Tired Malaysian professional at a desk with dual monitors and a tall stack of printed documents

Does AI Actually Reduce Workload, or Just Move It?

Short answer: AI does both, removing work from the doing and adding it to the checking, the learning and the managing. Whether the total falls depends on whether the team redesigns the work or just layers AI on top. In a 2024 survey of 2,500 workers across four countries, 77% of employees using AI said the tools had added to their workload, not reduced it.

The promise sounded straightforward.

Hand the tedious parts to the machine, and get your evenings back.

Years into the AI workplace era, the measured reality is more interesting than the promise.

The promise versus the measurement

In 2024, Upwork and Workplace Intelligence surveyed 2,500 workers across the US, UK, Australia and Canada, including 1,250 C-suite executives.

The gap between the boardroom and the desk was striking.

  • 96% of C-suite leaders expected AI to increase productivity.
  • 77% of employees actually using AI said the tools had added to their workload.
  • 39% of employees said they now spend more time reviewing or moderating AI-generated content.
  • 85% of respondents said their company was already using AI; 39% reported that their employer mandates the tools and 46% that theirs encourages them.

The leaders were not necessarily wrong about their expectations, and the staff were describing their own experience.

One reading of the gap is that the two groups were talking about different things: anticipated gains versus felt workload.

These are also self-reported impressions, not clocked hours.

Where the work often goes

  • Reviewing the output. AI drafts fast, but someone must check the figures, the tone and the facts. Review is real work, and it is rarely counted in the "time saved" claim.
  • Learning the tools. Every new tool has a learning curve, and the curve can repeat with every major update. That time comes out of the same working day.
  • Fixing confident mistakes. A plausible-but-wrong answer can cost more than no answer, because someone must find the error before it ships.
  • The checking gap is measured. KPMG's 2025 study found 58% of U.S. workers rely on AI output without properly evaluating it, and 57% said that reliance had led to mistakes in their work.
  • Feeding the machine. Writing a good prompt, gathering the right context, cleaning the data first: the preparation is work too.
  • Handling the extra volume. Faster drafting can mean more documents, and quicker replies can mean longer threads. When output gets cheaper, demand for it often grows.

Hands circling a sentence with a red pen on a printed draft document

The ten-second summary and the work that follows

Imagine a manager asking an AI tool: "Summarise this meeting transcript into five action items for the team."

The summary arrives in ten seconds. Neat, plausible, well formatted.

In this illustration, item three names the wrong owner, and the deadline the client mentioned verbally is missing.

Then the real work starts.

Did it catch the deadline that was mentioned once, in passing?

Did it assign the vendor follow-up to the person who actually owns it?

In this example, checking the summary against the transcript takes a few minutes. The missed deadline, left in, would surface next week and cost far longer to fix.

A practiced reviewer makes that check a habit. In a good course, the trainer has learners run the same check on their own summaries and spot the wrong owner themselves.

The draft was fast. The workload moved; it did not disappear.

The trap: cheaper output, more output

There is an old economic pattern that can apply here.

When a resource gets cheaper, people often consume more of it instead of banking the saving.

It does not happen every time, but it is common enough to plan for.

AI makes a first draft nearly free. So the report that used to be four pages can grow to twelve, and the weekly update can become daily.

When that happens, the work has not shrunk. The standard has inflated to fill the new capacity.

A hypothetical example, labelled as such: an executive saves thirty minutes drafting with AI in the morning, and spends forty minutes in the afternoon answering the extra threads created by everyone else's faster replies.

Her company may be more productive on paper. Her day is longer.

Why leaders and staff can see it so differently

The C-suite tends to see capability: what the tools can now do.

Staff tend to see the total: what their day now contains.

Both may be right about what they experience.

The risk is treating adoption as the finish line. When, as 39% of respondents reported, an employer mandates AI use, it is mandating the tool, not the outcome.

A tool dropped into an unchanged workflow often adds a step before it removes one.

