You announce at the Monday meeting that the company is "going to start using AI." Half the room nods. The other half just heard "we're figuring out which of you we don't need."

That second group isn't being dramatic. They're reading the same headlines you are.

The real fear you're dealing with

In a Pew Research survey of about 5,300 U.S. workers, fielded in October 2024 and published the following February, 52% said they were worried about how AI will be used in their workplace. 32% expected it to mean fewer job opportunities for them in the long run. Only 6% expected more.

So when your office manager goes quiet after the announcement, that's the normal reaction, not the dramatic one.

The fear is rational, and you can't train around it. You have to answer it first.

That means saying, out loud and specifically, what AI is going to do to each person's job. If a tool is going to take over a task entirely, say so. "The after-hours line is going to an AI answering agent. Nobody's hours change, because nobody was taking those calls anyway." If it isn't going to replace anyone, explain why in terms they can check for themselves. "It drafts the estimate. You still walk the property, you still set the price, you still sign it. It types faster than you. That's the whole pitch."

Vague reassurance ("nobody's job is at risk!") makes it worse. People have heard that line before, usually a few months before layoffs. Specifics are what they can believe.

Framing that actually works

"This is going to handle the part of your job you like least."

Every role in a service business has a grind task. The dispatcher re-keys the same job details into three systems. The paralegal spends Friday afternoons formatting discovery responses. The practice manager chases the same insurance pre-authorizations every week. Your field techs write up the same notes at the end of every job and hate every minute of it.

Don't guess what that task is. Ask.

Sit down with each person and say: "If I could make one repetitive part of your week disappear, what would you pick?" They'll have an answer in under ten seconds. They've been thinking about it for years. And it's often not the task you'd have picked, because you haven't done their job in a while.

Start there. One task, one person, chosen by the person. Now the tool is something they asked for.

A side benefit: the grind tasks are usually the ones AI is good at. Summarizing, reformatting, drafting from a template, pulling data from one place and putting it in another. The judgment calls and the hands-on work stay human, which is exactly what you told them.

The rollout approach that doesn't create chaos

The fastest way to kill AI adoption is a company-wide mandate. "Everyone will use ChatGPT for emails starting Monday." You'll get compliance from a few, quiet sabotage from a few, and a majority who open it twice, get a weird result, and never open it again.

Do this instead. Pick one person. Curious and comfortable with software. Not necessarily the youngest person in the office, and definitely not whoever's already overloaded.

Give that person one tool for one task. Their task, the one they named. For most offices the tool is a business plan like ChatGPT Business or Claude Team, nothing more exotic than that. Neither trains on your data by default, and both generally start at two seats, around $25 each month to month, so the second seat is yours. The dispatcher pastes the day's job notes in and gets the recap email back as a draft. The paralegal pastes a discovery response in and asks for it reformatted to the firm's template. Give them two weeks and permission to waste some time on it. Check in once a week.

When it works, and for a well-chosen task it usually does, they become the internal expert. Then they show the next person, and you stay out of it.

When the owner says "AI is great," the staff hears a sales pitch. When the dispatcher who's been there six years says "honestly, it saves me half an hour a day on the recap emails, here's how I set it up," that's a coworker talking. Peer-to-peer carries weight a mandate never will.

It also protects you. If the tool turns out to be junk for your workflow (plenty are), you've lost two weeks of one person's experimentation instead of a month of everyone's frustration.

Ground rules to set on day one:

  • Which tools are approved, and what never goes into them. Put it in writing.
  • A human reads everything before it goes out. AI drafts; people send.
  • Mistakes caused by the tool get reported, not hidden. You want to hear about the made-up price in the estimate before the customer does.
  • Nothing touches a customer-facing channel (the phones, the website chat, the estimates) until the pilot person says it's ready. That includes the after-hours answering agent from earlier: it goes through the same pilot first.

A medical practice or a law office needs to take that first rule further. Staff can't paste patient records or confidential client files into a free or personal chatbot. Consumer plans can train on whatever gets pasted in unless someone changed a setting, and the setting isn't the point: there's still no business associate agreement. A business plan fixes the training default but not the paperwork. Neither ChatGPT Business nor Claude Team comes with a BAA (that takes an enterprise-level arrangement), so keep patient information out of your pilot until you have one signed.

Law offices have their own line to check. ABA Formal Opinion 512 says lawyers generally need a client's informed consent before putting client information into a self-learning AI tool. Whether your tool counts, and what your state bar says, is a question for whoever handles ethics at the firm. Get that answer before real client documents go in.

What to measure and share with the team

One number: hours saved per week.

Have your pilot person track it, even roughly. "Recap emails used to take me 45 minutes a day. Now they take 10." That's just under three hours a week. Write it down.

Then tell the team. Skip the slide deck. Say it in the Monday meeting, in one sentence, with the real number.

And say where those hours went. Most owners skip this part. If three hours came back and the person just got three more hours of the same work piled on, you've confirmed everyone's fear. If those hours went to calling back the leads that used to go cold, or finally cleaning up the customer records, or leaving at five on Fridays, say that.

Skepticism doesn't drop because you promised AI would be helpful. It drops because the paralegal got Friday afternoons back and everyone can see it.

Be just as public about what didn't work. "We tried it for scheduling and it was a mess, so we stopped." That builds more trust than any success story, because it proves you won't force a bad tool on them to save face.

What this looks like after 90 days

Realistically: one or two tools in daily use by most of the team, a handful of tasks that take a third of the time they used to, and a staff that treats the next AI announcement like a normal software update instead of a threat.

That's it. Nobody got transformed or replaced; a few annoying things got less annoying, and your people trust you a bit more because you were straight with them.

If you'd rather have someone map this out before you pick the pilot, that's what the AI Audit is for. One week, $750 fixed fee: I sit with your team, write up the ten workflows they repeat most, and tell you which ones are worth handing to a tool and which aren't. The pilot person and the pilot task usually fall right out of that list. Read about the AI Audit.