The AI industry loves to talk about productivity.
Hours saved.
Tasks completed faster.
Documents generated in seconds.
And to be fair, those gains are real.
According to recent research from Workday, 85% of employees report saving between one and seven hours per week through AI. On the surface, that sounds like a huge success story.
But buried within the same research is a much more interesting statistic.
Nearly 40% of those productivity gains are being lost to rework.
Employees are spending significant amounts of time correcting errors, rewriting content, validating outputs, and checking whether AI-generated information is actually accurate. For every ten hours saved, roughly four hours are being spent fixing what the AI produced in the first place. (Workday)
As a CIO, I don’t find this surprising.
In fact, I think many organisations are measuring the wrong thing entirely.
The Productivity Illusion
Most AI discussions focus on gross productivity.
How many minutes were saved?
How many documents were generated?
How many tickets were closed?
How many reports were summarised?
The problem is that speed alone tells us very little.
If somebody generates a policy document in five minutes but another employee spends thirty minutes correcting it, have we really gained productivity?
If an AI-generated summary omits critical context and requires multiple reviews before publication, have we actually reduced effort?
The answer is often no.
We’ve simply moved the work somewhere else.
The Rise of the AI Janitor
One of the most interesting findings in the Workday research is that highly engaged employees appear to carry much of the burden of AI validation. These are often the people reviewing outputs, correcting mistakes, identifying hallucinations, and ensuring quality before work reaches customers, executives, or regulators. (investor.workday.com)
In many organisations, the best people are quietly becoming AI janitors.
Not because they’re resistant to AI.
Quite the opposite.
They’re the people experienced enough to recognise when something is wrong.
Unfortunately, this work is often invisible.
Management sees the faster output.
The experts see the cleanup effort.
Those are two very different realities.
AI Is Brilliant at the Wrong Things
The mistake many organisations make is assuming AI should be applied everywhere.
It shouldn’t.
AI is exceptional at:
- Drafting first versions
- Summarising routine information
- Generating ideas
- Handling repetitive administrative work
- Accelerating research
Where it becomes dangerous is when organisations expect it to replace expertise.
A meeting summary is very different from a legal opinion.
A first draft is very different from a board paper.
A suggested solution is very different from an engineering design.
The more specialised the work becomes, the more expensive errors become.
This is where many organisations discover that the final 10% of quality requires 90% of the effort.
Faster Isn’t Better
As technology leaders, we should be measuring net value, not gross efficiency.
The question isn’t:
“How much time did AI save?”
The question is:
“How much value did we create after accounting for verification, correction, governance, and risk?”
Those are very different measurements.
A process that takes ten minutes longer but produces higher quality outcomes may ultimately be more productive than a process completed instantly and corrected three times.
The Training Gap Nobody Talks About
Another finding from the Workday research stood out to me.
While two-thirds of leaders identify AI skills as a priority investment area, only a minority of employees experiencing significant AI rework report receiving meaningful training. (investor.workday.com)
This is a familiar pattern.
Organisations buy AI.
Employees are expected to use AI.
Very little effort is invested in teaching people how to use AI effectively.
Prompting, validation, source verification, risk assessment, and understanding model limitations are skills.
Like any skill, they require training.
Without that training, organisations shouldn’t be surprised when productivity gains evaporate.
My View
I remain extremely optimistic about AI.
I use it every day.
My teams use it every day.
The productivity gains are real.
But the conversation is finally becoming more mature.
We’re moving beyond asking whether AI saves time.
We’re starting to ask where it creates value and where it creates rework.
That’s a much more useful discussion.
The organisations that succeed with AI won’t necessarily be the ones using it the most.
They’ll be the ones that understand where human expertise still matters, where AI genuinely accelerates work, and where quality remains more important than speed.
Because in the end, nobody gets promoted for producing work faster.
They get promoted for producing better outcomes.
