Nobody is replaced. The busywork is.

Say the words "we are looking at AI for the back office" in an open-plan office and watch what happens to the room. Nobody argues. Everybody understood.

The fear is not irrational and it is not solved by reassurance. It is solved by being specific, early, about what is actually going to change.

The fear arrives before you do

By the time anyone announces a project, the people in the area have already read the headlines and drawn their own conclusions. What they hear in "we are automating this area" is a question about their own week, and until somebody answers it out loud they will keep answering it privately, badly.

This matters practically, not just decently. The people who do the work are the only source of information about how the work is done. If they think the measurement is building a case against them, the measurement will be wrong, and everything built on top of it will be wrong too.

What actually leaves

What goes is the part of the job nobody would defend if you asked them.

• Typing a number into a second system because the first one does not talk to it.
• Opening the same portal every morning to check whether anything arrived.
• Rebuilding the same report on Monday from the same four places.
• Hunting for a document that already exists somewhere.
• Reading a long email thread to work out what was agreed.

Ask the person who does these things what they would drop if their week were an hour shorter. It is always this list. Never the customers, never the tricky cases, never the part where they actually know something.

What stays, and why

The exception. The complaint. The customer who has had a bad month and needs a person. The supplier relationship. The judgement call about whether this particular case is worth bending the rule for. Anything that spends money or makes a promise.

None of that is left with people as a courtesy. It stays because it is the work that needs a person, and a system that took it over would do it worse and do it confidently. In an audit this gets written down explicitly: which processes stay human, and why. That list is as much a part of the result as the automated half.

The team decides whether this works

The measurement depends on people recording what they do and showing how they do it. The build depends on someone explaining the exception that only happens twice a year. The handover depends on the team picking it up and using it.

A team that has concluded the project is aimed at them will do none of that, and none of it can be forced. They will answer the questionnaire thinly, forget the recording, and remember the exception in month four. The project does not fail loudly. It just quietly comes out worse than it should have.

The customer we say no to

Sometimes the goal really is a smaller team. That customer exists, and we are not the right people for them.

Partly because of how we would rather work, and partly for a plainer reason: those projects go badly. The pressure lands on automating the things that need judgement, because the aim is not that the work goes well but that a person becomes unnecessary. That is exactly the pressure that breaks these systems. Then the audit says half the area stays human, and the conversation turns into an argument.

What you do with hours you get back is your decision, and it is a real one: same team handling more, or a delivery time that stops embarrassing you, or somebody finally doing the work you hired them for. We just do not sell the first version.

What to say before you say anything else

If you are going to measure an area, tell the people in it three things before you start, in this order.

What is being counted: the work, not the worker. What it is for: finding the tasks that a system should be doing instead of a person. And what will not change: the parts that need somebody who knows the customer, the exception and the history.

Then let them see the result. The list of what stays human is usually longer than they expected, and it is the most convincing thing in the whole document, precisely because nobody wrote it to convince them.