Let’s talk about the last time you had to fight a machine to reach a person.
Picture this. It’s a Tuesday. You’re buying something ordinary — groceries, fuel, a ticket. The card declines. You try again. Declines. You check the app. “Suspicious activity detected.” So you call the number on the back of the card.
A voice answers. Not a person. A machine that says it’s here to help. It asks you to say the reason for your call in a few words. You say “fraud.” It routes you. It asks for your ID number, your account number, your date of birth, the last four digits of something. Then it tells you the block was correct and to try the transaction again.
You try the transaction again. It declines again.
And now you’re in the loop that defines modern customer service: the machine that blocked you is the same machine judging your appeal. You press zero. You say “agent” until you feel ridiculous. You wait. When a human finally answers, they read a screen — and the screen says the same thing the machine said.
The human isn’t an appeal. The human is a speaker for the machine.
That’s the pattern nobody wants to name: the person is no longer the default. The person is the escalation. And escalation is priced.
Here’s what I mean. Every call centre, every bank, every telecom, every government line is being rebuilt around one assumption — the machine handles the volume, and the human is the exception you have to earn. Some companies don’t even pretend anymore. “Talk to a human” is literally sold as a premium perk. You pay more, you skip the queue, you get a person. The rest of us get the maze.
Now, I’m not anti-automation. I build software. I know what a good bot can do for a business. But there’s a difference between a machine that handles the routine and a machine that decides who gets heard. That difference is where dignity goes to die.
Because the machine gets to be wrong in a way only a person can fix. It blocks your account and there’s no one to argue with. It flags you as a risk and the flag follows you across companies, across years, across systems that talk to each other in the background. It cancels a payment that was supposed to keep your lights on, and its explanation is a screen you can’t reach.
Think about what that means for the moments that actually matter. The death in the family. The medical bill. The account that needs to be unlocked today, not in five business days. These aren’t routine. These are the moments where being heard by a person isn’t a convenience — it’s the entire point. And in those moments, the machine can’t hear you, because the machine was never listening. It was classifying.
The machine doesn’t need to understand your situation. It needs to put you in a box, and your job is to fit the box. If you don’t fit, you don’t get heard. You get more questions. You get “try again later.” You get the loop.
So here’s the principle: a system you can’t appeal to a person is not a service. It’s a toll booth. And the toll is your time, your patience, and eventually your right to be treated as more than a case number.
So what do we do? If you’re building — and I say this to every founder who asks me about AI customer service — don’t ask how much you can automate. Ask where the human stands, and whether that human can actually change the outcome. An override that can’t override is theatre. A human who has to read the machine’s verdict back to you isn’t a human in the loop. They’re a human in the way. The companies that win the next ten years aren’t the ones with the best bots. They’re the ones where, when the bot is wrong about a real person, a real person can fix it. That’s not a cost centre. That’s the trust architecture.
And if you’re just living — remember this the next time you’re on hold: the difficulty of reaching a person is a design choice. It is not an accident. Somewhere, a spreadsheet decided your time was worth less than the cost of employing someone to hear you. Call it efficiency if you want. It’s a statement about who counts.
The machines will get better at sounding human. They’ll get better at pretending to listen. But the test was never how well the machine talks. The test is whether, when the machine is wrong about you, there is a person you can reach who can change the outcome.
If that person exists, you still live in a world built for humans. If they don’t, you’re not a customer. You’re a case number, waiting on hold, hoping the machine makes an exception.
It won’t. The machine doesn’t do exceptions. It does categories.