This is more or less how the conversation went in my head. Which is to say, it’s more or less how the conversation actually went.
“We have to do AI. Everyone is. So the chatbot will help us. It’ll pick up some slack for the customer service department.”
I hear this a lot. Not always in those exact words, but the shape of it is always the same: AI as inevitability, chatbot as cost-cutting measure, customer service team as a line item to be quietly eliminated.
“No it won’t,” I said.
“Why not?”
“Because the training data you fed it is your website content.”
A pause. The kind that means they hadn’t considered this a problem.
“And what’s wrong with that?”
Here’s what’s wrong with that.
Your website — the one you’re now using as the foundation for a tool that’s going to interact with real customers in real time — is out of date by at least twenty years. It was built when your business looked different, your services were different, and your customers expected less.
Your website probably still has a phone number that rings a desk nobody sits at anymore, or a phone number that was disconnected when you went remote during the pandemic. It might still reference a product you discontinued in 2011. All things I’ve seen, by the way.
I decide to reply with a more polite summary instead: “That content — and that website — are out of date by twenty years at least. So the chatbot isn’t going to provide support. It’s going to provide slop, because it was given slop to train on.”
More silence.
I kept going, because I figured that’s what they wanted.
“Your customers are going to get angry and frustrated, and they’ll tell the chatbot they want to speak to a human. But there won’t be a human to connect them to, because you’ll have fired them all.”
This is the part where everyone starts to squirm in their seats like they’re feeling uncomfortable, because they are feeling uncomfortable because it’s true, and everyone in the room knows it’s true, and nobody wants to say it out loud.
The chatbot isn’t a supplement to the customer service team. In this scenario, it’s a replacement for it. Which means when the chatbot fails — and it will fail — there’s no fallback.
“So your clients will swear at the robot. Which, as a robot, is incapable of taking offence. But it will have been trained to close the conversation the first time it’s called a piece of shit fucking clunker.”
I’ll be honest: at that point, the bot probably deserves it. Whether I should have sworn in the boardroom is another argument entirely.
“So you’ll have a pretend-hurt robot and an actually pissed off customer. And that will be a great look for you and your brand — just like your website from 2005.”
I was shown the door.
I want to be clear about something: I am not anti-AI. I use it. My team uses it.
It is genuinely useful for a significant number of things, and the list is growing, and fast.
But AI is a tool or set of tools, or toolbox, and like every tool or toolbox, it performs exactly as well as the quality of inputs and the intelligence of implementation allow it to.
A chatbot trained on outdated, inaccurate, or simply bad content will produce outdated, inaccurate, and bad responses — but cheaply, and very confidently, and at scale, to every single customer who encounters it.
Garbage in, garbage out has been a principle of computing since before most of us were born. AI doesn’t change that. It just creates it quicker and makes the garbage sound more authoritative.
If you want to introduce AI into your customer-facing operations, I’d strongly suggest starting somewhere else: fix the content first. Update the website. Make sure the information the system will learn from is accurate, current, and actually representative of what your business does and how it works today.
Then talk about the chatbot.
And maybe keep at least one human on the payroll. Just in case.
Mark Debono is the founder of Systemato and an independent marketing and communications consultant.
He has strong opinions about AI, stronger opinions about bad websites, and will tell you both to your face, for a fee.
