Why customers are fighting their way out of chatbots, and how the smartest companies are building automation people don’t need to escape.
Everybody’s got a doom-loop story. You call about a double charge, and a pleasant synthetic voice offers to help with your balance, your due date or “something else.” You say “something else.” It offers your balance again. You say “representative.” It asks you to describe your problem in a few words, so you do. It offers your balance. Somewhere around the fourth lap, you’ve stopped being a customer with a question and become a person with a grudge.
The industry has a name for this now, which tells you how common it’s become. Gartner calls it the doom loop, and its latest State of the Customer survey found that a third of customers hit one in their most recent service interaction. It comes in three flavors: phone menus that never reach a person, chatbots that can’t work out what you want and multichannel loops that bounce you from one to the other.
Phone menus deserve a special mention. The modern voice system invites callers to “just say what you need” and then punishes them for doing it. Press one, say “billing,” read out your account number, hear that your call is very important and listen to the same twelve bars of hold music until the line drops. Every one of those steps is a chance to lose the caller, and plenty of callers take it. They hang up, try the website, dial a competitor or simply never call back.
What customers are telling us
The numbers aren’t subtle. When Gartner surveyed nearly 5,800 customers, 64% said they’d prefer companies not use AI in customer service at all, and 53% said they’d consider switching to a competitor over it. Their biggest worry wasn’t privacy or lost jobs, though both made the list. It was that AI would make it harder to reach a human being.
That fear has been earned. In a survey commissioned by the customer service company Parloa, nearly 44% of Americans admitted to shouting “human!” into their phones to escape an automated system, and 17% confessed to swearing at chatbots. A Five9 study found that fewer than half of consumers rated their latest chatbot experience positively, and that among those who gave up and asked for a person, nearly half did so because the bot kept misunderstanding them.
The damage doesn’t end with a ruined afternoon. Researchers writing in the California Management Review note that chatbot frustration erodes loyalty, trust and the likelihood that a customer will buy again. Omnisend found that 39% of online shoppers had abandoned a purchase because of a frustrating chatbot. And trust, once lost, is slow to come back: Gartner reports that only about a quarter of customers would try a chatbot again after a bad experience.
The math that built the maze
So why do companies keep building these labyrinths? Because on a spreadsheet, they look brilliant. The Berkeley researchers put the saving at roughly 70 cents per interaction compared with a human agent. Multiply that by a few million contacts a year, and you’ve got a figure that makes any finance department sit up straight.
The trouble is everything the figure leaves out. Many systems are graded on “containment” or “deflection,” the share of customers kept away from a paid human. By that yardstick, a customer who hangs up in disgust counts as a win. Even some vendors now say so out loud; one support-software company argues that deflection-based pricing pays the bot for brushing off a customer whose problem is still unsolved.
Klarna supplies the cautionary tale the whole industry now quotes. In 2024 the fintech boasted that its AI assistant was doing the work of 700 agents, and it estimated a $40 million profit improvement. By 2025 its chief executive conceded that cost had weighed too heavily in the company’s decisions, that quality had suffered and that people often still want to talk to people. Klarna started hiring human agents again.
What the good ones do differently
None of this means the bots should go. Customers like getting answers at midnight, and nobody misses waiting on hold to check an order status. The companies getting it right have simply stopped treating automation as a wall and started treating it as a front door. A handful of practices turn up again and again.
Make the exit obvious. Here’s the lovely irony: making it easy to reach a person makes people more willing to use the bot. Five9 found that 55% of consumers trust AI to resolve their issue when a human option is available, compared with just 26% when there isn’t one. Gartner’s advice runs the same way: tell customers at the start that the bot will connect them to an agent if it can’t solve the problem.
Hand over the whole story. Nothing sours a customer faster than explaining a problem to a bot and then explaining it all over again to a person. Gartner recommends that the chat turn seamlessly into an agent conversation that picks up exactly where the bot left off. Customers expect the representative to have read the transcript, and they’re right to.
Know when to quit. A well-built bot recognizes failure early. A second misunderstanding, a repeated request or a sharp change in tone should trigger a handoff, not another cheerful apology.
Keep the bot where it shines. Order tracking, password resets, store hours and appointment changes are perfect work for automation. Billing disputes, cancellations and anything emotionally charged usually aren’t. The California Management Review authors recommend setting clear expectations about what the chatbot can and can’t do.
Measure what the customer feels. Track failure rates and escalations, not just containment. As one customer-experience expert told Customer Experience Dive, “Don’t celebrate containment if customers feel trapped.”
Test with real people. Real customers mumble, cram three problems into one sentence and call from noisy airports with kids screaming in the background. A system tested only on tidy, polite scripts will stumble the moment it meets the public.
Where the profit really lives
Built this way, the bot still earns its keep, and then some. It absorbs the routine questions that clog the phone lines, frees human agents for problems that need judgment and answers at two in the morning when nobody’s on shift. The savings are real; they just aren’t the whole ledger. A frustrated customer who calls back angry costs more to serve, and one who quietly takes their business elsewhere costs a great deal more than 70 cents.
The companies winning this round aren’t the ones with the cleverest bots. They’re the ones that remember what the bot is for. It isn’t there to keep customers away from the company. It’s there to get them what they came for as quickly as possible, by whatever path works, and that includes a person when a person is what’s needed.
A note for smaller businesses
There’s good news in all this for smaller operations. A big company’s phone tree has usually grown by accretion over decades, one menu option at a time. A small business gets to start fresh: a simple chatbot that answers the ten questions customers actually ask, says plainly that it’s a bot and offers a phone number or a callback the moment it’s out of its depth. That isn’t a compromise. It’s precisely the design the research favors.
Customers aren’t against automation. They’re against being trapped. Build the door before you build the maze, and they’ll happily walk through it.
Sources
- Gartner, “Gartner Survey Finds 64% of Customers Would Prefer That Companies Didn’t Use AI for Customer Service,” July 9, 2024. gartner.com/en/newsroom
- Customer Experience Dive, “3 of the spookiest customer experiences” (Gartner State of the Customer survey on doom loops). customerexperiencedive.com
- Cybernews, “Majority of Americans shout and swear at AI-powered customer service agents” (Parloa survey). cybernews.com
- Customer Experience Dive, “Why customers are fed up with AI chatbots” (Five9 survey). customerexperiencedive.com
- Yuqing Ren and Rongjin Zhang, “Chatbot Frustration is Real: Hidden Costs and Best Practices,” California Management Review, April 2026. cmr.berkeley.edu
- Omnisend, “39% of shoppers have abandoned their purchase due to frustrating AI chatbots,” March 12, 2025. omnisend.com
- Renascence, “Gartner: Only 27% Retry Chatbots After a Bad Experience,” September 3, 2026. renascence.io
- Gladly, “Why AI vendors are designed to frustrate your customers,” July 24, 2026. gladly.ai
- eMarketer, “Klarna backtracks AI customer service plans”; Maginative, “Klarna Dials Back its AI Customer Service Strategy”; Finextra, “Klarna rethinks AI customer service push,” 2025.