On October 15, 2026, the Oliver Wyman Forum released The Reality Check: Beyond the AI Hype, Hope, and Hysteria. Its sharpest ROI number describes a management decision. Only 8 percent of workplace AI users say their organization has a company-wide standard for what happens to the time AI saves. Another 54 percent say individual employees decide, or that nobody makes an explicit decision at all. Many AI business cases count hours saved first. A saved hour with no destination leaves that claim resting on time nobody has spent on anything the business can measure.
John Romeo, CEO of the Oliver Wyman Forum, put it in one line: “AI creates capacity, but this time isn’t a return until someone decides what to do with it.”
Basil C. Puglisi reaches the same point through a requirement he calls Evidence-Based Value Realization, set out in his paper What AI Transformation Actually Requires. A claim of added capacity, he writes, “calls for evidence of what that capacity produced.” The same section sets the test plainly: “Time saved counts as value only when quality, rework, and risk hold steady or improve over the same period.” His advice to a client follows from that. Before an AI program reports hours saved as a return, a named person decides in writing where those hours go, and the business measures what changed there.
Who the Oliver Wyman Forum surveyed, and what the numbers can carry
The report draws on two waves of the Oliver Wyman Forum Global Voices Survey, fielded in April (12,883 responses) and August (12,837) 2026 across 15 markets. It also uses the Oliver Wyman Forum and NYSE CEO Agenda 2026 survey of 415 CEOs, fielded January to March. According to the methodology page, Global Voices respondents were adults aged 18 to 75 surveyed online, about 96 percent of them employed, and results are “self-reported and descriptive unless otherwise noted.” Oliver Wyman also sells AI consulting through Quotient, AI by Oliver Wyman, whose client work informs the report, so the recommendations are a consultancy’s view built on a large survey base.
The saved-time question went to 9,833 workplace AI users: “When AI saves time in your job, who usually decides what happens to that time?” In the full report, 54 percent chose individual or no explicit decision and 31 percent said a manager or team decides. Another 8 percent pointed to an organization-wide standard, and 6 percent reported no saving or were unsure. The authors add a fair caution that these are employee reports, not organization-level incidence estimates. A company can hold a standard its people have never heard of, and that would be a finding about how far the decision travels.
Saved hours carry a checking cost the business case rarely counts
The report’s ROI chapter adds a number that complicates every hours-saved slide. Workplace AI users report spending a mean of 22 minutes checking, correcting, or redoing output for every hour of AI-assisted work, based on 7,845 people who gave an estimate. The authors note that the survey doesn’t establish how much of that checking is new compared with the pre-AI process. Even so, a gross hours figure that leaves out review time overstates what a team actually has available to redirect.
The chapter’s CEO data points the same way. Only 12 percent of CEOs report AI-driven cost reduction or revenue uplift above 10 percent. Those CEOs are 4.4 times as likely as CEOs who cannot yet point to returns to have deployed AI at scale in new products, services, and revenue streams (40 percent versus 9 percent). Among them, 52 percent prioritize redesigning roles and workflows, against 33 percent of CEOs without returns. The report labels these descriptive comparisons that do not establish causation, and that caveat belongs in any client deck that repeats them.
Oliver Wyman names three destinations for AI-created capacity
The report’s answer is an “ROI harvester,” a senior owner accountable from the outset for the intended business result, with end-to-end authority to make the decisions that capture it. Each destination for freed capacity brings a hard choice of its own, the authors argue. A cost case often lands on people, through slower hiring, lower contractor spend, or layoffs. A reinvestment case asks which higher-value work takes up the hours and who will check what it produces. A growth case depends on customers, and on whether business owners will change the offer, the pricing, the sales process, or the customer experience.
Puglisi treats the harvester’s choice as a checkpoint. Under Checkpoint-Based Governance (CBG), a named human holds authority to accept, modify, or reject at a defined point before work moves forward. Choosing where saved time goes fits that definition, because the choice commits money or headcount in ways that are hard to reverse. A usage dashboard that counts hours is Responsible AI in Puglisi’s terms, a machine checking a machine. The routing decision reaches AI Governance only when a named person accepts, modifies, or rejects it and the record shows which.
CEOs and their employees describe different destinations
A second survey sits awkwardly beside the first. In the EY-Parthenon CEO Outlook Survey released October 1, 2026, 48 percent of 1,200 CEOs across 21 countries said they are reinvesting AI’s additional capacity into innovation and transformation. Almost a quarter (23 percent) said AI-driven gains are often absorbed before they reach the bottom line, and only 16 percent reported clear, real-time visibility of ROI.
The two findings come from different people answering different questions, so neither cancels the other. CEOs may well be directing capacity at the portfolio level while the hour inside a single job drifts to whatever the employee chooses. Puglisi’s position is that this tension should stay in the open and be settled with evidence. A client that hears “we reinvest” from the top should ask the teams doing the work where last month’s saved hours went, and treat the gap between the two answers as data.
