- operations
- team
- small business
Hiring when part of the job is already automated
If you automated anything this year, the role you were about to advertise no longer exists. A one-page inventory and a first week built back to front.
You hire someone for customer support, they turn up on Monday morning, and half the tasks you were going to hand them no longer exist. Appointment reminders go out on their own, email arrives pre-sorted, and the ten questions you get every day are answered before anyone reads them. Nobody warns you about this, because the job ad was copied from the one you posted three years ago — back when a person did all of that by hand.
If you've automated anything in the last few months and you're about to hire, this is the conversation worth having before the ad goes live. It covers two things: which job you're actually filling, and how to design a first week that isn't spent teaching someone to do manually what a machine already does.
The job you describe versus the job that exists
A typical small-business vacancy is a list of verbs: answer, schedule, remind, file, send quotes, follow up. When part of that list is already automated and the text never gets updated, three things go wrong.
You attract the wrong person. Someone who applies for "answering calls and booking appointments" is expecting execution work, and what you'll hand them is review-and-exceptions work — precisely the things the agent can't resolve. Those are two different candidates.
Then comes the month-two slump. Someone used to measuring their day in completed tasks finds a half-empty inbox and an uneasy feeling of not contributing much. If nobody has explained that their value sits in the fifteen odd messages rather than the two hundred normal ones, they read it as being surplus to requirements.
The third is the expensive one: you hire for a workload you no longer have. It's the classic mistake of sizing this year's team from last year's numbers.
The one-page inventory, before you post the ad
Before writing anything, take a sheet of paper and sort the role's tasks into three columns. Half an hour is plenty.
- An agent handles it end to end. Confirmations, reminders, FAQ answers, sorting whatever comes in. Nobody touches it unless something breaks.
- An agent drafts it, a person approves it. Quotes built from your price list, replies to complaints, overdue-payment nudges. The agent writes, someone presses send. This column is the heart of the new role.
- A person does all of it. Awkward calls, angry customers, negotiating with a supplier — anything needing judgement or a face.
Now add up the hours in columns two and three. That's the job you're hiring for, and it's almost always a different — and more senior — role than the one you were about to advertise. Write the ad from there: "review and approve", "handle whatever falls outside the script", "improve the system's answers", instead of "manage the diary".
One warning about that list: build it with whoever does the work today, not from memory. The gap between what you think reception does and what it actually does is always wider than it looks from the office.
The first week, back to front
Classic onboarding runs easy-to-hard: simple tasks first, tricky ones later. With part of the work automated that order collapses, because the simple tasks aren't available to practise on. You have to run it the other way round.
Days one and two: read, don't touch. Have them go through real conversations from the past few weeks — the ones the agent handled and the ones that ended up with a human — and note the exact moment each switched over. It's the fastest way to understand where the boundary sits. As a bonus, you get a list of cases you stopped noticing years ago.
Day three: the exception rules. Walk them through what the agent does when it doesn't know something: who it pings, how quickly, what it never promises. If those rules are written down, this is a one-page session. If they aren't, you've just found the most important item on your own to-do list.
Days four and five: approving with a spotter. Let them start reviewing and sending what the agent drafts, with someone looking over their shoulder on the first day. This is where the house tone gets learned properly — far faster than writing from scratch, because there's a draft in front of them to argue with.
Week two: start fixing the system. When they spot a weak answer, they shouldn't just patch that one thread — they should write the good answer into wherever the agent reads from. That's the mental shift separating someone who plugs holes from someone who closes them.
Teaching someone to supervise is harder than teaching them to do
Here's the uncomfortable bit. Reviewing an agent's work is harder than doing the work, and it's duller. When something comes out right ninety times running, number ninety-one gets approved unread. That's human, and it happens everywhere.
Three concrete things offset it. First, keep the approval queue short and give it a fixed slot in the day rather than letting it trickle in. Second, be explicit about which messages get a glance and which get read in full — large amounts, new customers, anything resembling a complaint. Third, ask them, out loud, to argue with the agent. A new person sees oddities the rest of you normalised months ago; if the day-one message is "this all works, just hit send", you lose that fresh pair of eyes permanently.
When to skip all of this
Two clear cases.
If nobody on your team really understands the process you want to automate, hire first and automate second. Automating a mess only gives you a faster mess, and nobody will be able to tell whether the agent is getting it wrong.
And if you're considering not filling the vacancy at all because "the agent covers part of it", check what happens to the rest first. Work that isn't automated doesn't vanish; it redistributes to whoever is left, usually as interruptions. An agent that saves ten hours of repetitive messaging doesn't replace the person who also chases suppliers and cashes up.
The upside nobody mentions
The most useful side effect of hiring while part of the work is automated is the free audit. Someone arriving from outside who spends a week reviewing what your agent does will tell you, unfiltered, which replies sound robotic, which rules contradict each other, and which common case nobody accounted for. You can't see it any more — you've been reading those messages for months.
Keep that list. It's worth more than their first performance review.
If that's where you are — about to hire, unsure which parts of the role are already covered by what you've built — the operations agent is a decent way to see where the repetitive work ends and the work that needs a person begins.
