Inside the $1,800 Missed Call Agent I Built for a Truck Repair Shop

I sold a voice agent to a real truck repair shop for $1,800. Its job is simple to describe: pick up the calls the team can't get to, answer the basic questions, pull out the details that matter, and email them to the front desk. Here's what went into it, and what I'd want any shop owner to know before buying one.

The job: catch the calls nobody can pick up

The agent is the receptionist for Big Sky Truck and Trailer Repair whenever the team is busy or the shop is closed. That's it. It's not trying to replace the front desk. It covers the calls that would otherwise ring out, including the middle of the night when nobody is there to man the phones.

Every caller goes down one of three paths. A new service request. A status check on a truck that's already in the shop. Or a general question that needs a service advisor to follow up. Each path gets its own instructions, because a driver asking "is my truck done yet" needs a totally different conversation than someone booking a brake job.

On a new service call, the agent gets the caller's name, the make and model, what the truck needs, the best callback number, whether it's drivable or needs a tow, and when they want to drop it off. Then it gives the shop hours and wraps up.

What the shop gets after every call

This is the part owners care about most. When the call ends, the important details get pulled out of the transcript automatically and sent two places. A new row lands in a Google Sheet the managing director can scan. And a clean email hits the front desk: you missed a call from a potential customer, here are the details. Name, number, call type, vehicle, drivable or not, drop off time, a summary, and the recording.

Nobody has to listen to a whole call to figure out who called and when they want to come in. That would be super inefficient.

One small detail I'm proud of. The agent asks if the number they're calling from is the best one to reach them. If the caller says "actually, my wife's number is better," that's the number that shows up in the email. Otherwise it uses the number they called from. Small thing, but it's the difference between a lead you can call back and one you can't.

Built for truckers, not a demo room

A lot of the decisions came from who's calling. Truck drivers call from the road with a ton of background noise. So I turned noise filtering up to aggressive, and I gave the agent permission to say "sorry, I can't hear you, can you speak up?" instead of guessing. Without that, it'll try to answer anyway and probably get it wrong.

Speed mattered too. I chose a slightly less smart model because it's faster. The smarter one pushed response time to almost 1,300 milliseconds versus about 1,000 for the one I picked, and a normal phone call with a friend runs around 500 to 600. When you're talking to truckers, I'll trade a little intelligence for a lot less waiting.

I also kept it lean with no knowledge base. In my experience they add around 50 milliseconds, and the agent gets worse at finding the right answer than if the info just lives in its main instructions.

Guardrails that protect the shop

The agent only answers what it's been told. If the info isn't in its instructions, it says it doesn't have it and writes the question down for a service advisor. That one rule is the biggest thing I've found for keeping an agent from making stuff up.

Pricing gets the same treatment. Ask "how much for a brake job?" and it explains the shop wants to look at the truck first to give an accurate estimate, then asks if you can bring it by. Same answer every time.

It tells the truth if someone asks whether they're talking to an AI. And if a salesperson calls and keeps pitching, two or more times, it politely ends the call instead of burning minutes and wasting the owner's time.

Testing was 40% of the build

About 40% of the two week build went into testing. I ran simulated callers against the agent, like "Bob" calling to check on his truck, and graded each run on whether it followed its steps. Then I tested the whole pipeline from call to inbox, because a great conversation means nothing if the email never reaches the front desk.

That's what I'd push any business owner to ask about. Not just "does it sound human," but "how did you test it, and what happens after the call ends?" And none of this is specific to trucks. Catching missed calls, a few clear paths, and a clean handoff to your team works for pretty much any business that lives on the phone.