Hannah found out about the refund on a Tuesday, from an angry customer. Her small marketing agency had switched on an AI assistant three weeks earlier to draft client emails, and it had cheerfully promised a full refund on a project that, according to her contract, didn't qualify for one. Nobody had checked the draft. Nobody had been asked to. The tool did exactly what tools do. It sounded confident. Hannah is a made-up owner, and so is everyone around her, but if you run a small business and you've bought an AI tool lately, you may recognize the feeling in her stomach.
This post is about the gap that causes moments like that. It isn't the technology. It's that most small companies buy the tool and forget to hire the person who runs it. Here's what that person does, where to find one, and how to start without a big budget or a bigger headache.
The Tool Wasn't The Problem
Hannah called Joao, a contractor who'd handled her operations for two years. He listened to the whole story and then asked a question she didn't enjoy. "Who owns it?"
"The AI?"
"The workflow. Who decides what it's allowed to write, who reads it before it goes out, and who fixes it when it's wrong?"
Silence. She'd assumed the software would handle it. "Everyone does," Joao said, "at first." It can draft all day, but it doesn't know your refund policy unless someone tells it, can't tell when a client is upset, and can't notice it got something wrong. Somebody has to stand in the gap between what the tool writes and what customers read. Rarely the founder, because the founder is already doing six other jobs.
What An "AI Operator" Actually Does
Hannah asked what the job was even called. Joao shrugged. "Nobody's settled on a title. I say AI operator." Not a computer scientist, not a manager. A careful person who treats the tool like a fast, eager, slightly reckless new hire.
Joao ticked the duties off on his fingers. One, they write and tidy the instructions so the tool stops guessing about your policies. Two, they read everything before customers do, at least until trust is earned. Three, they keep a plain log of mistakes, because patterns show up fast. The tool always muddles dates, or gets pushy in the third paragraph. Four, they keep your data clean. Five, they help you decide which jobs it should never touch.
"It's mostly editing and noticing," Joao said. "People who are good at it are the ones who read everything twice."
"So, a very patient proofreader," Hannah said. "With opinions," said Joao.
Joao had a story for this one. A friend's company had let an AI tool summarize its weekly sales calls, and for a month nobody noticed it kept getting the follow-up dates wrong by a week. Reps chased leads late, and deals quietly cooled. Nobody was careless. Nobody was looking.
That doesn't sound glamorous. It's also what separates the businesses that save hours from the ones that save hours and then lose a client.
Why Remote Talent Fits This Job So Well
"Does this person have to be in my office?" Hannah asked. Joao laughed. "Or your time zone." The work is reading, checking, organizing and improving, all of which can be done from anywhere, and a lot of it benefits from happening while she's offline.
Cost matters here too, because you're adding a checker, not a department. Hiring in another market can bring that price down without dropping quality, as long as you hire carefully and train properly. Hannah's first question was whether she could afford it. Her second was how she'd ever find someone she'd trust with customer emails. Her first reviewer, Teodoro, worked nine hours ahead of her, which turned out to be a gift. She'd go to bed with forty drafts waiting and wake up to forty drafts checked, with a short note on the three he'd rewritten.
How did she find someone she'd trust? By testing, mostly. Reliability, attention to detail and clear written English mattered far more than any technical degree.
Hannah asked the obvious thing before moving on. Couldn't she skip the hiring and just buy a better tool? "Buy the better tool," Joao said. "And still hire." Better software makes fewer mistakes, but it makes them with the same confidence, and it still doesn't know your business.
Three Roles Worth Considering
Joao sketched it on the back of an invoice. "You don't need all three," he said. "Pick the one that matches your biggest headache."
Role | What they do with AI | A good first test |
|---|---|---|
Virtual assistant | Runs the tools day to day, drafts, reviews, files and flags problems | Edit five AI-written emails against your policy |
Reviewer or QA person | Checks outputs for errors, tone and accuracy before they go out | Find the planted mistakes in a sample batch |
Technical specialist | Connects tools, builds automations, handles data pipelines | Sketch how they'd connect two of your systems |
If you want the technical end, remote AI engineers can build the integrations and automations while a human reviewer keeps an eye on what comes out the other side. Most small teams start with the first two. The third can wait until the volume earns it.
Set The Guardrails Before The First Task
Hannah's refund disaster would have been avoided by a single page of rules. When Joao wrote hers, it was embarrassingly short.
Joao dictated while she typed. Which customer information can go into an AI tool, and which can't. Nothing with payment details, ever. Which messages a human must read before they're sent, and for refunds, legal wording and anything angry the answer was always. What to do when unsure, which is ask. And who's allowed to change the instructions. Joao printed it and taped it to her monitor. "That's it?" Hannah asked. "That's it. If it needs two pages, nobody reads it."
Hand the page to whoever you hire on day one, and ask them to argue with it. The good ones will. If you're working with people in other countries, remember that employment and contractor rules differ from place to place. This post is general information, not legal advice, so have someone qualified look at your setup.
A Thirty-Day Pilot You Can Copy
You don't need a grand strategy. Joao suggested a small experiment, and Hannah ran it almost exactly.
In week one, she picked one narrow job, client emails, and wrote down what a good one looks like, with three real examples, including the refund one she still can't look at. In week two, she hired a reviewer on a short paid trial and gave them the one-page rules. They checked every AI draft and logged every mistake. In week three, they fixed the instructions based on the log, and the number of fixes dropped enough that Hannah stopped wincing at her inbox. In week four, Hannah let the reviewer approve the simple emails alone, while the tricky ones still came to her. She hovered over that first batch like a nervous parent.
"Start with one job," Joao said. "If you can't make one job reliable, ten jobs will just be ten problems."
How You'll Know It's Working
Hannah tracked four things on a sticky note. How many AI drafts needed a fix before sending, because that number should fall. How long a typical email took from request to reply. How many messages had to be escalated to her, which should shrink as the instructions improve. And how many mistakes reached a customer. Her target for that last one was zero. In month one it was one, a typo, and she taped the sticky note up anyway. Teodoro's weekly note, three lines long, turned out to be the most useful report she got all quarter.
No movement after six weeks? Blame the instructions before the tool or the person. Joao's rule of thumb is that nine times out of ten, that's where the trouble lives.
Where To Learn More
If you've never hired a remote assistant before, the free guides at Hire A Virtual Assistant cover training, tools and contracts in plain English, which is handy before you write your own one-page rulebook.
Back To Hannah
A couple of months later, a client wrote to say how fast the agency had become. Hannah smiled, because she knew the truth. The AI hadn't made her fast. The person reading everything twice had made her safe, and safe was what let her go fast.
She still flinches when a customer's name lights up her phone on a Tuesday. But now when a draft goes wrong, it goes wrong on someone's desk, gets noticed, gets logged and never reaches the customer. That's the whole difference.
If you've bought the tool, hire the owner. Start with one job, write down the rules, and read everything twice.
Copyright © 2026 California Business Journal. All Rights Reserved.
For California Business Journal Disclaimers, go to https://calbizjournal.com/terms-conditions/.
