Straight answer: an AI "employee" (a chatbot, lead-qualification agent, or automated assistant) costs far more than the monthly subscription once you count the hours you spend checking its work, fixing its mistakes, and rewriting what it produces so it doesn't sound like a robot wearing your logo. For most small businesses that hidden cost runs to five or ten hours a month, which at your own billable rate is often more than hiring a real virtual assistant would have cost in the first place.
The number on the invoice is never the number that matters
Every AI vendor sells you the same story: replace a £2,000-a-month employee with a £29-a-month bot. I fell for it too. Last year I set up an AI agent to qualify inbound leads on my own site before they hit my calendar, using a stack that cost £47 a month between the chatbot tool and the automation that fed it into my CRM. On paper that's a rounding error compared to a part-time assistant.
What nobody tells you is what happens in month two, when the agent starts confidently telling a warm lead that I offer services I don't, because it pulled an answer from an old page it wasn't supposed to reference. I spent three hours untangling that, another two rewriting its scripts so it stopped improvising, and then had to personally email the prospect to apologise for the confusion. That's five hours of my time, at whatever you think my hour is worth, spent fixing a £47 tool.
What happened when I ran the experiment for 90 days
I kept a log, because I wanted to know if this was a one-off or a pattern. Over 90 days:
- The bot handled 214 initial enquiries without any human involvement.
- Of those, 38 needed correction after the fact because the answer was wrong, out of date, or too generic to be useful.
- Fixing those 38 took roughly 9 hours across the quarter, which is 3 hours a month.
- Retraining the bot after each fix (updating its instructions so the same mistake didn't repeat) added another 2 hours a month.
- Total hidden labour: about 5 hours a month, on top of the £47 subscription.
At my own rate that 5 hours is worth considerably more than the subscription fee. A part-time VA in the Philippines, at £8 to £12 an hour through most freelance platforms, would have cost me £40 to £60 a month for the same 5 hours, plus she would have caught her own mistakes without me noticing. A UK-based VA doing the same job would run £15 to £25 an hour, so £75 to £125 a month for those hours alone, before you add her actual output.
The AI wasn't cheaper. It just moved the cost from an invoice I could see to hours I didn't track until I forced myself to.
Where the maths flips in the AI's favour
This isn't an argument against AI employees. It's an argument against buying one without doing the sum. There are three situations where an AI agent beats a human hire outright:
- Pure volume, low stakes. Answering "what are your opening hours" or "do you ship to Ireland" a thousand times a month. A wrong answer here costs you nothing serious, and no human wants that job anyway.
- Overnight and weekend coverage. If your best leads come in at 9pm because that's when your ideal customer finally sits down after their kids are in bed, an AI agent that catches them then is worth its fee even if it's clumsy, because the alternative is silence until Monday.
- First-pass sorting before a human takes over. Use the AI to tag and route enquiries, not to close them. That's the version I run now, and my monthly correction time dropped from 5 hours to about 1.
Notice the pattern: the AI earns its keep on volume and speed, and loses money the moment you ask it to represent your judgement, your pricing, or your tone. That third one is where most small businesses trip up, because they don't think of tone as a cost until a customer replies "this doesn't sound like you at all."
The uncomfortable bit nobody selling you the tool will say out loud
Here's the part that made me uncomfortable enough to write this post. Buyers, especially anyone spending more than a few hundred pounds with you, can tell when they're talking to an AI wearing your brand's coat, and a growing number of them trust it less for exactly that reason. A survey-backed piece in Forbes on AI-generated customer interactions has flagged the same pattern I saw in my own inbox: polish without a person behind it reads as evasive, not efficient. People don't mind AI doing the boring bits. They mind AI pretending to be a relationship.
That distinction matters more than any productivity stat. If you've spent years building a brand voice the way GoPro built theirs around real customer footage, not scripted polish, an AI agent that flattens every reply into the same neutral corporate tone is quietly undoing years of work. I've seen it happen to my own email replies before I caught it: technically correct, grammatically clean, and completely missing the bluntness people come to me for.
The actual math you should run before you buy anything
Do this before you sign up for any AI tool that touches customers directly:
- Estimate your realistic error and correction rate. Ask the vendor for their actual accuracy figures, not their marketing copy, and assume it's worse in practice.
