the work bottleneck has moved and where organizational habits still guard execution capacity. One names an outcome; the other proposes a cause inside the organization.
Sources
- McKinsey, The state of AI (2026 survey, published Aug 25, 2026)
- Fortune on MIT NANDA, The GenAI Divide: State of AI in Business 2025 (Aug 18, 2025)
- Reuters via The Star: Over 40% of agentic AI projects will be scrapped by 2027, Gartner says (Jun 25, 2025)
- Constellation Research: Cognizant aims to solve "AI velocity gap" (Feb 5, 2026)