“Please put me throughto a human.”

Everyone knows the scene. You call a company, and a friendly voice answers.

At some point you hang up – or yell “HUMAN!” into the phone, which actually works more often than not.

So many people have been through this that, for most of them, “AI on the phone” has become shorthand for “an obstacle in front of the actual conversation.” A business owner who introduces a voice assistant today isn't starting from zero. They're starting in the red. Their customers expect the worst.

And honestly: that reflex isn't irrational. It's the result of twenty years of bad phone robots that weren't built to help customers, but to keep calls away from expensive staff. Customers noticed. They weren't being served, they were being brushed off – just politely.

So the interesting question isn't whether customers like AI on the phone. It's this: What separates the assistants customers stay relaxed with from the ones they hang up on?

Look at when callers bail out of automated systems, and it's almost always the same four moments.

All four reasons have one thing in common: it's never about the voice. It's about whether the system was built for the customer –

The good news: all four mistakes can be avoided. Not with more technology, but with a different mindset when you build. Five rules – each with the line the customer hears:

In every row, the good version is the one where the customer has less work and more certainty. That's the whole trick. Not “sounding more natural,” not “doing more” – but letting the caller feel, with every sentence, that their request is actually moving forward.

No assistant can do everything. The question isn't whether it hands off to a person, but how. This is where it's decided whether the customer remembers the call as help or as a hurdle.

Everyone knows what a bad handoff looks like: you get transferred, it rings, someone picks up – and asks, “So what's this about?” Back to square one. The three minutes with the assistant were wasted, and the customer knows it.

A good handoff has three qualities:

1. It comes early enough. The assistant recognizes on its own when it's reaching its limits – a complaint, a request for a goodwill gesture, an emergency – and offers the handoff before the customer has to demand it.
2. It goes to the right person. Not to “the main line,” but to the colleague who's responsible and available right now. If no one's there, it offers a callback and schedules it – at a time the customer chooses.
3. It brings the context along. Before she picks up, the colleague can see who's calling, what the request is and what's already been settled.

That third quality is the most important and the rarest. It turns the assistant from a bouncer into a colleague who has set up the conversation.

Many companies hesitate to introduce their assistant as AI. The worry: customers will hang up the moment they hear it. Our experience shows the opposite. Customers hang up when they notice it after it was kept from them.

An assistant who introduces herself this way sets the right expectations. The customer knows: I'm talking to a system that can book appointments and answer questions, and that will pass me on to a person when it needs to. That isn't a drawback, it's a clear statement. And it takes away the tension that builds when you're not sure who you're actually talking to.

Then there's a second kind of honesty that counts even more day to day: admitting when you don't know something. “I can't see that from here, so I'll have someone call you back” is a good sentence. “Yes, we definitely have that in stock,” without a look at the system, is a bad one – even if it sounds friendlier.

And one more thing: the EU AI Act requires companies to tell people when they're interacting with an AI. Treat that as a box to check, and you hide the notice in the fine print. See it as a virtue, and you simply let Anni introduce herself – and earn trust before the conversation has really begun.