Share article
The UK’s financial ombudsman has found that a large share of the complaints it currently receives are written largely by artificial intelligence. They tend to be confident and long, yet often completely off the mark. And how did the service deal with it? It picked up the good old telephone and started calling.
When I first read the story, I laughed out loud. Then my smile slowly froze. It dawned on me that this is no longer a joke about chatbots, but the everyday reality of customer care. And it’s only a matter of time before it knocks on the doors of Czech companies too.
37 out of 39
The UK’s Financial Ombudsman Service (FOS) works much like the Czech Financial Arbiter. It’s an independent body that resolves disputes between people and their banks or insurers free of charge. Its chief, James Dipple-Johnstone, recently described to the Financial Times a case that would fit a comedy sketch better than an official file. They received a complaint in which 37 out of 39 references were wrong.
I had to read that twice. This wasn’t one typo or one misquoted clause. Almost everything was wrong.
And it’s no isolated curiosity. In March this year, the service reviewed a small sample of cases and found that up to a third of the responses people sent in were written by artificial intelligence, or at least heavily assisted by it. And it’s not just ordinary people. AI is also used by claims management companies that pursue compensation on behalf of clients, and even by lawyers. Some of their complaints run to more than 200 pages.
Two hundred pages over a single complaint. That’s no longer a complaint, that’s a serialized novel.
Imagine Mr. Novák
(A hypothetical scenario for illustration, not a real case.)
His insurer has rejected Mr. Novák’s claim for a flooded bathroom. It’s eleven at night, he’s angry, so he opens a chatbot and types: “Write me a proper complaint.” A minute later he has three pages full of statutes, court rulings and phrases like “gross breach of contractual obligations.” It sounds as if a lawyer wrote it. Mr. Novák finally feels someone is taking him seriously.
But half of the cited laws don’t apply to his case at all, and one of the court rulings the AI simply made up. And the most important thing, that the technician promised to come and never showed up, got lost somewhere on page two.
Meanwhile, on the other side sits a person who has to check every single reference properly. They spend an hour searching for rulings that don’t exist. I genuinely feel for them. And for Mr. Novák too, who is waiting for a reply and getting angrier by the minute. They both lose.
That’s exactly what the UK ombudsman describes. What really matters gets lost in a flood of inaccurate text, and staff spend their time checking things that have nothing to do with the case.
An answer from the last century
So what did the FOS do about it? It started calling people more. Caseworkers phone complainants to find out straight away what they actually want, before they get buried under pages of generated text.
I’ll admit this made me really happy. The most modern problem in customer care is being solved with one of the oldest tools we have. And someone has finally realized, once again, that a five-minute conversation often says more than two hundred pages of generated text.
AI isn’t the villain
It would be easy to turn this into a story about evil artificial intelligence. But that’s not how it is, and it would be unfair of me to write it that way. The service itself acknowledges that AI can help people write their complaints more clearly and understandably. Plenty of people struggle to describe what’s really bothering them, and AI genuinely helps with that.
So the problem isn’t the tool. It’s that people don’t check anything the AI produces. The result is texts that are needlessly long, don’t make sense, or are simply wrong from start to finish.
What to take away
- A long complaint doesn’t mean a justified one. But it doesn’t mean an unjustified one either. Look for what the customer actually wants, not how many pages they wrote.
- When the text doesn’t make sense, call. One call saves hours of fact-checking, and the customer feels someone is listening.
- Don’t let AI answer AI without a human. Otherwise two machines will end up talking to each other and the customer will be left standing somewhere in between.
- Write clearly. The clearer your terms and replies, the less AI will have to fill in on its own.
We don’t yet know how many complaints like this Czech companies are already receiving. And I’d quite like to find out. If you’re dealing with this in your customer care, get in touch, we’d be glad to write more about it.
The future of customer care may not be about who has the better AI. But about who still knows how to pick up the phone and talk to another human being.
