AI in the Bay: How Shop Owners Are Building Custom Bots That Actually Work

Welcome to Ratchet and Wrench Radio,

produced by Endeavor Business Media,

a division of Endeavor B2B,

bringing you strategies and inspiration

for auto care success.

A.I. is everywhere.

But for many shop owners,

it still feels unclear how it fits

into day-to-day operations.

In this episode of Ratchet and Wrench Radio,

Seth Thorson, president and CEO

of EuroTech Auto Service in Minnesota,

cuts through the noise with a hands-on approach

to artificial intelligence in the shop.

After a sold-out session at Vision earlier this year,

Thorson explains how to create custom A.I. agents,

what he calls putting A.I. in a box

to handle specific, repeatable tasks like HR questions,

technician story rewrites, and maintenance recommendations.

Rather than replacing employees,

he positions A.I. as a time multiplier

that helps teams work smarter and faster.

From implementation lessons to common mistakes

and first steps, this conversation offers

a practical roadmap for shop owners

ready to experiment with A.I.

and actually make it work.

Let's listen in.

Hi, Seth, thank you so much for joining us today.

Thanks for having me here today.

Not a problem.

Seth Thorson, president and CEO of EuroTech Auto Service

in the Minnesota area, Minneapolis area, I should say.

Your session at Vision got the most response

that we had seen on the floor.

People were absolutely raving about it, it was sold out.

What do you think made this topic resonate

so strongly with shop owners?

I think A.I. is a hot topic right now

and I think there's a lot of misunderstanding

of what it can do, what it can't do,

how it will change things, how it won't change things.

I think that's a big part.

And I think being focused on how to make it work

in shop operations and use it day-to-day shop operations

versus I think a lot of the classes tend to focus on A.I.

and marketing and things like that,

where not a lot of them are actually diving in

of what it does for shop operations

and how to do different things inside your shop with it.

So I think that really resonated with people

more than some of the other A.I. classes

for the ones that attended it, right?

If you're struggling with marketing,

the marketing A.I. classes probably are gonna knock it

out of the park for you.

It just, I think this happened to resonate with people

that were seeing a lot of different topics

but weren't seeing a specific topic

focused on exactly what we were trying to teach

and accomplish.

What kinds of questions were people asking?

What did you notice?

Were there any trends?

Yeah, I think people want more of it.

They wanna learn more about the cloud

and the kind of the vibe coding stuff

where you can build some custom apps.

But really, it's a foundational class of start here.

And it's probably the third time I've done that class.

And the biggest thing is the first time I did it,

I did it for our Transformers group

that I'm part of with Greg Munch and those guys.

And so I did that as just a demo as a lecture

and everybody was amazed and like,

how do we do this?

And then they start working on it

but they don't get very far.

So the next time I did it for the BIMRS group

and Euro train for Sherry and Sherry said in the class

and said, well, you have to do this at Vision.

That one I did as a live workshop

where everybody had to bring a laptop,

a paid subscription and they left with 90% working product.

They still had some refinement to do when they got home.

There's still some homework to do, right?

In a four hour class, they're not gonna be done.

But that really resonated well

and the same thing at Vision, I think resonated well

because we really set it up as,

if you're taking this class,

you have to have a paid subscription,

you have to have your laptop,

you have to, you know, Sherry,

I need power at every computer so everybody has power.

We really spent a lot of intentional time

making this a workshop where I lead you through

actually creating something for you in your shop.

And I think that really resonated with people

because rather than just hear somebody speak,

they're building it with me live

and they're walking away with something

they can take back home.

So for someone who wasn't in the room that day,

what exactly is a custom chat GPT agents?

Sure, so what people generally do with AI

is they just go talk to it

and they try to train AI to think like them, do like them

and they try and do it on a variety of topics

and that gets really messy

because it doesn't always respond

and you get answers and things back

that you're not looking for.

What a GTP does is creates a very tight control

around a very specific topic.

And so when we teach it,

we teach how to do an HR bot, right?

That you give it your HR docs,

it can't go search the internet,

it can't do anything,

but answer based on the documents you give

and if it doesn't have the answer, it says, we don't know.

Otherwise AI tries to make something up

to make you happier,

goes find something random on the internet and says,

well, I found this and that gets really, really messy.

So really what we talk with the agents

is how to build something

or if you're building a service writer story

or the technician story rewriter,

how to rewrite a technician story,

you give it really clear defined boxes, right?

