I’ve always noticed that when people are very keen to tell you about their “AI,” it usually pays to ask one or two awkward questions. In customer experience technology, this has become something of a sport.
Every vendor seems to have an “AI-powered” solution. But lean in, and ask: “When you say AI… what kind are we talking about?” That’s when the room goes quiet, the sales deck gets shuffled, and the truth is often far less glamorous than the label.
This blurring of definitions — known as AI washing — is spreading fast. And you see it most clearly in places like chatbots, routing, analytics, and even personalisation.
Example One: Three Chatbots, One Label
Chatbots are the poster child of AI in CX. But under the hood, they come in three very different flavours.
- Scripted Chatbots – Decision trees with a friendlier face. Reliable for FAQs, but quickly fall apart if a customer strays off-script.
- Generative AI Chatbots – Powered by large language models. They can improvise and sound natural, but sometimes hallucinate answers with alarming confidence.
- Agentic AI Chatbots – The real deal. They don’t just chat, they act: rebooking flights, processing refunds, or updating accounts without human handover.
The problem? All three are often lumped under the same “AI” banner. Which means a buyer expecting agentic capability might end up with little more than an expensive FAQ page.
Example Two: AI Routing vs. Old-School Rules
Another favourite playground for AI washing is call routing.
Legacy routing has been around for decades. “If customer selects option 1, send to Sales. If they’re VIP, send to Priority Queue.” Simple, effective, but not exactly cutting-edge.
Now, vendors rebrand this as AI routing. True AI routing should analyse intent in real time, predict needs from history, and route customers to the best-fit agent or channel. But in many cases, it’s still the same rules engine with a new coat of paint.
It’s a bit like me putting on a tracksuit and insisting I’m an Olympic athlete. The outfit has changed, but the performance hasn’t.