G-P’s 2026 AI at Work Report Shows the AI ROI Reckoning Has Arrived

Why this is a confidence problem, not an adoption problem

The report does not show companies walking away from AI. In fact, G-P says all surveyed executives are using it ^[3]. The shift is more specific: leaders appear to be moving from enthusiasm about deployment to pressure for measurable business value.

That is a meaningful change from the adoption-first mood of the prior cycle. G-P’s 2025 AI at Work Report emphasized acceleration, with 91% of executives actively scaling AI initiatives and 74% saying AI was critical to company success ^[14]. The 2026 report still shows widespread use, but its framing has moved toward accountability, pressure-testing, and proof of ROI ^[3].

Other research points to a similar value gap. Boston Consulting Group reported that 60% of companies were not achieving material value from AI at scale, while another 35% were seeing some returns but not moving far enough or fast enough ^[7]. McKinsey likewise found that 92% of companies planned to increase AI investments over three years, yet only 1% of leaders described their organizations as mature enough for AI to be fully integrated into workflows and driving substantial business outcomes ^[13].

The hidden cost: AI can make work faster and heavier at the same time

One reason AI ROI can disappoint is that speed at one step can create verification work at another. G-P reports that 69% of executives say employees are spending more time monitoring, reviewing, or updating AI-generated work ^[3]. In practice, that means a tool may produce drafts, answers, code, or summaries quickly while shifting the burden to humans who must check accuracy, rewrite output, manage risk, or clean up errors.

That matters because gross output is not the same as net productivity. If AI helps a team create more material but also requires more review, the real return depends on the full workflow, not just the automated step. Separate Workday research summarized by Channel Insider makes a similar point: time saved by AI can be offset by rework such as fixing mistakes, rewriting content, and double-checking AI outputs ^[11].

Performative productivity is becoming a real management risk

G-P’s report also highlights a softer but important risk: AI activity can be mistaken for business value. The survey found that 88% of executives are concerned about employees using AI to appear productive or comply with AI-use expectations without producing meaningful outcomes ^[3]. Nearly half, 47%, are very or extremely concerned this is already happening ^[3].

That concern should make companies cautious about measuring AI success through surface-level signals such as tool logins, prompt counts, number of AI-generated drafts, or employee self-reports of AI use. Those metrics may show activity, but they do not prove that work became better, faster, safer, or more profitable.

The human-worker paradox

The most sensitive workforce finding is that 82% of executives said AI has lowered the value they place on human employees ^[3]. That is striking because the same report also shows that humans are still being asked to review, monitor, and update AI-generated work at significant levels ^[3].

The lesson is not that people are irrelevant. It is that many organizations may be undervaluing the human judgment required to make AI useful. McKinsey’s workplace AI research argued that companies should focus on practical applications that empower employees in daily work and connect AI to measurable outcomes, rather than treating AI as a standalone technology rollout ^[13].

How companies should rethink AI success metrics

G-P’s findings suggest that companies need to measure AI by outcomes, not by adoption. A stronger AI scorecard would include:

1. Business impact: revenue contribution, cost reduction, customer experience improvement, risk reduction, or faster delivery of validated work.
2. Net productivity: time saved minus time spent reviewing, correcting, rewriting, or updating AI-generated output.
3. Quality: error rates, escalation rates, compliance issues, customer satisfaction, and the percentage of AI outputs accepted without major rework.
4. Workflow integration: whether AI is embedded in a high-value process, not just available as another tool.
5. Human leverage: whether AI helps employees make better decisions and complete more valuable work, rather than simply increasing output volume.
6. Budget accountability: whether each AI use case has a clear owner, baseline, success metric, and review date before spending expands.

The practical shift is simple: companies should stop asking only whether employees are using AI and start asking whether AI is improving the validated work that matters.

The bottom line

G-P’s 2026 AI at Work Report is not an anti-AI story. It is an accountability story. The same survey that found universal AI use also found that 73% of executives were underwhelmed by at least some AI investments and that nearly 70% may scale back spending if goals are not met ^[2][3].

Because the report is based on executive survey responses, it should not be read as audited proof that AI has failed. It does show something important: the burden of proof has shifted. For workplace AI, the next phase is less about deployment and more about measurable, trusted, human-validated business value.