When AI is more likely to reduce workload

None of these conditions guarantees a lighter workload.

They are the conditions under which a real saving seems most likely.

  • The task is a repetitive first draft. Routine updates, summaries, standard replies: where a fast first version was always going to be reviewed anyway, it can save time when drafting plus review together beat the old method.
  • The review habit already exists. Teams with a fast, reliable checking habit are more likely to keep the speed. Teams without one often lose it to rework.
  • The saved hour is banked on purpose. If the freed time is quietly refilled with more of the same, the diary never feels lighter, even when output genuinely rose.
  • The workflow is redesigned, not just accelerated. Removing a step beats speeding it up. If AI makes a weekly report unnecessary rather than faster, the workload actually falls.

The numbers worth watching in your own company

You do not need a research firm to find out whether AI is saving or moving your team's work.

Three practical measures help tell the story, though no single one settles it.

  • Draft-to-approved time. How long from the AI's first draft to a document someone would sign? If review regularly takes longer than drafting used to, the saving deserves a second look.
  • Rework rate. How often does AI-assisted work come back for correction? A rising rework rate is a warning sign worth investigating; weak checking is one common cause, but not the only one.
  • Volume versus value. Are you producing more output, or better outcomes? Twelve-page reports nobody finishes are not obviously a productivity gain.

Three actions that bank the saving

  • Cut one standing meeting. If AI now produces the summary the meeting existed to hear, the meeting can go. That is a redesign, and the saved hour is real for everyone in the room.
  • Retire one report. If a dashboard or an AI summary answers the same questions, stop producing the report instead of producing it faster.
  • Name the owner of the saved time. "We saved five hours this week" needs a destination, or the hours can quietly evaporate into more threads.

Graphic showing the 77 percent Upwork 2024 statistic and the draft, review, fix flow

What this means for a Malaysian team

The Upwork survey above was run in the US, UK, Australia and Canada, and the KPMG figures come from the United States, so treat these percentages as reference points rather than Malaysian measurements.

The underlying question travels well, though.

The question to ask is not "are we using AI?" It is "which two or three workflows have we actually redesigned around it?"

Measure hours, not vibes.

Pick the workflows where drafts and summaries dominate, build the review habit first, and protect the time that gets freed. That is when the promise has a chance of showing up in the timesheet.

For the related question of where Copilot genuinely saves time, see our practical take on Copilot at work.

For the checking habit that protects the saving, see what to check before you trust AI analysis.

Frequently asked questions

So is AI productivity a myth?

No. The productivity is real but conditional.

In our judgment it is more likely to show up in redesigned workflows with a checking habit, and more likely to get eaten in workplaces that simply add AI on top of everything they already did.

Why do employees say AI adds work while vendors promise savings?

One explanation is that the saving and the new work land on different people.

The tool may save the company capability while the reviewing, learning and fixing land on the employee's day. Both statements can be true at once.

What is the one thing to fix first?

The review habit.

Fast, reliable checking is what converts AI speed into actual saved time instead of rework. It is a trainable skill, and it is worth testing against your own review times and error rates.

So judge AI by the total workload, preparation, checking and fixing included, not by first-draft speed.

The teams that count honestly are the ones with the best chance of keeping the saving.

Michael Cheong runs hands-on AI training for Malaysian workplace teams, focused on exactly these habits: where AI genuinely saves time and how to keep the saving. Details are on the AI Trainer for Workplace Productivity page.

Basis: this is an opinion piece anchored on two published studies, both linked. The Upwork and Workplace Intelligence 2024 survey covered 2,500 workers (1,250 C-suite executives, 625 salaried employees and 625 freelancers) across the US, UK, Australia and Canada; the KPMG figures quoted are the U.S. findings of its 2025 study. Figures are quoted exactly as those sources state them; the examples are hypothetical illustrations, labelled as such. Claims about when AI reduces workload are judgment, framed as likelihoods rather than guarantees.

Similar Posts

Leave a Reply

Your email address will not be published. Required fields are marked *