Behind the Scenes
·
10 min read
·
Generative AI Won't Grow Your Shop. The Boring Automation Found $289,000.
By
Noah Satho
·
Milpitas, California
The Advice Everyone Gives Is Pointed at the Wrong Thing
When most people hear generative AI they picture the machine that makes the stuff, the images, the video, the captions, the posts. That's a fair read, because that's what's been filling everyone's feed for two years now. It's also what gets sold hardest to small business owners, and it's what doesn't work. The research on this has become hard to argue with. Gartner surveyed 1,539 US consumers and half of them said they'd rather give their business to a brand that doesn't use generative AI in the things they see. The Harris Poll found that 54% of people are actively annoyed by how much they hear about AI, and around two thirds said they'd love it if brands stopped mentioning it altogether. My favorite finding is from the Journal of Consumer Research, because it explains the mechanism instead of just measuring the damage. Researchers at USC ran eight experiments and analyzed over a million TikTok posts, and content labeled as AI made got roughly 7 to 8% fewer likes and about 7% less engagement overall. It's a modest effect, and I'd rather call it modest than inflate it to win an argument. The interesting part is why it happens, because people rated those creators as having put in about 15% less effort, and the penalty disappeared completely when the AI tool involved took real skill to use. So it was never really about the technology. People punish work that looks like it was cheap to make, and generative content aimed at an audience looks exactly that cheap, because it was. You're spending your credibility to save an hour. That same technology, running quietly in your back office, has no audience at all. Nobody's judging the effort that went into a script that moves a quote from one system into another at three in the morning. It only has to work.
What Actually Worked Was Boring
Our flagship example is quoting, and it's the least exciting thing imaginable. We do paint protection, and the price depends on the size of the car. A lead arrives in one system, our pricing lives in another, and the two don't talk to each other, so for years somebody had to read what car the person drove, work out its size class, look up what that costs, build the proposal and send it, about five minutes each. My sales rep works around 50 people on a busy day, so that's 250 minutes, more than half his shift spent moving details from one screen to another instead of talking to anybody. Before we automated it, 23.4% of the people who asked us for a price ever got one. The rest just fell through, because there weren't enough hours in the day. In the last full week I audited, 91 of 113 new leads got a quote within seconds of arriving, with nobody touching it at all. The hard part wasn't the AI, and this is the thing I most want other owners to hear. Most of the work went into a list of over 1,300 vehicles we'd been keeping by hand for years, because size isn't something you can calculate. A Lamborghini Urus is an SUV by any measurement and prices as an exotic. A 1966 Mustang and a 1972 Suburban price identically once age overrides shape, and those two cars have nothing else in common at all. Ask a general purpose chatbot to classify either of those and it'll answer confidently and get it wrong. The model was the cheap part, and what we already knew was the asset. The part I'm proudest of is the refusal. If the script isn't confident about a car it doesn't guess, it holds the quote back and hands it to a person to sort out. We learned why the hard way, when a Porsche came through, the matcher fell back to guessing off the make, and we underquoted the car. Once a customer is holding a number, that number is real, and you don't get to quietly take it back. The rest of the list is just as dull. A job that wakes up every five minutes and works out who's owed a follow up. A morning call list that gets generated instead of assembled. Waivers and walk around forms built from templates instead of retyped. Footage from a shoot that copies itself off the card, checks itself, and tags itself by vehicle. A daily pass that corrects stale prices before anyone sees them. Ten of these run without anybody watching, and on the last check 167 out of 167 customers who'd already bought were correctly taken off follow up, which is exactly the sort of thing that costs you a relationship when it goes wrong. The one I didn't see coming: the bridge I built between those two systems, purely so I'd stop retyping things, turned out to be something other shops want to buy. The boring plumbing became a product. I should be straight about what this doesn't do. The quoting system has sent hundreds of proposals and it closes about 1 in 50 entirely on its own. That sounds like a failure rate and it's the honest core of this piece. It never replaced my sales rep and it was never going to, because nobody hands over four figures on a car they love without speaking to a person first. It also doesn't mean the $289,000 is in the bank. That's a list of names and amounts somebody now has to work through, and that somebody is a human being.
