There is a familiar headline making the rounds again: customers are turning against AI in customer service. The surveys look damning. AnswerConnect’s tracking research across the US, UK, and Canada found preference for a real person climbing while preference for AI fell to a sliver, with frustration rising alongside it. Read fast, and you would conclude the market has rendered its verdict on AI.
I read it differently, and the data agrees with me. After years running support organizations at scale, most recently leading Device, Digital, and Alexa support for more than 200 million subscribers, and earlier in regulated financial services at Visa, Capital One, and USAA, I have watched customers react to automation in real time. They do not hate AI. They hate not getting their problem solved. They hate repeating themselves. They hate hitting a wall with no door. When automation causes those things, customers revolt.
The backlash is a verdict on service, not software
Look past the headline number and the picture sharpens. The same research that shows people prefer humans also shows that roughly four in five find AI genuinely helpful for simple, repetitive questions, and that most will start with a chatbot if they can reach a person when they need one. That is not a technology people reject. It is a tool people accept when it works and resent when it traps them.
What drives the frustration is friction. Glance’s 2026 CX Trends research found customers reporting more loops and more time re-explaining themselves as companies rushed to automate. Sixty-eight percent said a complete resolution mattered most, ahead of speed, and only seven percent rarely had to repeat themselves across channels. None of that is about artificial intelligence. A human associate who did the same would earn the same contempt. The technology is new. The grievance is old.
What I learned optimizing the wrong number
The most expensive mistake I have seen leaders make, and one I had to correct in my own organization, is optimizing automation for deflection instead of resolution.
Deflection rate, sometimes dressed up as containment rate, measures how many contacts the bot handled without passing them to a person. It looks like efficiency and moves cost off the financial statements. It is also easy to game. A bot can contain a contact by being too hard to escape, or by closing a session it never resolved. The dashboard turns green. Meanwhile that customer calls back twice and quietly takes their business to a competitor.
When teams stop rewarding containment and start rewarding resolution and low customer effort, two things happen. The automation gets better, and the associates get more effective, because the contacts that reach them are the complex, human ones where judgment and empathy create value and customer delight. When a customer faces a moment that is confusing, emotional, or high stakes, put a person on it.
The handoff is where most of the damage happens. Customers do not object to starting with automation. They object to reaching a live associate and being told to start over, as if the previous ten minutes never occurred. Fixing that one seam, so the associate opens already knowing what the customer tried and why it failed, lifted our scores more than any single piece of automation.
Financial services raised the stakes
Earlier in my career, in payments and banking, the tolerance for a loop was effectively zero. When a card is declined at checkout, an account is frozen on a fraud flag, or a dispute sits unresolved, the emotional temperature is high and the trust is fragile. Those environments taught me what AI is for.
The customers I served then never complained that an algorithm had helped them. They complained when the system could not tell them why their card was declined, or when authentication kept them spinning. The AI they welcomed was the AI they could not see: routing to the right associate, fraud models that caught problems early, account history surfaced so no one had to ask twice. The AI they resented stood between them and a resolution.
That distinction travels across every industry I have worked in: AI that informs and accelerates a person earns trust, and AI that blocks a person destroys it. Well over half of consumers say their trust would fall if a business relied mainly on AI.
Three principles that earn trust
Briefing a leadership team today, I would keep it to three things.
1. Measure resolution, not deflection. If your automation cannot point to problems solved, you are buying a quieter queue, not a better experience.
2. Keep the door open and obvious to a person. The companies generating backlash are the ones hiding the exit. Customers will use automation first when they trust they can reach a human the moment they need one.
3. Never make the customer carry the context. Pass it. The handoff from machine to associate, and from one channel to the next, is where loyalty is won or lost. I wrote about how trust is built and broken at these seams in The Trust Algorithm, and the seams are almost always operational, not technological.
The verdict customers have always delivered
The cost of getting this wrong has not changed even as the tools have. Qualtrics and ServiceNow put $1.9 trillion in US consumer spending at risk from poor experiences, with roughly four in five customers reporting they switched brands after a bad one. Customers are unforgiving of bad service and generous toward good service, regardless of who or what delivers it.
The so-called AI backlash is not a referendum on AI. It is the verdict customers have handed down for decades. The winners over the next several years will not be the companies that use the least automation, or the most. They will be the ones that use it to make service feel effortless and human, in that order. Customers were never going to love our technology. They were going to judge us on whether we solved their problem and treated them like a person. That has not changed, and it never will.