- Multiply the hours you'll spend fixing errors by what your hour is worth, not what you wish it were worth.
- Add that to the subscription price to get your real monthly cost.
- Compare that real number against a part-time human doing the same job, including a UK VA at £15 to £25 an hour or an offshore VA at £8 to £12.
- Only proceed if the AI is still cheaper once the correction hours are counted honestly.
If you want to see this play out as a number rather than a guess, build it as an actual calculator rather than a gut feeling. I've written before about how interactive calculators convert better than static content because people trust a number they worked out themselves over one you handed them, and this is exactly that situation: put your own hourly rate and error estimate into the sum and look at what comes out.
Where I landed after the 90 days
I kept the AI agent. I just stopped asking it to do the one thing it was bad at, which was sounding like me under pressure. It now sorts and tags, a human (me, or my assistant) closes. I also stopped running it on my highest-value pages entirely, the ones where a prospect is deciding whether to spend four figures with me, because that's exactly where the tone mismatch cost me the most. Email marketing is a good comparison point here: the businesses getting this right, like the ones running smart automation through platforms similar to how Mailchimp built its brand around approachable, human-sounding automation rather than sterile corporate copy, tend to keep a human eye on anything that touches money.
If you're weighing this up for your own business and the sums aren't obvious, it's worth thirty minutes with someone who's already run the experiment rather than guessing your way through month one the way I did. I cover the full breakdown, tool by tool, in my piece on what an AI consultant costs and what you get for it, which is the same maths in more detail.
Want AI doing the heavy lifting in your marketing?
I build the systems that handle the boring 80 percent, so you get your week back. Done properly, with the human kept in.
The bit most guides skip: what happens to your team when the AI is "always on"
There's a second cost that never shows up in any pricing table. When your AI agent works nights and weekends, your customers quietly start expecting your humans to as well. I noticed replies from my own inbox creeping later into the evening once people got used to instant answers from the bot at 10pm. Nobody set that expectation on purpose. It just happened, the way scope creep always happens, one reasonable-seeming reply at a time. If you're rolling out an AI employee, decide now, before launch, what hours your humans are off, and say so in the same message the bot sends. "I'm an assistant and I'm here 24/7, our team responds 9 to 5 Monday to Friday" costs you nothing and saves you from training your customers to expect a human clock that never stops.
This is the same discipline good content strategists have always used. Joe Pulizzi's whole approach to content was about consistency you can sustain, not maximum output regardless of cost, and the same rule applies to AI employees: sustainable beats impressive every time.
A short checklist before you hire your first AI agent
- Write down what a wrong answer costs you in this specific role. A wrong shipping estimate is cheap. A wrong price quote is not.
- Test it on your worst-case customer question before you test it on your easiest one.
- Set a monthly limit on how much correction time you'll tolerate before you pull it back to a narrower job.
- Keep it away from anything where your brand voice is the actual product, not just the delivery method.
- Tell customers plainly when they're talking to it, and when a human takes over. Nobody minds AI. People mind being tricked into thinking it's a person.
Free resource: The AI Agent Stack Decision Cheat Sheet.
Frequently asked questions
Is an AI employee cheaper than hiring a virtual assistant?
Often not once you count your own correction time. In my own 90-day test, a £47-a-month AI agent cost roughly 5 extra hours a month in fixes and retraining, which at most small business owners' rates works out more expensive than a part-time offshore VA at £8 to £12 an hour doing the same job.
What's the biggest hidden cost of running an AI agent in a small business?
Time spent correcting errors and retraining it, not the subscription fee. Vendors advertise the price of the tool. Nobody advertises the hours you'll spend fixing what it gets wrong, which is usually the real number that matters.
Where does an AI employee save money?
High-volume, low-stakes work: answering repetitive questions, catching leads outside office hours, and sorting or tagging enquiries before a human takes over. It loses money the moment you ask it to represent judgement, pricing decisions, or your specific brand voice.
Should customers know they're talking to an AI, not a person?
Yes, plainly and upfront. Trust research keeps pointing the same way: people are fine with AI doing the boring first pass, but they trust a business less if they realise afterwards they were misled into thinking a human was on the other end.
I go much deeper on this in the AI marketing guide.
Related reading: Business Lessons from Dale Carnegie and Business Lessons from Mary Barra.
