AI Support Is Broken and Nobody's Saying It
Your AI Support Bot Is Driving Customers Away
Hi Team,
HOW IS IT JULY ALREADY?
Before I jump in, two important things:
If you wish you traveled more and want to travel for free this summer, get yourself a Chase Sapphire card and check out seats.aero to start looking at transfer options for your summer vacay. Anyone can do it, GO SEE THE WORLD. I will keep saying this until I am blue in the face. It’s free money.
If you have been to Kotor or Dubrovnik lately and have suggestions on where to stay, please let me know. Debating an October sprint with the kiddos. :)
Now for this week, I want to dive into maybe the hottest and most unnerving topic discussed here, which will certainly get me in trouble.
AI in CX.
Something I’ve been thinking about lately:
I still have yet to meet a single customer who told me they loved an AI support experience.
Every brand I talk to is implementing AI all over their CX stack. Your efficiency metrics look incredible on paper. Response times are down, handle times are shrinking, and your team can process way more volume than they could a year ago.
But nobody is talking about the fact that your customers hate a lot of it, and according to data from this week’s sponsor, >50% of bot-resolved tickets get reopened.
Not all of it and maybe not always, but enough of it that a meaningful chunk of them are one bad bot loop away from abandoning your brand entirely.
They are not even waiting long enough for the CSAT survey to hit, they already slammed the phone down and are jogging to your competitor instead.
This week, I want to talk about where AI actually works in CX and where it actively destroys trust, brought to you by real data.
Let’s get into it.
This week’s newsie is brought to you by Kim.cc.
Nearly 60% of consumers welcome AI for simple tasks, but demand a human when the stakes are higher.
Everyone keeps saying customers want AI support.
Turns out that’s only half true.
Kim.cc surveyed 1,000+ U.S. consumers to understand what people actually think about AI customer service.
Some findings were baffling:
• 50% blame company leadership, not the AI, when a chatbot makes an error.
• Over one-third have already abandoned a brand because of a bad AI support experience.
• Nearly half repeatedly type “agent” just to escape the bot.
The takeaway isn’t “don’t use AI.”
It’s that customers want AI for simple questions, and humans the second things get complicated.
Kim.cc is built around that exact idea: AI customer support with a human eye.
Through their Sentinel service, AI handles the work, while trained experts supervise the system - reviewing quality, handling exceptions, and continuously training the AI. They work with 200+ Shopify brands to help them move faster without turning support into a locked room with a chatbot in it.
If you’re building customer support for your Shopify store, this report is worth reading.
I Watched Brands Get This Backwards
I’ve started using two questions before letting AI touch anything, because I’ve seen enough brands roll out AI support, celebrate the efficiency gains, and then wake up a few weeks later to reviews about how impossible it is to get help.
The two questions:
Can the AI execute this task flawlessly?
Are the stakes low enough that if it goes slightly wrong, the customer doesn’t lose money, miss something important, or feel ignored during a crisis?
If the answer to both is yes, AI is probably fine. If the answer to either is no, you need a human close by.
Simple tracking requests, basic FAQs, store hours, return policy lookups. These are low-stakes, high-repetition tasks where speed actually matters and accuracy is mostly binary. Nobody needs to wait 18 hours for a human to say their package is still in New Jersey.
That’s where AI works.
According to Kim.cc’s survey, 44.5% of consumers welcome AI for simple tracking, and another 42% accept or tolerate it. For basic info like store hours, return policies, or FAQs, 42.9% welcome AI and 43.2% accept or tolerate it. Pretty clear green light.
But the second the issue gets expensive, emotional, or messy, people want a human. Only 15.9% welcome AI for financial transactions like refunds, billing disputes, or plan upgrades. 51.7% demand a human. For high-emotion issues like a terrible experience or flight cancellation, 59.5% demand a human.
Essentially, the lower the stakes, the higher the tolerance. The higher the stakes, the faster people bail. I know this might sound obvious to most of you all, but then half the brands I talk to are “99% AI,” so apparently we still need to say it.
This is where brands keep getting themselves into trouble. They see AI work for “Where is my order?” and somehow decide it should sit in front of every customer issue.
That is like saying because a microwave is good enough for leftovers, it should handle your wedding catering. Useful tool, wrong damn job.
The Stakes Are The Strategy
A tracking question and a refund dispute both count as one ticket. A return policy lookup and a damaged item escalation both count as one ticket. A subscription skip and an angry cancellation after three bad deliveries both count as one ticket.
But they are not the same thing. Some tickets are low-stakes and annoying. Others are trust moments.