The Best Thing I Built Just Asks Me Questions
If you take one idea away from this, take this one. Most of what a small business knows is locked in the owner's head, and it never comes out, because writing it down is miserable and there's always something more urgent. So I built a tool that interviews me instead. It asks one question at a time, it saves after every answer so I can quit halfway through and pick it up later, and what comes out the other end is a written process my team can actually follow. It sounds like this: "When a lead says they want the cheapest option, what do you actually say back? Not the polished version, what you'd say standing in the shop." Then I answer, and it asks me the follow up I'd have skipped if I'd been writing it on my own, which is the whole trick. What I'd been avoiding was never the knowing, it was the typing. The same idea at a bigger scale is where our single biggest win came from. We had years of customer conversations sitting in a system nobody could read at volume, so we ran an analysis across 301,861 messages sorted into deals we won and deals we lost, another 55,081 emails, and 170,644 outbound messages from ten different people. That's the sort of project you'd hire an analyst for and then wait six months. It took about 90 minutes. None of that generated a single piece of content. It generated answers about our own business that we'd been guessing at for years. And here's the part that actually matters to me, which is why I'm writing any of this down. I don't want AI to replace human connection, I want it to free us up from mind numbing tedium and allow people more time to connect with one another. I can afford to speak with my clients at pickup for 30 minutes, because I can get so much more done with agentic AI and vibe coded automations handling admin work for me, instead of trying to cut our conversation short because I have too much work piling up on my plate. That's the whole return, which was never about throughput or payroll. My sales rep spends his day in conversations because the typing left his desk, and I get to stand in the lot with somebody and talk about their car for half an hour without a pile of admin building up behind me. Automate the queue, not the relationship.
So you're saying don't use AI at all?
Not even close. I use these tools every day and the scripts described here were written by one of them. The distinction I'm drawing is about direction, not technology. Aimed at the things your customers see, generative AI costs you credibility to save you an hour. Aimed at the admin nobody wanted to do, it's the best money we've spent.
What should a shop owner automate first?
Whatever you personally retype the most. Not the thing that sounds most impressive, the thing that's most repetitive. For us it was moving a customer's details from the system where leads arrive into the system where prices live, which is about as unglamorous as it gets, and it turned out to be the biggest single win we've had.
Do I need to know how to code to do this?
I'm not a professional developer. I describe what I want in plain English, an AI model writes the code, and then I test it against real cases until it stops getting them wrong. The genuinely hard part is explaining exactly what your business already knows, which is knowledge you have and a contractor doesn't.
Is the $289,000 money you earned?
No, and it matters that I'm clear about this. It was already ours. It's quotes people never came back on, deposits that had gone past due, and pipeline our CRM was showing as zero. It isn't new sales and it isn't collected cash, it's a list of names and amounts that a person can now work through, and that list didn't exist before.
Does an AI talk to your customers?
Only the first quote is automatic. Everything after that is one of us. There's no model writing replies, negotiating, or sitting in the middle of a conversation, and there isn't going to be, because that's the part of this business worth protecting.
What does vibe coded actually mean?
It's the informal name for describing what you want in ordinary language and letting an AI model write the code, rather than writing it yourself line by line. It's how nearly everything here got built. The catch people miss is that it doesn't remove the need to know your own business, it just removes the need to type.
Keep reading
More from the shop
Owner check-ins, install standards, and straight answers about what a coating does.
- The Honda Fit we were not even supposed to restore
- What a proper Icon Rocklear installation should look like
- Icon Rocklear quality control: what happens after installation
- A Nissan R35 GTR, start to finish
- Deep scratch repair: what a coating can and cannot fix
- A track driven ZL1 Camaro gets film repair and a permanent finish
- Behind the scenes on a 2021 Toyota Tacoma
- Saving carbon fiber from sun damage on a 2022 BMW 330i
- Icon Rocklear Reddit reviews: what Reddit gets right, and what it misses
- Is Icon Rocklear worth it? An honest answer from the shop that installs it
- What happens to your deposit if you change your mind
- Our quotes take five seconds now. The hard part was not the AI