The sentence that causes the problem #
"We are looking at AI to make the business more efficient."
Every person hearing it does the same arithmetic, and the answer they reach is about their own job. It is not an unreasonable inference, because the sentence contains no information that would rule it out. Efficiency with no object is heard as headcount, and once that is the frame, the rest of the meeting is people managing their own anxiety rather than listening.
The fix is not reassurance. It is specificity.
Four things to say, in this order #
-
The job, by name
Not "admin". The Monday morning reconciliation. The certificate chase. Re-keying the booking form into the system. A named job is a boundary, and a boundary is what tells everyone else this is not about them.
-
Why that one
Because it eats a day a week, because it is the thing people complain about, because it gets dropped when you are busy. Reasons that match what people already experience are believed. Reasons about strategy are not.
-
Who is involved in working it out
Name the person who does the job now. They know the nine exceptions nobody documented, and involving them is both the practical route to a working result and the clearest signal that this is being done with them rather than to them.
-
What you do not know yet, and when you will
This is the one people skip, and it is the one that buys credibility. "I do not know yet what this means for how the role is shaped, and I will tell you in October" is a far stronger position than a comforting answer you may have to take back.
What not to say #
- Do not promise nobody will be affected unless you mean it. A withdrawn promise is remembered for years and poisons the next announcement, whatever it is about. If the honest answer is that roles will change, say roles will change.
- Do not say it will free people up for more interesting work without being able to name the work. Said without a name it reads as management language for something worse, and people are right to read it that way.
- Do not present it as already decided when it is not. If you are still working out whether the job can be automated at all, say that. Plenty of jobs turn out not to be worth it, for reasons set out in when not to automate something, and the person who does that job is usually the first to know why.
- Do not let it leak. A project everybody finds out about through a calendar invite has already become a rumour, and the rumour is always worse than the plan.
If the system will record what people do #
This is the part most businesses miss, because it does not feel like monitoring. An automation that logs how long a task took, who handled what, how many items each person approved or when somebody last touched a record is processing personal data about workers, and the ICO has specific expectations about it.
Its guidance states that employers must identify a lawful basis for monitoring, must make workers aware of how and what personal information is being collected, and must check that the monitoring is necessary for the purpose with no other reasonable and less intrusive way to achieve it.
A data protection impact assessment is required where processing is likely to be high risk, and the ICO says you should do one even where it is not. The page carries a notice that it is under review following the Data (Use and Access) Act, so check it rather than this summary.
The surrounding duties are in what UK GDPR asks of a business using AI.
When this stops being a conversation and becomes a process #
There is a real line between exploring a change and proposing redundancies, and it has legal consequences. GOV.UK sets out collective consultation rules that apply where 20 or more employees may be made redundant at one establishment within any 90 day period: consultation with representatives, and minimum periods before the first dismissal of 30 days where 20 to 99 are affected and 45 days where 100 or more are.
Smaller numbers still carry duties around fair process and genuine redundancy.
Most business automation is nowhere near this. It matters anyway, because once you know where the line is, you can be confident that an early honest conversation is not a legal step, which is exactly the anxiety that makes people say nothing and let it leak instead. What actually tends to happen to the person whose job was half admin is covered in will AI replace my staff.
After the announcement #
Two things are worth doing that nobody does.
Tell people what is allowed while the project is under way. Staff who hear the business is interested in AI will start experimenting with it, and if there is no line they will invent one that is more generous than yours. That is a live data protection exposure rather than a hypothetical one, and it is covered in your staff are already using AI.
Then close the loop. Come back in a few weeks and say what was found, including the parts that did not work. A project that goes quiet after the announcement is assumed to have gone badly, or to have gone well in a way nobody wants to describe.