If I call Delta to change my seat and an AI bot can move me from a middle seat to an aisle seat in 30 seconds, incredible. Give me the bot 10/10 times. I do not need a human, I do not need fake empathy, and I sure as hell do not need someone to say, “I understand how frustrating middle seats can be.”
Just move me away from 26B and let us both continue our lives.
But if I have an emergency, my flight leaves in 30 minutes, and I need to change the name on a ticket or get rebooked before I miss something important, please do not make me explain my life to a bot that can only offer three buttons and a cheerful apology.
Same airline, same general support category, completely different stakes.
That’s the part most AI CX rollouts miss. The better question: what happens if AI gets it wrong?
If your AI screws up a tracking update, the customer is mildly annoyed. If your AI screws up a billing dispute or gives wrong information about a return policy, the customer loses money or time, and now they’re gone.
The survey found that when AI gives bad information that costs customers time or money, 50.3% blame company leadership for deploying an unready tool. Only 17.7% blame the AI itself.
Customers see bad AI support as a deliberate cost-cutting decision. Once they see it that way, the damage compounds. They don’t think the bot is broken. They think the brand is cheap.
The Gatekeeper Problem
The Kim.cc survey calls one of the worst AI behaviors “The Gatekeeper”: refusing to give customers an option or button to speak to a human.
It is brutal because it is exactly what customers hate, and exactly what many brands are optimizing for. One of my least favorite metrics in CX is “deflection rate.”
Imagine telling a customer the goal of your support bot is to deflect as many people with issues as possible from getting to their ideal resolution in a painless manner.
Sounds less sexy on the vendor slide.
Most AI is designed to solve the issue before escalating. On a slide, that sounds reasonable. Why route to a human if the bot can handle it?
That assumes the bot actually can handle it. More importantly, it assumes the customer trusts the bot to handle it.
Those are different things.
When the bot misses once, maybe the customer rephrases. When it misses twice, the customer starts getting annoyed. By the third attempt, the customer has usually decided two things: the bot is useless, and the brand is making it hard to get help on purpose.
According to the survey, 47.6% of consumers have typed “agent,” “human,” or “representative” repeatedly to bypass the bot.
That is a damn revolt.
A customer should not have to trick your support system to reach someone who can make a decision. They should not need to type nonsense, threaten to cancel, pretend they want to buy something expensive, or call a separate phone number because the chat experience turned into a hostage situation.
It is an obstacle course with a CSAT survey at the end.
If your AI cannot solve the issue in two prompts, it should offer a human. Not after five attempts. Not after the customer starts typing in all caps. Not after they threaten to report you to the Better Business Bureau like it is 2006.
Two prompts.
Fake Empathy Makes It Worse
Customers hate fake empathy.
When a bot says, “I’m so sorry to hear that, I understand how frustrating that must be,” 32.3% of consumers say they find it manipulative or annoying because a machine cannot actually feel empathy. Another 17% say it actively angers them because it feels like a fake delay tactic. Only 18% find it comforting.
So roughly half of customers are actively put off by the exact language many brands are adding to make AI feel warmer.
AI cannot feel empathy, and customers know this. Pretending otherwise does not make the interaction better. It makes it feel dishonest.
If the bot is going to apologize, it should apologize for the delay or inconvenience and then immediately solve the problem.
“I’m sorry this is taking longer than expected. Let me pull up your order details now.”
That works because it is transactional and honest.
“I totally understand how frustrating this must be for you” does not work because the customer knows the bot does not understand anything. It is reading a script.
The brands that do this well do not try to make AI sound human. They make it sound fast and accurate.
The bot is a tool, not a therapist. Let it be a good tool.
What Actually Works
If the task is simple, repetitive, and low-risk, AI is fine. Order tracking, FAQ lookups, store hours, basic account changes, return policy questions, address changes before fulfillment. Let the bot handle it. Speed actually wins there.
When the task involves money, emotion, urgency, or any level of interpretation, route to a human quickly. Don’t make the customer fight for it. Don’t make them type “agent” five times. Build the escape hatch into the first or second prompt, and make it obvious.
Stop trying to make your AI sound empathetic. It’s a bot. Let it be fast, accurate, and honest about what it is.
Solve the problem or get out of the way. That’s the whole job.
Right now, too many brands are using AI like a blanket solution, and the data is screaming that it’s backfiring.
Your efficiency metrics look great. Your customer sentiment does not.
Fix the framework before you lose more people.
Again, highly suggest checking out the report from Kim.cc
That’s it for this week!
Any topics you’d like to see me cover in the future?
Just shoot me a DM or an email!
Cheers,
Eli 💛
