AI employees for a service business: what they can do today, and what they still can't

Every expert business owner has now been told some version of the same pitch: hire an AI employee and stop paying people to do repetitive work. Some of that is real. A lot of it is a screen recording of a demo that would fall apart on your second Tuesday.

If you sell a $5,000 coaching program or a $15,000 consulting engagement, the question is not whether AI is impressive. It is which specific jobs in your business it can take over without you checking its work every hour.

Start with the job description, not the tool

The fastest way to waste money on AI is to buy a tool and then look for work for it. Do it the other way round. Write down every recurring task your team does in a normal week, how long it takes, and what a mistake costs.

You will usually find three kinds of work. Work with a clear input and a clear output, like turning a call recording into notes.

Work that needs judgment but follows a pattern, like replying to a support question. And work where the relationship is the product, like a coaching session or a sales call.

AI is strong at the first kind, decent at the second with supervision, and a liability at the third. Most disappointment comes from pointing it at the third kind first because that is where the founder feels the most pain.

What it handles well today

Here is the work we see AI doing reliably in service businesses right now, with a human glancing at the output rather than redoing it.

• Call summaries. Recording a sales or onboarding call, then producing notes, objections raised, next steps and a CRM update. This used to be a setter's least favourite hour of the day.
• First drafts of content. Turning a podcast episode or a long voice note into an email, a set of social posts or a newsletter outline in your existing voice, if you give it enough real samples.
• Lead research. Checking an applicant's website, socials and business size before a call so the closer walks in prepared.
• Inbox triage. Sorting incoming email and support messages into urgent, routine and spam, and drafting replies to the routine ones for approval.
• Reporting. Pulling ad spend, booked calls, show rates and closed revenue into one weekly summary instead of five open tabs.

Notice what these have in common. The output is checkable in under a minute, and a mistake is cheap.

A clumsy call summary costs nothing. A clumsy reply to an unhappy client who paid you $12,000 costs a lot.

Give AI the work where a mistake is cheap and checking is fast. Keep the work where the relationship is the product.

Where it still breaks

The weak spots are predictable once you know where to look. AI is confident when it is wrong, which is a bad trait in anything customer facing. It will state a refund policy you do not have, promise a feature that does not exist, or invent a date.

It also struggles with context that lives in your head. It does not know that one client is a referral from your best customer, or that another has complained twice already. Unless that history is written down somewhere it can read, it will treat everyone the same.

And it cannot sell a high ticket offer on its own. A buyer about to spend $10,000 wants to feel understood by a person who has solved the problem before.

AI can prepare that person brilliantly. It cannot replace them, and buyers can tell when you try.

How to roll it out without a mess

Treat an AI employee exactly like a new junior hire. Give it one job, a written process, real examples of good work, and a review period where every output gets checked.

Say your setter spends ten hours a week writing pre call research notes. Hand that single task to AI for a month. Have the setter review every note for the first two weeks, then spot check.

If the notes are good, you have bought back roughly forty hours a month for one person. If they are not, you have lost very little.

Only then move to the next task. The businesses getting real value are not the ones with ten AI agents. They are the ones with three that each do one boring job properly.

The honest scorecard

A sensible rule for now: let AI draft, sort, summarise and report. Let humans decide, promise, apologise and close. Anything that commits your business to money, a deadline or a policy should pass a human first.

That line will move. Models get better every few months, and tasks that need supervision today will not need it next year. But the businesses that win will be the ones that moved one job at a time, measured the result, and kept a person where trust is the thing being sold.

The goal is not a company run by software. It is a company where your best people spend their week on the work only they can do.