Silence about saved time pushes employees to hide their gains
The workforce numbers explain why the destination question can’t wait for a later phase. When Oliver Wyman asked employees what should happen if AI doubled their team’s output, they ranked “fewer people on the team” last. Asked what they expect will happen, they ranked it first. Sixty-two percent of daily AI users said they have played down how good they are with AI at work. Ana Kreacic, chief operating officer of the Oliver Wyman Forum, warned that if leaders don’t explain how freed-up time will be claimed, “employees may assume the goal is to eliminate their roles.”
Masked gains are a measurement failure as much as a morale one. An hour an employee hides never reaches the baseline, and a written destination gives people a reason to report it.
What Puglisi would put in place this quarter
For a client with AI already in daily use, the work starts with one workflow at a time. Measure gross hours saved against a pre-AI baseline, subtract the checking and rework time the team reports, and call the remainder net capacity. Then a named owner chooses the destination in writing: cost, reinvestment, or growth. The same owner names the business result expected to move, such as cost per resolved issue or cycle time. That choice is the checkpoint, and it’s recorded with the date and the reasons so a later reviewer can see who decided and why.
Two numbers belong on the executive dashboard. The first is the share of AI-assisted workflows with a written destination for net saved time, with a target of every workflow in the funded portfolio. The second is the business result attached to each destination, reported against its baseline at every review, with the hours figure shown beside it as supporting context.
Where the next AI return argument will happen
The Reality Check moves the AI return question from the tool to the person who decides what happens to the time the tool frees. That move leaves two problems open. Nobody has yet shown a reliable way to choose a cost destination without confirming the job fears that make employees hide their gains. The 8 percent figure suggests most organizations haven’t tried to choose at all. The next Global Voices wave will show whether that share moves, and the next round of CEO surveys will show whether executives start asking where the hours went as well as whether hours were saved. Companies that can answer the first of those questions will likely have an easier time with the second.
Sources
- EY. (2026, October 1). AI drives productivity, but CEOs struggle to turn gains into growth [Press release]. EY Newsroom. https://www.ey.com/en_gl/newsroom/2026/10/ai-drives-productivity-but-ceos-struggle-to-turn-gains-into-grow
- Fine, A., Kreacic, A., Ardeshna, A., Anderson, E., Czervionke, E., Romeo, J., & Luong, S. (2026, October 15). The return on intelligence (ROI). In The reality check: Beyond the AI hype, hope, and hysteria. Oliver Wyman Forum. https://www.oliverwymanforum.com/artificial-intelligence/agentic-ai-strategy-governance-workforce/roi.html
- Oliver Wyman. (2026, October 15). Rapid AI capability development is creating opportunities for those companies with leadership that translate productivity gains into business value [Press release]. Oliver Wyman Media Center. https://www.oliverwyman.com/media-center/2026/oct/rapid-ai-capability-development-is-creating-opportunities-for-those-companies-with-leadership-that-translate-productivity-gains-into-business-value.html
- Oliver Wyman Forum. (2026a, October 15). Methodology. In The reality check: Beyond the AI hype, hope, and hysteria. https://www.oliverwymanforum.com/artificial-intelligence/agentic-ai-strategy-governance-workforce/methodology.html
- Oliver Wyman Forum. (2026b, October 15). The reality check: Beyond the AI hype, hope, and hysteria [Report]. https://www.oliverwymanforum.com/content/dam/oliver-wyman/ow-forum/template-scripts/templates/ai-agenda/2026/pdf/The_reality_check.pdf
- Puglisi, B. C. (2026, September 21). What AI transformation actually requires. basilpuglisi.com. https://basilpuglisi.com/ai-transformation-requirements/
Questions readers ask
What did the Oliver Wyman Forum’s Reality Check find about time saved by AI?
Released October 15, 2026, the report found that only 8 percent of 9,833 workplace AI users say their organization has a company-wide standard for what happens to AI-saved time. Fifty-four percent said individuals decide or no explicit decision is made, and 31 percent said a manager or team decides.
What is an ROI harvester?
The Oliver Wyman Forum defines an ROI harvester as a senior owner accountable from the outset for the intended business result of an AI investment, with end-to-end authority to make the decisions that capture it. The harvester decides whether freed capacity becomes a cost case, a reinvestment case, or a growth case, and then acts on that decision.
Why do 22 minutes of checking matter for AI ROI?
Workplace AI users in the Oliver Wyman survey report a mean of 22 minutes checking, correcting, or redoing AI output for every hour of AI-assisted work. The survey can’t say how much of that is new work, but a gross hours-saved figure that ignores review time overstates the capacity a team can redirect.
How does Basil C. Puglisi’s Evidence-Based Value Realization apply to saved time?
In What AI Transformation Actually Requires, Puglisi writes that a claim of added capacity calls for evidence of what that capacity produced. Each claim needs a baseline, an attribution rule, and a stop rule, and time saved counts as value only when quality, rework, and risk hold steady or improve over the same period.
What should a company measure once AI starts saving time?
Track net capacity per workflow, meaning gross hours saved minus checking and rework time. Record a written destination for that capacity, chosen by a named owner, and report the business result it was meant to move, such as cost per resolved issue or cycle time, against a pre-AI baseline at each review.
#AIg
Leave a Reply