Enterprise AI ROI Survey Finds Faster Work, Slower Transformation

Speed has arrived. Reinvention hasn't.

Deloitte's newest enterprise AI survey describes an industry that has stopped asking whether the technology works and started asking what it is actually worth. The firm's latest edition of The State of AI in the Enterprise, published on October 29 under the subtitle "Built Differently," surveyed 3,500 leaders from director level up to the C-suite. The respondents came from 25 countries and six industries, and all had direct involvement in their companies' AI programs.3 Fieldwork ran in May and June 2026.3

The main finding is a split. On one side, 81% of companies say AI has sped up decision-making and execution.3 On the other, only 25% say they use AI to deeply transform the business, meaning new products or services, reinvented core processes, or changed business models.3 Deloitte calls this "the new AI bottleneck": work gets done faster, but it is not getting done differently.3

The most quotable number is about money. In this edition, 28% of companies report revenue gains from AI, up from 20% in Deloitte's January report.3 Deloitte reads that as a sign that financial value is becoming more solid. It also says the companies seeing revenue gains report wider benefits beyond productivity and cost savings, which suggests the payoff comes from working smarter as well as faster.3

How the numbers have moved

The new results make more sense next to the earlier 2026 edition. That report, called The Untapped Edge, surveyed 3,235 leaders in 24 countries between August and September 2025.13 Two-thirds of respondents (66%) reported productivity and efficiency gains. In descending order, the other reported benefits were better insights and decision-making (53%), lower costs (40%), better customer relationships (38%), and revenue growth (20%).21 Revenue was mostly something companies hoped for: 74% wanted AI to grow revenue, but only 20% said it already did.2

On that baseline, an eight-point rise in revenue gains over about nine months is a real change. A large gap between ambition and results remains, though. Roughly seven in ten companies still don't report revenue from AI, while about three-quarters hoped to.

The January report also split companies into three roughly equal groups. About 34% were using AI to deeply transform, 30% were redesigning key processes around it, and 37% were using it at the surface with little or no process change.2 The October release gives 25% for deep transformation.3 The samples differ in size and country mix, and the two reports may not define the category the same way. Still, the direction counts against any story of steady progress. It looks like respondents became more cautious about what counts as transformation once they had more deployment experience. When AI is everywhere, claiming you have reinvented the business requires more evidence.

The cost problem nobody is measuring

The most useful new detail for practitioners is about cost visibility. Leaders can now see their total AI bill, but many cannot see what is driving it. Seventy-one percent track total inference and API spending, yet only 51% track token volume and 44% track per-unit inference cost.3

This matters more as agents spread. An agent that runs multistep workflows, calls tools, and loops through reasoning steps uses compute in ways that are hard to predict from a monthly invoice. If a company can't price a single unit of AI work, it can't easily tell whether a given use case pays off.

The Deloitte AI Institute's mid-year pulse check, which polled almost 3,700 professionals, found the same pattern. Most organizations can measure what AI costs, but far fewer can measure what it is worth.6 That report warned that companies judging AI only by cost savings will struggle to justify more spending as boards ask for evidence of strategic, not just operational, value.6 Another compilation of 2026 survey data shows the gap is not unique to Deloitte. It cites KPMG findings that only 12% of organizations consistently weigh AI value against cost, and notes there is no published benchmark for the fully loaded cost of a completed workflow.9

Organizational change is the real constraint

Deloitte names companies' capacity to redesign work and governance as the main obstacle to getting value from AI.3 Its other 2026 research points the same way.

In an August survey of more than 500 US business and IT leaders, Deloitte found that 74% expect nearly half their business processes to be redesigned or rebuilt around AI agents within four years. Only 5% said their processes were highly prepared for agents, and just 15% had scaled, orchestrated multi-agent systems.14 Only one in five said their organization was ready to redesign processes for autonomous operation. Many were instead adding agents on top of existing workflows to chase quick returns.14 The most common obstacles were data that isn't unified or accessible (72%), lack of confidence in governing agents (70%), and integration that is too costly or complex (67%).14 Even among companies that have scaled agents, only 46% said their processes were prepared.16

Governance follows the same pattern. The January report found that close to three-quarters of companies planned to deploy agentic AI within two years, but only 21% had a mature model for governing agents.11 Deloitte also argues that companies whose senior leaders actively shape AI governance get much more business value than those that leave it to technical teams.1

Read together, these studies tell a consistent story. The technology now delivers speed reliably. What separates the 25% who transform from everyone else is mostly management: whether a company will redesign workflows, set limits on what agents can do on their own, and decide who is accountable for the results.

Where the reporting diverges

Deloitte's numbers have always looked better than some outside measures, and the gap should be stated plainly. An analysis of the January edition compared Deloitte's finding that 74% of companies' most advanced initiatives met or beat ROI targets with MIT NANDA's widely cited claim that 95% of generative AI pilots fail.7 It explained the difference by method. Deloitte asks leaders about their flagship projects against targets they set themselves, while MIT counted every pilot against a stricter standard of rapid revenue growth.7 The same analysis pointed to self-reporting bias and to targets that were lowered after the 2024–2025 pilot cycle.7

Deloitte itself has acknowledged the tension. Its June research on technology investment found that 84% of AI investors reported gains, then pointed to MIT's finding that only 5% of pilots deliver lasting value at scale. It said organizations may be overestimating their success.5 Other benchmarks are lower still. KPMG reported that just 7% of leaders have established ROI, and McKinsey found that about 6% of respondents attribute at least 5% of EBIT to AI.9

The sensible reading is that none of these figures is wrong. They measure different things. Deloitte's 28% revenue figure records what leaders believe about their own organizations, and its sample leans toward companies already active in AI.13 That makes it a useful directional signal but not an audited result.

Timelines, not verdicts

Deloitte presents the ROI picture as returns arriving at different speeds, with financial gains taking longest.3 That fits its October 2025 survey of 1,854 executives, which found that most organizations needed two to four years to get satisfactory ROI from a typical AI use case. That is much longer than the seven to 12 months usually expected for technology investments.8 In that same survey, only 10% of agentic AI users said they were seeing significant ROI.8

In that light, the rise in revenue gains to 28% looks like the early payoff from projects started during the first generative AI wave. Speed improvements show up within months, process redesign takes years, and revenue follows the redesign.

The takeaway

"ROI reckoning" is a fair description, though the real story is less about spending than about how organizations operate. Investment has not slowed. Earlier this year, 84% of leaders said they were increasing AI spending.4 The reckoning is that the easy gains are mostly used up. Most companies have already captured the speed. The next gains depend on unit-level cost accounting, rebuilt workflows, and governance that can keep up with autonomous agents, and the latest survey suggests most enterprises are not ready on any of these yet.

The companies to watch are the minority already reporting revenue. If Deloitte is right that they are working smarter as well as faster, the gap between them and everyone else is likely to grow before it shrinks.