It's putting AI in a box

and you give it what we call the rubric

which is a really tight control.

And so we really teach and talk about how to box AI

in the do very specific tasks

and a lot of people get frustrated with prompting

where they prompt and it doesn't do it at once.

Well, if we create a bot or an agent

that all it's supposed to do is take this file

and do this with it, you don't have to prompt it.

You just dump the file in it and it automatically does it

and you don't have to worry about all these prompting

and things like that.

We do talk some general AI,

we talk some prompting,

how to do different acronyms I teach

on how to do prompting in the AI,

but that's kind of the thousand foot overview

of what that looks like

and what we kind of covered in class.

What's a simple example

of how one might actually work

or be applied in a repair shop?

You know, one of the ones that we talked a little bit

about was the technician story rewrite.

I think I've seen some chatter in their net

where techs are getting upset

that it's rewriting stories and not their story.

And one of the things we talk about

is we're gonna rewrite technician stories.

It has to maintain all their documentation

and it can't add things that weren't done.

And so one of the ways is we can tell AI

that anytime a technician type copies

and pastes a story in or writer copies

and pastes a story in, these are the parameters.

You cannot add tests that weren't done

because that's not what we wanna do on a document.

You can't add things that weren't done.

You can expound on technical abilities

and you really just type in what you're looking for it to do

in your shop and how you want it to run.

And that's another use of it.

Another use of it is we teach how to copy and paste

your Carfax report and spit out maintenance

based on different maintenance guides

and it'll spit out how do you want your maintenance

to look, do you want it to be red, green, yellow

when it's due and you can copy that in.

I know there's programs out there that do that

but if you're not buying one of those programs

and you wanna do sign yourself,

we teach how you can do that within chat GTP

where you just say, okay, boom,

you copy and paste this in,

it automatically knows that that's from Carfax

and that automatically spits out the maintenance

that's due based on the mileage of the vehicle.

So some of those things are other things

we teach and talk in the class

and we talk about how to make some service writer assistance,

how to put some of your SOPs in that draw some assistance

out or just how to build an agent

for very specific case in your shop

that you wanted to do something very specific

every single time and your shop may have a process

or something different than anybody else

and teach you how you can box AI in to do that.

And then we get a little bit in at the end,

a little bit in the what you can do with Claude

and some of the custom stuff

and then building Chrome extensions and all the stuff,

like I have some really crazy extensions

we use in our company.

But the basics start with how do you box AI in?

Once you look at a GTP working,

then you can move on to the next step in custom coding

and all sorts of things that work better

or more advanced uses of it.

But I tell people, let's start here,

this is kind of the starting point.

You used your own shop, EuroTech,

as your example in the session.

What was the first thing you created with it?

The first thing we created with it was our HR,

which just answers all the HR questions.

Can I take PTO tomorrow?

Well, no, you can't take PTO tomorrow,

but it will tell you why,

but CAI will do things that,

you know, in a small repair shop, right?

Your HR is generally still your managers.

And there are sometimes sensitive topics,

things like that,

that maybe employee doesn't want to talk about.

So if it says, well,

tell me a little bit more about why you want tomorrow off.

Okay, well, my mother died.

Okay, well, then this applies to, you know,

we have Minnesota safe and sick and earned safe and sick time.

So every state's gonna have different laws.

You gotta make sure that you put in all the parameters

for your particular law

and it's in your handbook, things like that.

Well, it will then tell the employee that,

yes, that would be qualified for Minnesota safe and sick.

You would be able to have tomorrow off

for bereavement or whatever.

And so, you know, here's how you would request that.

And it takes some of the pressure off having

to have those uncomfortable conversations

or employees that may not open up

and they'll just type into this bot

and it'll give them answers

as long as you've tested and validated

that you'd like everything at spitting out

and you've thrown everything or in the sun at it, you know,

that's one of the ways we've,

that was one of the first ones we did.

Then we did a technician assistant,

which we'll rewrite the technician story,

but I've also trained it on five or six different online

resources and databases and how to go to NHTSA

and search for any service bulletins that the car has

so that if detect types of story

and there's a bulletin and he missed it,

it will say, hey, did you see there's a service bulletin

for this issue?

Or it'll ask them some questions,

kind of like one of our master tax

and say, hey, your story is really good,

but I noticed you may have missed this, this and this.

Do you wanna check these things

before you submit this or submit the story?

And that way it'll guide them through checking

some things they may have missed.

And that was one of the second ones we wrote.

