AI chatbot for customer service. We said “No” and the bot replied the issue was resolved
Uber Eats delivered someone else’s groceries to us, and the support chat closed the case after our “No”. When an AI chatbot must hand over to a person, and a ten minute test.
In this article
In short
At the end of October Uber Eats delivered someone else’s groceries to us, and after we answered “No” the bot said the issue was already resolved. In a Gartner survey of 350 executives at large firms, cutting staff for AI did not go together with better returns. An AI chatbot can answer simple questions fast, but complaints, money and the word “No” should go to a person.
On Friday evening, 30 October, an Uber Eats courier brought us a bag of groceries. Not one item was ours. We reported it in the app straight away, with a photo.
On Saturday at 12.51 we asked in the chat what happens next. The app said it was connecting us to an agent and that it would take about 4 minutes. Before anyone joined, the chat was closed. The bot asked whether the issue was resolved. We answered “No”. A minute later the next message said the issue had already been resolved.
We kept screenshots of the whole chat. This is not about one company. It is simply what an AI chatbot in customer service looks like from the customer’s side when nobody thought about one word.
What went wrong in that chat
The bot did what it was set up to do. It asked a question and closed the case. The trouble is that our answer changed nothing. Asking “is your issue resolved” only makes sense if the answer “No” triggers something. Here it triggered nothing.
For a customer this can be worse than having no chat at all. Without a chat you call or send an email and wait. With a chat that closes the case after “No”, you feel brushed off. People remember that longer than the wrong order itself.
The worst part is that in the company’s reports this chat probably looks fine. Case closed in a minute, customer answered the question. If you measure a bot only by the number of closed cases, chats like this will look like success. It is better to count how many cases customers had to open a second time.
Firms cut staff for AI and the returns did not follow
In May 2026 Gartner described a survey of 350 executives at large organisations that were piloting or using autonomous systems such as AI agents. About 80 percent of them had reduced their workforce. Yet firms with strong returns and firms with weak or negative results cut staff at almost the same rate. Gartner analyst Helen Poitevin put it this way. Workforce reductions may create budget room, but they do not create return (IT Brief, 6 May 2026). The firms with better results were the ones investing in people’s skills and in how the work is organised.
Orgvue surveyed 300 HR managers in the US. In that study, 32 percent of firms that cut staff expecting savings from AI later had to rehire (TechIntelPro, 14 March 2026).
These are large American firms, not a local garage or an accounting office. The lesson still fits a small business. AI does part of the work well, but it rarely replaces the whole person who talks to customers.
When a chatbot must hand over to a person
When you set up a chat, spend more time on when the bot should stop than on what it should say. Decide this before launch and write it down. The conversation goes to a person when one of these signals appears.
- the customer answers “No” or asks the same thing a second time
- the chat mentions a complaint, a refund or money
- the customer asks for a human, even rudely
- the bot does not know the answer and starts guessing
- the topic is personal data or a contract
A handover must not end the conversation. If nobody is in the office, the bot says honestly when someone will reply and leaves the case open. Only a person or the customer can close it.
Where AI in customer service really helps a small business
It works best where questions repeat and the answer is already written down somewhere. Opening hours, directions, order status, the next free appointment, which documents to bring. In the evening nobody at the firm answers these, and the customer wants to know now.
The second good role is behind the scenes. AI reads the message and drafts a reply, and a person sends it. The customer gets an answer sooner, and someone still stands behind every sentence. As a first step this is often better than a bot that talks on its own.
Before you paste customer emails into an AI tool, check what happens to that data. We wrote about it in Client data in ChatGPT. If the chat collects names, phone numbers or a description of the problem, agree the storage rules with a lawyer.
How to test your own bot in ten minutes
Open your website as a customer and send the chat three messages. First a plain question about opening hours. Then a complaint, for example “I received the wrong order”. Finally answer “No” when it asks whether the issue is resolved. Check whether anyone at your firm gets a notification after that last word.
If nobody does, that is the first thing to fix. And if you want to go through customer service in your firm with someone from outside, book a free AI audit. It is a conversation of about 30 minutes, and afterwards you get a written list of places where AI can save time, with notes on client data.
What would you like to change in your business?
Leave your number. We will talk about your process and see whether we can help. The first call is free.
The service this text is about
Automation and AI / OCRAutomation of document re-typing, OCR and checking rules, and your data stays with you.