A colleague watched me work with an AI assistant recently. After a few minutes, they stopped me. "It's alarming how much you talk to that thing like it's a person." They weren't wrong. I speak to AI tools conversationally, with context and follow-ups and the occasional thank you. It's not because I think they're people. It's because that's how the tools work best. But the discomfort on their face was real, and it's a reaction I've seen more than once.
That reaction matters. It tells you something about where most people are with AI right now: somewhere between curious and unsettled. And if that's where your customers are, the way they first encounter AI in your business will either move them toward trust or confirm every doubt they already have.
The Call
I called the Snohomish County non-emergency line. I've called that number plenty of times for work. I know what to expect: a harried, overworked 911 operator who happens to not be on an emergency call at that moment. You give them the basics, they file the report, you move on.
An AI answered. Her name is Ava. My immediate reaction was anger. Not mild annoyance. Genuine frustration. I'm calling the police, and I get a machine? I was already mentally preparing my standard "agent, agent, agent" mantra to bypass it.
Then she asked me to explain why I was calling. Not from a menu. Not "press 1 for theft, press 2 for noise complaint." Just: tell me what's going on.
So I did. And the response was natural. It came back at the speed of the conversation, no awkward pauses while it loaded a script. She didn't try to guide me through a decision tree. She listened to what I said, understood the situation, and asked relevant follow-up questions. She filed the report, gave me a case number, and told me a deputy would follow up if there were questions.
I've called since. When Ava determines that a non-emergency call is actually urgent enough, she transfers you immediately to a human dispatcher with full context. She knows the difference. Since launching in 2024, that system has handled over 220,000 non-emergency calls and reduced the load on dispatchers by 90% during peak hours. The overworked operators I used to reach are still there. They're just spending their time on the calls that actually need them.
Why One Worked
My colleague's discomfort and my initial anger came from the same place: we weren't ready to interact with AI in a context where we expected a human. That instinct is valid. It's also the exact instinct that a poorly deployed AI system will trigger in your customers.
The sheriff's system won me over because it was designed around what the caller actually needs, not around what the technology can do. It didn't announce itself as AI and then fumble through a scripted experience. It just handled the interaction competently, at a human pace, with the right questions at the right time. And when the situation called for a real person, it got out of the way.
That's the difference between AI deployed well and AI deployed because someone had a budget for it. The technology was the same in both cases. The design thinking was not.
The Cost of Getting It Wrong
AI is in its early trust-building window right now. Most of your customers have interacted with bad chatbots, useless phone trees, and automated systems that made a simple task harder. They are not starting from a place of trust. They are starting from a place of "great, another machine that can't help me."
The first time a customer encounters AI in your business, they are forming an opinion that will be difficult to reverse. A system that fumbles the interaction, gives generic answers, or makes the customer feel like they've been downgraded from human service to a cost-cutting measure will do real damage. Not just to that interaction, but to the customer's willingness to engage with your business the next time.
A system designed around the customer's actual experience, one that handles the right tasks at the right pace and knows when to hand off to a person, builds the kind of trust that turns a skeptic into an advocate. That's the system worth building. The other kind isn't just a waste of money. It's a step backward.