Deloitte's "2026 State of AI in the Enterprise" report was released this week, containing a set of numbers that give pause:
74% of enterprise executives expect AI to drive revenue growth. Those who have actually achieved it: 20%.
The 54-percentage-point gap is not due to a lack of investment, but rather the entire industry measuring the wrong things.
Efficiency is real, but hasn't translated into money
The report's most contradictory comparison of numbers:
| Metric | Percentage of companies reporting improvement |
|---|---|
| Efficiency and productivity | 66% |
| Decision quality | 60% |
| Revenue growth | 20% |
Efficiency has improved, decisions are getting better, but only one in five companies has actually seen changes in revenue on the books.
Why the gap? Fortune's analysis is blunt: most companies report AI value in terms of "hours saved," but CFOs want to see changes in the P&L.
Deloitte itself uses an internal AI assistant called Sidekick, saving employees an average of 2 hours per week. That sounds good when reported to senior management, but it's zero on the P&L.
Were those 2 hours saved converted into new products, new customers, or new revenue? Most companies don't track that step. This is the "efficiency trap": you measure what you can measure, not what you need to measure.
From experiment to production, twice as hard as expected
Another striking number:
- 54% of companies expected to move more than 40% of their AI projects into production within 3-6 months
- Actual achievement: 25%
More than half of companies overestimated their ability to deploy, or underestimated the gap between "POC works" and "production is reliable."
The report found one explanation: only 20% of companies expressed confidence in "talent readiness," while 42% believed their AI strategy was fully prepared.
The strategy is ready on paper, but the execution team hasn't caught up—this is one of the main forms of AI deployment failure today. It's not that the technology isn't good enough; it's that after AI is installed, no one knows how to run it.
But budgets are still increasing
Despite ROI being hard to quantify and deployment rates falling short of expectations, 84% of companies say their AI budgets will continue to increase or remain stable this year.
This shows that companies are betting on AI not based on proven ROI, but on their assessment of the future competitive landscape—fear of being left behind by peers if they don't adopt AI. This is defensive spending, not return-driven investment.
This model can hold for a year or two. But if ROI still can't be clearly quantified after three or four years, the political pressure on budget approvals will become increasingly difficult to resist.
One relatively positive signal: 25% of leaders say AI has already had a "transformative impact," double the 12% from a year ago. So the direction is right, just slower than what the PowerPoint slides promised.
Alternative measurement framework recommended by the report
Deloitte proposed metrics to replace "hours saved":
- Decision cycle speed (how much time is shortened from raising a question to making a decision)
- Customer interaction quality (NPS changes, customer service resolution rates)
- Time to market for products
- Employee satisfaction and skill growth
- Organizational capability accumulation
These are not smoke screens; they can indeed capture some of AI's value. But for CFOs who need to report to the board every quarter, these are stories, not evidence.
The core question for enterprise AI in 2026 has shifted from "can it be used" to "has it made money." Deloitte's numbers are blunt: most companies are still stuck on the left side of this equation, using the wrong ruler to measure something that is happening.
Sources: CocoLoop, The hidden ROI of AI: What leaders should actually measure (Fortune); Deloitte 2026 State of AI in the Enterprise Report