Executives Say AI Saves Time, Workers Say It Does Not

AI was sold as a shortcut to productivity, but new surveys suggest a widening perception gap inside companies: leaders report hours saved, while many employees say the tools slow them down or add new work.

What the New Surveys Reveal About Workplace AI Time Savings

In a survey of 5,000 white-collar professionals by AI consultancy Section, 33% of executives said AI saves them 4 to 8 hours per week and 19% claimed more than 12 hours saved. Only 2% of executives reported no time savings at all.

For non-managers, the story flips. Section found 40% of workers said AI saves them no time, while 27% reported less than two hours saved weekly and just 2% said they save more than 12 hours. Most respondents—85%—reported either no work-related AI use cases or only beginner-level ones, and 40% said they would be fine never using AI again.

A separate survey by Workday adds a twist: among employees who said AI does save time, 85% reported spending that time back correcting AI errors, as highlighted by reporting in the Wall Street Journal.

Why Leaders Feel Gains Employees Do Not With AI Tools

Executives tend to use AI for tasks where the tools shine: summarizing long documents, drafting emails and memos, shaping presentations, and scanning market chatter. They also enjoy better access to premium models, cleaner data, and hands-on support from IT or vendors—all of which boost perceived efficiency.

Individual contributors often face the opposite conditions. They deal with compliance checks, strict data access, messy handoffs, and “pilot purgatory” where tools aren’t fully embedded in core systems. Quality assurance becomes their responsibility: fact-checking model outputs, standardizing formats, and aligning with brand or regulatory rules. The result can feel like extra work disguised as automation.

Productivity Depends on the Job and the Data

Section’s report shows the technology sector leading AI use, while retail lags. That tracks with real-world outcomes. Developers report faster completion of repetitive coding tasks with code assistants, although bug risk rises without rigorous review. A GitHub study found developers completed a constrained coding task 55% faster with an AI assistant, but those gains rely on guardrails, tests, and clear specs.

Outside engineering, the evidence is mixed. A widely cited National Bureau of Economic Research working paper by Stanford and MIT researchers observed a 14% productivity boost for customer support agents using AI, with the biggest gains for less-experienced workers. That nuance matters: AI can compress the learning curve in some roles, yet offers marginal returns—or even creates rework—in others.

The Measurement Problem Behind AI Productivity Claims

Time saved by drafting a first pass is not the same as time saved end-to-end. If employees spend minutes generating a summary but then spend more minutes verifying facts, adjusting tone, and jumping between systems to fix formatting, net savings evaporate. Many firms track “prompts sent” instead of defect rates, cycle time, or customer outcomes.

Data quality is another culprit. Models are only as good as the context they’re given. Without secure connectors to knowledge bases, CRM notes, policies, and style guides, AI produces plausible but wrong outputs that demand human repair. Training gaps compound the issue: prompt patterns, retrieval techniques, and evaluation workflows are still new to most teams.

How to Turn AI Perception Into Reality for Workers

Start with high-volume, narrowly scoped tasks where accuracy can be measured—claims triage, invoice matching, meeting notes with source links, or code test scaffolding. Pair models with retrieval from vetted data and require citations so reviewers can check facts quickly.

Instrument the workflow end-to-end. Measure cycle time, error rates, rework, and downstream outcomes, not just draft speed. Let frontline teams flag where AI adds friction, and adapt or roll back use cases that don’t clear a quality bar. Incentives should reward verified results, not vanity metrics.

Invest in enablement: short, role-specific training; reusable prompts; model choice guidance; and clear escalation paths for questionable outputs. Treat human review as a feature, not a bug—define when AI can auto-ship and when a second pair of eyes is mandatory.

The Stakes for Social Permission and Workplace Trust

Even AI’s most prominent backers warn that public support hinges on tangible, broad-based benefits. Microsoft’s Satya Nadella has urged the industry to prove gains in areas like health, education, and public services—not just in executive workflows. If average employees feel only more oversight and cleanup, enthusiasm fades and scrutiny grows.

Can AI save time? Yes, in the right jobs with the right data and guardrails. But until workers experience dependable, end-to-end savings—not just faster drafts—executives will keep celebrating time wins while many employees still say the clock hasn’t changed.