And then we built the service writer one,

which is the maintenance and how to overcome objections.

And if the customer says this, say this,

and we built some sales types controls around it.

But the service writer one I haven't used for,

oh boy, 68 months now,

because we've moved it over to our Chrome extension

that interacts automatically

with our point of sales and everything

and just reads the data off the screen.

How long did it take you to implement?

Like even the HR one for starters, how long did that take?

In our company, usually once I build something,

implementation is handed off to Daniel Gravely,

who's my COO and he is my implementation guru.

And so, he will then look at the process,

look at timing, he may test it at one store

and then wide roll out.

But as far as implementation, he is the master of that.

And he's my, if you read RocketFuel or EOS,

he's my integrator to my high visionary.

And he will then go do the thing he does

and he makes those things happen.

So, I create it and if he likes it, validates it, tests it,

he will figure out how to roll it out.

I don't have to deal with it.

Did he share any feedback with you as far as

when you did start implementing these things,

did your team have any reaction?

Yeah, he always has feedback, oh yeah.

This doesn't work, this is causing issues, fix this, fix this.

There's always feedback.

We run Slack for all our teams.

So we have a service writer AI Slack channel

and some of our other tools have a separate Slack channel

and so there's feedback there.

And when I'm building this stuff,

I'll build it as version one, version 1.2, 1.3, 1.4, 1.5,

typical software coding.

And so like our service writer Chrome tool,

I just was working on last night

and I'm releasing version 2.1,

which there was a couple pain points

that were identified, were solving.

So version 2.1 will get pushed to the team.

Probably later today after the meeting,

after I do a little more validation test

after this podcast, but that is how we do it.

I can't say how it works for everybody else.

Everybody's shop's so different

and that's where the AI stuff becomes so unique

because forever we have had to deal

with all the software providers out there

and all the software, SAS, software as a service,

all the software as service providers,

you're stuck with how they built it

and you have to work,

you have to change your business to how they built it

where with this type of stuff,

now you can really build what you want

and how your shop works

and you don't have to try to work

within the parameters of some other software

that does 70% of what you want,

but doesn't do it in the way you want to do.

So I think unless you have a product anymore

that is so good and such a database

that nobody else can get access to

or it's such a kick ass software

that nobody can say no,

that I think a lot of these softwares of service companies

are frankly in trouble

and you actually see that on Wall Street

and all over with some of these software service companies

dropping evaluation and things like that,

I think you're gonna see more and more of it

and you're also gonna see an explosion of cash grab.

I mean, I would say,

I'm sure you've seen it in the past three months,

there's companies advertising all over software

and you look at it and how they vibe code it

and how these cloud code and some of that

and the reality becomes anybody could build that software

really, really quick.

So somebody can launch a software in two or three months,

somebody else can build that in a month or less.

It's very true.

I like how you added the details on it though,

where you kind of layered it and curtailed it

to your technicians, your specific shop

that they can, like you said,

they can kind of follow along a checklist

that these are things we look for also.

That's interesting that you can kind of layer that on

as needed.

Yeah, you can layer that on and you can change it

and I tell people like you're better

not to make one agent do too many things,

you're better to have five or six agents

to do very small defined things.

It's easier to control what it does in that box

and the agent just sit on a tab.

One of the reasons we use the GTPs

is you can distribute that to your team

and push a link out where they all use the same one

and you can share that where Gemini

is a very, very powerful AI

and I really like Gemini, I use Gemini

for a lot of things too, but for the longest time

they're gems, which was Gemini's version of GTPs

could only be used by the person that created

and not shared out.

So that gets frustrating because then you have to create one

for every team member, which is not,

it's not gonna work.

Where the GTPs get super powerful

is the shop owner or admin or main guru in the shop

wherever it does it can create one

and then share it to the rest of the team

and then they all use it.

They can't change how it works, only the creator can.

And so I can make changes, but they just use it

and then they have access to be able to use the tool

across the shops and so that is, you know,

that's a difference and that's why we use that

as our test case.

Now, is there plenty of other AI,

lots of other things you can do,

perplexity, co-pilot, like I get a million questions on AI.

And I said, guys, there's too many out here

know everything about everything,

but I know enough about the big players

and here's where I see different things.

We're gonna concentrate on this one today

because I'm gonna teach you something on this

and you can take that and use whatever you want,

but this is what we're gonna talk about today

and how we're gonna work on it.

And I wanna show you one really tight example

and something you can build and walk away with.

