You want the people who know the work to shape your first AI job, not quietly work around it. You run the firm day to day, and you rely on a few people in the office who handle the scheduling, the calls and the paperwork.

The trouble often starts with the announcement. You mention AI at the Monday meeting before anyone knows which job you mean. By Wednesday your longest-serving office lead has decided it’s about her role, and she stops telling you where the schedule really goes wrong.

Once that happens, you lose the knowledge the job needs, and the first attempt fails in a way that makes a second attempt harder to start. You’ll come away with the first conversation to have, what to say in it, and what you can honestly promise.

Talk to the team before you choose a tool

I believe the first conversation should happen before you ask any supplier for a demonstration. At that point you can still change which job you pick, what information it uses and how it gets checked, without defending money already spent.

Leave it until after the tool is bought and the process is designed, and consultation feels like a request to approve a finished decision. In a close office team, people can cooperate in public while protecting themselves in private. They keep the old method going, fix the new output without mentioning it, or stop passing on anything that sounds like bad news. The job then loses the knowledge it needs while looking as though it has support.

Talking early also lets you be honest about what isn’t settled. You can say what’s decided, what’s still open and when you’ll decide the rest. That’s more believable than pretending every answer is already known, and it gives the team somewhere to put their concerns.

Describe one job, not AI in general

Introduce AI by describing one job, why you’re looking at it, and the next decision you’ll make. Keep the tool in the background until the team understands the work.

Start with the job as it runs today. In most offices, a good example is turning the weekend’s out-of-hours messages into Monday’s handover note. It starts when the out-of-hours phone comes back to the office. What comes in is the notes of each call and message. What the business needs out is a clear list of changes, complaints and follow-ups for the week. Because those messages hold customers’ personal details, decide what the tool may see before anyone tries it. How to write an AI use policy covers that. “Using AI in the office” covers too many activities for anyone to respond to properly.

Then say why this job. Imagine the person covering the out-of-hours phone spends 2 hours every Monday morning typing up the messages while the phones ring. Tie the reason to work the team recognises, not to a general claim about keeping up with technology.

Finally, say what hasn’t been decided. You may not know which tool fits, how much of the note it can draft, or whether a trial will justify carrying on. Naming those open questions shows the team exactly where they can influence the answer.

Name the owner and what stays with people

Describe the handover note as one bounded step with a named owner and a clear point where a person decides. The team should be able to see where the AI step starts, where it stops and who answers for the result.

Name the owner early. They keep the instructions and the boundary current, act on corrections, and stop the job if it no longer meets the standard. Here that might be the office manager or the longest-serving office lead. Who owns the job covers the authority and protected time that person needs.

Be just as clear about what stays human. In this example, any complaint, any question about a customer’s account and every call back to an unhappy customer stays with a person in the office. The AI step might draft the handover note. A person reads it against the messages, adds what the messages don’t say, and decides what happens next.

Avoid describing the tool as a new member of staff. That language invites people to picture a whole role being replaced when you’re looking at one step. Talk about the input, the output, the check and the next action instead.

Let the people who do the job shape it

The people who do this job every week should shape the rules, the exceptions and the checks before a tool touches live work. Their experience turns a neat process description into one that survives a real Monday.

Ask them to walk through three or four recent handover notes and point out where they made a judgement. Look for the call they treated differently, the detail they corrected before anyone else saw it, and the moment they stopped to ask a manager. Writing down the unwritten rules sets out how to capture those decisions from real cases rather than memory.

Turn each decision into a rule the job can follow, or a condition that sends the work back to a person. The person who does the job should be able to read what’s written and recognise their own process, including the parts that stay outside the AI step.

Invite challenge while the design can still change. Ask what would make the job unsafe, slower or harder to check, and write down what you decided. People keep raising problems when they can see where the last one went.

Keep the same person involved in the trial, because they can compare the draft with the standard and spot corrections that keep coming back. How to test an AI job with the people who do it explains how to run that trial and decide what happens next.

Be straight about roles and hours

Promise the office team a clear process for decisions, not an outcome you can’t yet know. Before the job has been tested, you can say how people will be involved, when they’ll hear updates and who makes decisions about changed work. When you don’t know yet, say when you’ll decide, what evidence you’ll use and who will tell them.

Don’t promise that nobody’s work will change. The point of the job is to change how a step gets done, so that reassurance breaks the first time the process moves. Don’t promise a number of hours saved before real work has been tested either.

If someone asks whether this is about cutting jobs, answer the question directly. I believe the right aim is efficiency, not headcount: the hours a job frees should go somewhere named. There’s rarely a shortage of places, such as calls back to customers, chasing overdue invoices or schedule gaps that currently get filled at the last minute. Ask the office team where the time should go. What to do with the hours AI saves covers how to decide.

Talk about tasks and roles separately. Taking time out of one task doesn’t remove a role on its own. If a change could affect a role rather than a task, say so plainly and follow the proper employment process. The redundancy question covers that conversation.

Keep the team updated once it’s running

Make updates about the handover note part of how the job runs, rather than waiting for a problem or a formal meeting. The team needs to know what the job is doing, what has changed and which questions are still open.

Once the job goes live, keep one route for questions and one current version of the process. Tell the team when the boundary or the information rules change, and when the evidence leads you to continue, narrow or stop the job.

Share the result in terms the office can see. If the handover note is now ready by 9 o’clock rather than 11, tell them that, in those words. It might also be fewer interruptions for the office lead who holds all the knowledge, or calls to customers that used to wait. Use the measure you agreed for the job, not a broad claim that AI is saving time.

Hold a 30-minute team conversation this week

On Monday, before you speak to any supplier, write three lines: the one job you’re looking at, why you’re looking at it, and what isn’t decided yet. That takes about 15 minutes.

Next, book 30 minutes with the office team to walk through those three lines and name who will own the job. Follow it with 20 minutes one to one with the person who does that job most often, to start listing where they make judgements. By the end of the week you’ll have a job, an owner and the person who’ll help shape it.