Can AI replace admin staff? What the evidence actually shows

The short answer

AI is taking on admin tasks, not admin roles — and the evidence for that is better than most of what gets written about it. The Office for National Statistics now measures it directly. Among UK businesses using AI in June 2026, roughly half reported no change in headcount, and around one in twenty reported a reduction.

What changes is the shape of the job. The sorting, re-keying, booking and chasing move to software; the judgement, the exceptions and the relationships stay with the person. The businesses that get value from this plan for that change deliberately rather than hoping a tool will absorb a role.

Where AI use is heading

Adoption is rising quickly. The share of businesses with ten or more employees using at least one AI technology rose from 11.9% in late September 2023 to 34.9% in late June 2026 — almost three times its starting level, measured every quarter in between.

That is a measured trend, not a forecast, and we have not extended the line. Its direction is clear enough: within a few years, using AI somewhere in the business will be ordinary rather than novel. The ONS also found that improving business operations is the most common reason businesses give for using it.

What the evidence says about jobs

The same ONS analysis asked businesses using AI what it had done to their headcount. Across every size band, no change was by far the most common answer, and reductions were reported by a small minority.

Two details are worth reading carefully. First, the largest businesses were the most uncertain: over a third of those with 250 or more employees said they were not sure what effect AI had had, which suggests the effects take time to show. Second, the ONS found that businesses using AI specifically to improve operations were more likely to report reductions than those using it for other purposes only. Operations is where admin lives, so the honest answer is that some admin roles do shrink — but far fewer than headlines imply.

Which tasks move, and which stay

Tasks that move well

• Sorting incoming requests by type, urgency and owner.
• Re-keying the same details from one system into another — often plain automation, not AI at all.
• Booking and rescheduling against real availability.
• Chasing overdue replies, documents and signatures, on a schedule.
• Drafting routine summaries and first-draft replies for a person to check.

Tasks that stay with people

• Anything involving judgement, an exception, or a customer who is upset.
• Decisions with money, safety or legal weight attached.
• Knowing that this particular request is not what it looks like.

We explain the difference between the tools that do this in AI agents, AI employees and copilots.

The hours, in arithmetic

Take an illustrative full-time administrator on a 37.5-hour week who spends 27 hours on routine tasks. With the routine parts automated, the same work takes about 9.5 hours.

That frees 17.5 hours a week. Multiply by 4.33 for a month (about 76 hours) and divide by 162.5, the hours in one UK full-time month (37.5 × 52 ÷ 12): about 0.47 of a full-time person. In most businesses those hours are worth more redeployed — to customers, exceptions and the backlog nobody gets to — than removed.

Replace every input with your own. Ask the person who does the job to log a fortnight of tasks; the method is in how many hours automation can save, and the CRM-specific version in how much time a CRM saves.

How to introduce it without losing good people

1. Start with the people who do the work. Ask which tasks they would most like rid of. They know where the time goes.
2. Move one task at a time, measure it, and keep the person checking the output until it has earned trust.
3. Write down what the software may decide alone, what needs a person, and what it must never touch.
4. Redeploy the hours on purpose. If nobody decides what the freed time is for, it evaporates.
5. Keep a named person accountable and an audit trail of every automated action.

The honest conclusion

If you are hoping AI will let you remove an admin role outright, the evidence suggests most businesses do not end up there, and those that try often find the exceptions were the job. If you are hoping it will give your best administrator back a day or two a week for work that matters, that is realistic, measurable and increasingly common. For how we approach this, see AI employees and what AI employees are.