And, you know, I would say a lot of guys

bought the cheap personal subscription

and by the end of the session, they were going,

how do I upgrade to the business one and merge it

so I can share this out and do more with it?

But, you know, my goal as an instructor, right,

is because the class has a prerequisite

of buying something to come to class,

I wanted to make sure I gave people the option

to buy the absolute cheapest subscription

they could to come to class, which was 20 bucks.

So, you know, I will say my class

had a little bit of barrier to entry, right?

You had to buy something to get in,

but unfortunately, most people liked it,

or if they didn't, you know,

they could cancel their subscription,

but that is the tricky part,

is some of these softwares can add significant costs

depending on what you're buying.

So, we try to work with something

that's very basic, very inexpensive to get you in,

play with it, try it.

We also talk about not sharing your data

to the outside world.

You know, some people are using the free version,

they're dumping P&Ls into it

and they're dumping private sensitive data,

which is absolutely crazy with no data controls to me.

The paid versions all give you data controls

and unlimited ability to share some of that out.

So, there's a lot of things we kind of touch on,

as well as, you know, I've collected now 200 some surveys

from people I've done this class for

and I have a pretty good idea of where the industry is,

AI readiness, things like that,

that where people I've identified

that they see holes, advantages, things like that.

So, I'm collecting some good data

that eventually will share with people

and say, here's where industry sits

and honestly, when I share it in class,

people are like, oh, so I'm not that far behind.

I said, no, you'll hear the top one or two percent

of the advanced people talking all the time, right?

Just like anything.

Sure.

The top people talk, everybody else feels

like they're miles behind and I try and share the data,

like here's where everybody's at.

You're not far behind if you're just starting this journey.

And where are we sitting as an industry?

Can you give us a sneak peek of what you've uncovered?

Yeah, I mean, I think as an industry,

I think the top, you know, I think there's a 5%

that are really super aggressive at AI

and it looks to be, you know, 20 or 30% are still exploring it

and still learning what it can do

and still have basic things built.

One of the guys in my class said,

well, I thought this was gonna be about automated agents.

I said, no, this is about agents that we control

to some degree.

He goes, well, I already have GTPs.

I said, okay, so, you know, let me see what you have

because he was in class early and I looked

and I said, yeah, you did a pretty good job.

You have a really good start.

I think by the end of class, I think you're gonna,

some stuff's gonna be repeat, but I think by the end of class,

you're gonna learn some stuff.

But if this isn't the class for you, you know,

we can get you registered for another class,

I'll help you out, I'll go to the registration desk.

And by the end of class, he had said he had redone

some of his GTPs and he says they're much better.

You know, so I'm always open.

I mean, there's guys that are doing a lot of stuff

and maybe I have a trick or two that can tweak something,

but everybody will.

I mean, am I the end-all, be-all of AI?

Heck no.

I mean, I'm learning every single day

and there's guys out there that do things

that I can't even dream of with it.

We all start somewhere and I think that's the biggest message

for people is you're not that far behind

if you haven't even played with it yet.

Yeah, for those people who wanna get started,

where do you recommend getting started

with the agents, for example?

Yeah, I think the GTPs are the easiest thing to start with

because it has the biggest impact

to be able to share out to your team.

So that's primarily where I tell people to start.

And that's been where I push people to start to begin with.

Relatively cheap subscription,

relatively decent controls.

And once you get something working there,

you know you can copy and paste that

into other programs and other things to make it work.

So that's kinda my starting point now.

I mean, obviously, just like everything,

all the AI stuff, all of them became political

and certain people won't use certain AIs

because massive exit of Claude or GTP is better,

or I have fortunately somehow

that everything became political,

but also some people won't use things

out of spite for different things.

And so now, you know, can you use Claude to do a lot of it?

Yes, I mean, you know, I use Claude co-work a lot.

I mean, it sends me a daily brief based on my calendar,

based on everything else.

It can go into my, I mean,

it auto does stuff for me at 7 a.m.

At the end of the night, I have Claude set up.

It goes to my inbox and any emails that I haven't opened

that are spam, it will unsubscribe me from them.

There's a lot of things you can start doing

that you have autonomous agents,

like simplifying your life over time,

and that gets in the way more advanced stuff,

but that is, there's getting a lot of requests

for part two of what I did.

And so we're working on that along with refining the workbook

to have some better prompts in it,

not use my exact prompts, give them some more generic ones

that they can build off for their store,

because we were showing mine and they were trying

to copy mine, which I don't mind sharing,

but my prompts might not be 100% what they want.

