The AI Time Savings Nobody Is Counting

How to measure where AI nets you time and where it adds hidden work to your team

A team lead tells you the group saves ten hours a week with AI. The number is clean. It goes straight into the ROI deck and nobody questions it.

Spend a few days watching the actual work and that ten splits in two. Five hours genuinely leave the week. The other five go to catching what the AI got confidently wrong, reworking a draft that read well and said nothing, and re-checking figures that looked plausible. The headline still says ten. The number you can trust is five.

The figure most teams report is the gross one

Workday put a measurement on this gap. In a January 2026 study of 3,200 workers and leaders, the company found that for every ten hours of efficiency people gain from AI, nearly four hours go back into fixing the output. They called it the AI tax on productivity.

The surface story looks great. Eighty-five percent of workers said AI saved them somewhere between one and seven hours a week. Dig one layer down and it gets sober fast. Only fourteen percent came out consistently ahead once the rework was counted. Most teams are reporting the first number and living the second.

There’s a name for what creates the tax

Researchers at BetterUp Labs and Stanford’s Social Media Lab named this in Harvard Business Review back in September 2025. They called it workslop: AI-generated content that looks like good work but carries no substance to move the task forward.

It reads fine. It passes a glance. Then the person who receives it spends close to two hours working out what’s missing and rebuilding it. The effort doesn’t disappear when AI produces the draft. It moves downstream to whoever has to make the draft real.

The ROI question worth asking

Whether AI saves time overall is a question with an easy answer, and the easy answer tells you nothing useful. It almost always saves time somewhere. That fact alone gives you no way to decide where to point it.

The useful question is narrower. Which task types come out net positive after rework, which break even, and which cost more than doing the work by hand? Answer that and you stop debating AI in the abstract and start making decisions about specific work.

Run the audit by task type

Some categories reliably net out. Rule-bound data tasks net out when the inputs are clean and the rules are explicit, work like reformatting, sorting, or pulling fields from a structured source. So do templated communications that follow a known format, and first-draft outlines built from a detailed brief. The common thread is constraint. Give the task tight edges and the model has less room to be confidently wrong, so the rework stays low.

Other categories bleed time. Nuanced brand writing leans on voice and judgment the model can’t fake. Work that depends on current institutional knowledge runs into the simple fact that the model never had it. And anything that has to match one specific person’s voice closely enough to publish usually comes back needing a rebuild. The draft looks finished. It isn’t. Someone reworks most of it while the timesheet still records a saving.

The honest test for any task is small. Track the hours the AI saves on it, then track the hours your team spends correcting, verifying, and rewriting the output. The gap between those two numbers is the only productivity figure that matters, and it varies wildly by task. Two tasks that both “use AI” can sit on opposite sides of break-even.

What to do with the breakdown

Once you can see which tasks net out, the workflow decision gets simple. Point AI at the low-rework work and let it run. Put a tighter review gate on the high-rework work, or keep it human until the brief and the source material are good enough to change the math. Stop counting hours saved and start counting hours saved minus hours spent fixing. That second number is the one that shows up in your margins.

There’s a cost beyond the math. The Workday data found that the most engaged people absorb the most rework, because they’re the ones using AI hardest and reviewing it most carefully. Run that for a year without redesigning the work and your best people are doing two jobs at once, generating output and then validating it. That’s how a productivity tool becomes a burnout engine.

The teams getting real leverage from AI have narrowed it to the work where it nets out, measured the rework honestly, and pointed the tool there on purpose. The ten-hour number felt good. The five-hour number is the one you can build on.

. . .

Want to save hours each week by turning work into repeatable AI workflows?

The Fortune 100 AI Skills Library™ includes plug-and-play prompts built to save leaders time and money. Copy, paste, and edit in 60 seconds, then apply them across planning, execution, and reporting.