Why does the fear arrive before you have announced anything?
Somebody forwards a newspaper piece about a large firm cutting graduate intake. A client mentions their own AI programme on a call. A partner says, without thinking much about it, that the firm is "having a look at AI for reporting". By the time you sit down to explain anything properly, your people have already had the conversation several times, in kitchens and group chats, without you in the room.
That is the first thing to accept. You are not deciding whether to raise the subject of job security. You are deciding whether your account of it arrives before or after everybody else's. Arriving second is expensive, because you then spend your credibility correcting a story rather than telling one.
What are people actually afraid of?
It is rarely one fear. It is usually three, and they need different answers.
- Losing the job. The straightforward one. Fewer hours of work means fewer people needed to do it.
- Losing the craft. Quieter and often stronger. People who are good at drafting a report or a set of file notes take pride in it. The worry is not unemployment, it is being reduced to tidying up output they did not write and cannot fully vouch for.
- Being measured. Anything that touches how work is recorded looks like surveillance. A rebuild of timesheet capture is the obvious case, but reporting and file notes raise it too. If the system knows how long each stage took, who sees that, and what happens to the person whose stage takes longest?
If you answer the first fear and ignore the other two, you will think you have handled it and wonder why nothing improves.
Why does blanket reassurance fail?
"Nobody is losing their job because of this" is a sentence people have heard before, sometimes from firms that later made redundancies for unrelated reasons. It costs nothing to say, which is exactly why it carries little weight. A promise that cannot be checked is not reassuring, it is just noise before the next thing.
What lands instead is specific and falsifiable. Name the process. Say what it is today and what it will be afterwards. Say who reviews the output and whose name goes on it. Say what you will do if the thing turns out not to work. A team can test every one of those statements within a month, which is the point.
What can you honestly promise, and what can you not?
You can promise the scope. One process, rebuilt, not a programme that quietly expands into everything. You can promise that a named person still reviews and signs off anything going to a client. You can promise the people doing the job will be the ones who help design it, because a rebuild is impossible otherwise: the real process lives in their heads, not in the manual.
You cannot promise the firm will never change shape. Nobody can, and partners who try it get caught out by a lost instruction or a bad year that has nothing to do with software. Say the honest version: this piece of work is about capacity, and the firm's size will continue to depend on the work it wins. Treating people as adults is not a risk. Being caught in an overclaim is.
What actually changes for a person on an ordinary Tuesday?
This is the part most firms skip, and it is the part that settles people. Describe the day, not the strategy.
Today a fee earner finishes a site visit or a client meeting and faces an hour of writing that repeats what already happened. Afterwards a structured draft is waiting, in the firm's own format, and they spend fifteen minutes correcting the points that need judgement. The judgement calls do not move. The typing does. The same shift runs through client reporting and file notes: author becomes editor.
For more junior people the change is sharper and worth addressing on its own terms, which we do in what changes for junior staff. For everyone, be clear about the work that stays human, because knowing where the line sits is what stops people assuming it will keep moving.
What makes the fear worse?
Running the build in secret. Bringing in outside help who interview people without explaining why. Launching a finished system nobody saw being made. Each of these tells staff that they were the subject of the exercise rather than part of it.
The other thing that makes it worse is a vacuum. When a firm says nothing official, people improvise. They start pasting client material into free consumer tools because they have decided the firm is behind and they should get ahead. That is shadow AI, and silence is the main cause of it.
How do you know whether people believe you?
Watch for two signals. The first is whether anyone tells you the output is wrong. A team that has quietly decided this is a cost exercise will nod and work around it. A team that believes it will complain loudly about a draft that misses the firm's house style, because they expect to be listened to.
The second is whether anybody nominates the next job. When someone says "you should do this to the bid pack next", the argument is over. Until then it is still being had, whatever the room looked like in the meeting. Getting a team to adopt AI goes further into that, and why AI rollouts fail sets out how often the failure is social rather than technical.
What if the firm really is planning to be smaller?
Then do not dress a reduction as a technology project. People reconstitute the story from fragments, and once they decide the automation was the cover, you have lost the cooperation you need to make anything work, including the next three things you try.
In practice, firms of twenty to two hundred people are rarely short of cost. They are short of senior capacity, which is the pattern behind the partner bottleneck, and the realistic comparison is set out in automation versus hiring. If the hours come back and the work is there, the question becomes what you do with the capacity, which is a better problem and the one utilisation targets after the admin goes deals with. If you want a neutral starting point for the conversation with your team, the audit names the process worth tackling first and shows every assumption behind the number.