So we're refining the class a little bit.

It's only the third time we've done it, so.

Is there any mistakes that you've seen people making

that you could recommend or steer people away from things

to avoid when experimenting with these types of things?

Yeah, trying to make it do too much.

That's the biggest mistake.

They're trying to make one agent do technician,

service writer, or just trying to train a model

to do what they want.

Well, if they get that model trained on them

and they get that model doing what they want,

it doesn't help the rest of their team.

But that's where the GTP agents are super powerful.

And the downside I see a lot of guys

trying to make it do HR, service writer,

trying to make it do too many things.

I did that and I had a jumbled mess

and it wouldn't do anything I wanted to do

by the time I was done.

I'm like, this is terrible.

I mean, then I really learned that, you know what?

I can create 10 agents that do 10 very specific tasks.

And I'm better to do that and just make sure

I label what they specifically do.

And my guys can plug away at putting things

in the different boxes to do different things.

So make sure you segment everything

according to where it's needed in the shop.

Yep, yeah, try and make it,

don't try and make it do too much.

That's the biggest downfall

because it starts to confuse itself or it will do,

it just doesn't work.

You can't box it in enough to try to do too many tasks.

Now looking ahead, people have concerns

about AI taking over people's jobs,

things along that nature.

But if it's used effectively,

where do you see AI playing a part in auto shops

to their advantage, not necessarily to their detriment?

As far as using it to their advantage,

I think you're gonna see an influx of people doing that.

You know, it's not gonna replace people.

I think it's, I look at AI currently,

unless something drastically changes for our industry,

it's not a replacement.

I call it a time multiplier.

So maybe you won't need,

if you're a shop that's trying to run

what we call two and a half people, right?

Two people plus a junior writer or a reception or CSR.

Maybe that CSR gets replaced,

but a true service writer isn't gonna get replaced.

Technicians aren't gonna get replaced,

but it's much more a time multiplier.

So how do we use it to multiply the time

and speed up the delivery to the customer?

I think that is where AI excels right now.

Could it change? Sure.

But as it currently stands, that's where I see it excelling.

What's one small step a shop owner could take this week

to start experimenting with AI in their business?

I think just playing with the apps,

I always recommend the paid version of anything.

And then making sure you're not sharing it with there and in it.

I think just start playing with it, seeing what it does,

what it spits out that you don't like.

I think just even if you're only asking it

for specific insights, if you upload an Excel sheet

or you're asking it to check my policy,

if you upload a policy,

if you just start using it for small things,

see what it does.

And the things that you don't like that it does,

that is a great way to box it into your GTPs next

and things like that.

But some of it just start playing with it.

There are some good podcasts.

Well, probably this podcast now at this point,

but there are some good YouTube videos.

There are some other things in the market for it.

I have a feeling there'll be other people

starting to teach classes like this.

I think based on the feedback from Vision,

I think more people want something like this

and I'm not going everywhere.

So the more people that want to teach something,

they may bring a different spin.

Like I think there's other really smart people

in this industry that can teach this as well as I do,

that can get out there and do it too.

So I think you're going to see more

and more industry training like this

as well as giving people options.

But it sounds like you're already gearing up

for a part two, perhaps?

I am working on a part two

for some of the custom coding stuff.

Yeah, I'm working on some PowerPoints

and things like that as well as,

I mean, generally I'm known as a technical trainer, right?

I'm usually teaching BMW and Tesla

and advanced vehicle technology stuff.

But once in a while,

I'll step into the management side of the speaking world,

but for the most part, I generally speak on technical.

Well, we're definitely looking forward

to hearing what you come up with.

Ladies and gentlemen, Seth Thorson,

thank you so much for joining us today.

This is a very hot button topic right now, as you well know.

Thanks for having me.

You are welcome.

That's going to do it for us today at Ratchet and Wrench Radio.

Be sure to check out other episodes

on your favorite podcast player

and leave us a review if you found the episode to be helpful.

Also, follow Ratchet and Wrench

on Instagram, Facebook and LinkedIn.

You can get our Ratchet and Wrench Insider newsletter

delivered to your inbox four times a week

at ratchetandwrench.com forward slash subscribe.

Thanks for listening

and may the rest of today be the greatest day of your life.

AI in the Bay: How Shop Owners Are Building Custom Bots That Actually Work

retrieval-augmented generation (RAG)

NHTSA

service bulletins

service writer workflow

About this episode

Original notes