We Stopped Laughing At The Chatbot

We used to joke with our friends about how bad the chatbot was, because we all knew the trick: if you held on long enough, or complained enough, it would eventually put you through to a real human. I am not laughing anymore, and I am currently living through an example so bad that it has gone all the way round to being funny again.

Four channels and not one answer

I have a problem with a big company, and I have now been through every channel it has.

I ordered something that combines a product and a service. They came to deliver the service and they were supposed to give me the product with it, but the product does not work and needs to be replaced. On their system it looks like the service has been completed, so the ticket has been closed, and it turns out that you can complete a service but there is no way to get a product replaced once that has happened. Every time I raise it, it is automatically closed again.

I started on WhatsApp and was met by a chatbot. I tried typing my problem in my own words, but it just kept making me pick from a list, so I picked an item, could not find what I really wanted, went to "other" and was told, sorry, I can't help you. That was about 5 minutes gone. I thought I would try again, so I went in through a different menu option and managed to get to something close to my problem. It asked me to authenticate myself, which I did, but for some reason my account was not linked to my ID, so it could not help me that way either, and that was another 15 minutes.

So I called the contact centre, which put me into an automated process that uses my voice to authenticate me, and that failed as well. I called again, chose a different option and waited 15 minutes to get through to a person. Another 15 minutes later they could not resolve my problem, but they guaranteed that somebody would call me back. I waited 72 hours and nobody did. I called again, spent another 20 minutes and was told somebody would call me that day, and after a few days of going round that cycle I sent an email instead.

The automated response came back immediately, and 48 hours later a real response arrived telling me to go into a store. At the weekend I found the time to go into a branch, where I was told they could not resolve anything in store. The most hilarious part is what happened next, because I sat down with a customer service agent who typed an email to somebody else in the organisation and confirmed to me that the issue would be resolved within 48 hours.

Of course it wasn't, and many loops of the same process later it still isn't.

The money moved before the outcome did

When I talked to other people about this it turned out to be a common experience, and there is a number that explains why. Gartner surveyed 199 customer service and support leaders this year and found that their AI spending has surged by 38% while the overall budget for the function has risen by just 2%. Those two numbers only fit together if the money is moving out of people and into technology, and Gartner's own analyst Kim Hedlin says exactly that: leaders are "increasingly redirecting spending away from labor and overhead and instead toward technology. The challenge is ensuring those investments produce measurable business value."

Moving the spend is a fair decision if the customer's problem gets solved more often as a result. My experience says the technology has got ahead of itself, and more than that, it has got ahead of anyone actually understanding the problem from the customer's point of view.

The humans failed just as much as the AI

It would be easy to make this an article about a bad chatbot, but look at what actually happened. The chatbot failed me twice and the voice authentication failed me once, and then every person I reached could not solve it either. The person on the phone could only promise a call back, the person answering the email could only send me to a store, and the person in the store could only send an email.

Data & Society published research in July on why a human in the loop is not the same thing as oversight. It was written about people supervising AI agents, but the test works just as well for the person at the end of a customer service process. A person needs adequate knowledge of what the system can and cannot do, sufficient observation of what it is doing and meaningful control over it, and the researchers are clear that "intervention becomes meaningful only when users have enough knowledge, visibility, and control to act". Nobody I dealt with appeared to have the information or the authority to fix my problem, which means the human in the loop was just another step in the loop.

Don't automatically blame the AI

So my first takeaway is not to automatically blame the AI, because it might be that the business was the problem before the AI arrived. As the end customer I have no way of knowing whether the human operators were solving these problems before, and I doubt the company knows how it is doing now. As far as their system is concerned my service was completed and my ticket is closed, which is exactly the difference between closing a ticket and solving a problem. When Dataiku asked 685 CIOs this summer, 72% said they cannot consistently confirm whether their AI agents are delivering the business outcomes they were built for.

The second is that if you are implementing AI, you have to really understand the pains of the existing process first, and be honest about whether you are going to fix them or just expose them even more. For some businesses AI will never achieve the outcomes, because the business itself does not understand what its customers really need.

The third is about the people you leave in the process.

What I would do before automating anything

I would go through my own customer service as a customer, with a real problem, on every channel I offer, and see where I end up. I would work out where the process was already failing before putting AI on top of it. And I would give the people in the loop what they need to actually resolve something, then measure whether the problem got solved rather than whether the conversation ended.

When did you last test if your company can fix a problem for a customer?

Sources

Gartner. Gartner Survey Finds AI Spending by Customer Service Leaders Has Surged by 38%, Despite Overall Service and Support Function Budgets Rising by Just 2%, 26 August 2026, for the 38% against 2% and the Kim Hedlin quote. 199 service and support leaders, surveyed April to May 2026. Gartner advises the leaders it is surveying here. Figures as at 26 August 2026.

Data & Society. The Oversight Fallacy: Why AI Agents Require More than Humans-in-the-Loop, Samir Passi and Ranjit Singh, 29 July 2026, for the conditions of effective oversight. It draws on fieldwork in a computational biology laboratory, so I am applying its test to customer service rather than quoting a finding about customer service.

Dataiku. 81% of Global CIOs Say They Have Lost Oversight of Their Own AI Agents, 24 September 2026, for the 72%. The Harris Poll surveyed 685 CIOs across 8 countries between 9 and 29 July 2026. Dataiku sells a platform for governing AI, so treat it as commissioned research. Figures as at 24 September 2026.