I needed a garage door fixed last week. I called the company everyone in my area knows. They advertise on the radio constantly and their reviews are excellent. If you were picking a garage door company on name recognition alone, they would win.

Twenty minutes later, I hired someone else.

What happened on the call

An AI voice agent answered. For the first minute I was impressed. It sounded natural and asked reasonable questions, and I remember thinking it might be a real differentiator for them.

Then it started looping.

It asked what was wrong with the door. I told it. It asked again, phrased a little differently. I told it again. It could not give me a price range, could not tell me when a technician was available, and could not schedule anything. Every path through the conversation ended at the same questions I had already answered.

Eventually it offered to transfer me to a live person. I sat on hold for more than five minutes and hung up.

I called a smaller competitor. A woman picked up, listened to my problem, assigned a technician, and gave me a quote in a few minutes. I spent twenty minutes with the first company's clone and a few minutes with a person, and the person got the job.

The spreadsheet is measuring the wrong thing

I understand why the first company did it. Answering phones costs money. An AI agent picks up every call, never takes a break, and never calls in sick. On paper it looks like an easy win.

In garage door repair, though, the phone call is the sale. The person calling has a broken door, a car stuck behind it, and a decision to make in the next ten minutes. They are not calling to be qualified. They want a price and a time.

If your AI cannot quote, cannot schedule, and cannot hand off to a human quickly, you have not cut your intake costs. You have built a leak between your advertising and your revenue. Every dollar saved on a receptionist gets spent again on radio spots that drive calls into a system that loses them.

I was their ideal customer. I called them first, already sold by their marketing. The phone tree gave that away.

The other company used AI too

Shortly after I booked, the company I hired sent me a text with a photo of the technician who was on his way and a short note about him: he served in the Air Force and speaks Mandarin.

I would bet that message was generated by AI. It pulled a profile, wrote a friendly sentence, and sent it at the right moment. I liked the company more for it, and I felt better about who was about to show up at my house. It cost them almost nothing.

Both companies used AI. One used it to replace the moment where trust gets built. The other used it to add to that moment after a human had already handled the hard part.

Where AI belongs in your customer journey

Before you put AI in front of a customer, ask three questions.

Is this the moment the customer decides? If so, put a person there, or make sure the AI can actually complete the transaction: quote, schedule, confirm. Anything short of that is a wall.

Can the AI finish the job, or only start it? An agent that gathers information and then drops the caller into a hold queue is worse than no agent at all. The caller has invested time and gotten nothing back.

Does it add something a human would not have time to do? Personalized confirmations, technician introductions, follow-ups after the visit, review requests written in plain language. This is where AI earns its keep, because it improves the human experience instead of standing in for it.

What to take from this

In home services, and in most local businesses, a live person who can quote and book in a few minutes will beat an AI agent that asks a lot of questions and delivers nothing. The cost savings are real, but they are small next to the customers who hang up and call the next name on the list.

Use AI to make your customers feel known. Do not use it to keep them from reaching you.

The company on the radio spent real money getting me to dial. Then it spent twenty of my minutes talking me out of it. That is fixable, and it starts with deciding which moments are too important to automate.