Unlocking AI’s Potential in Manufacturing: A Conversation with Adam Marsh (Part 1)

All right. Welcome back to the Just Some BS podcast. Today we have Adam Marsh on. Adam, thanks for coming on.

Absolutely. I appreciate the opportunity.

So Adam, I like to read from people’s LinkedIn bios. It’s president at Ledge Inc. CEO, 8020 Quality, leading practical AI adoption, which we’ll get a ton into. The one thing I would like to ask is, can you summarize your career in under 90 seconds?

Sure. Industrial engineer out of Penn State, left and started working in manufacturing and consulting primarily at St. Orange Company. After five years, had the opportunity to go out on my own and work with not only myself, but my dad doing quality and manufacturing consulting all over the country. Since then, we’ve grown to a team of 10 focusing on quality management systems, implementations and support for manufacturers. And now we’re really pivoted and been able to jump on some of the AI options and I’m sure we’ll dive into how we’re able to help there and be the right kind of connector for that industry.

Yeah. Well, that’s amazing. So you have a huge engineering manufacturing background, so you probably walk through shops and just go, “Problem, problem, problem, problem efficiency, efficiency, efficiency.”

It’s hard not to be like, “Oh, hey, I got 500 ideas for you right now.” I have to slow down sometimes and deliver.

Yeah, because it’s probably a little scary, especially for customers that you’re working with because that’s a lot of change, especially when they’ve been doing things for a long time.

Yeah. And you’d be surprised, especially when you hit the manufacturing floor. Most of them are really good on the manufacturing floor. We spend more time fixing the backend processes than where they’re … And that’s stuff that’s costing them a lot of money that is kind of hard to see.

Is that more front office or quality on the backend or yes?

Yes, absolutely. It’s a lot of opportunities for them to look for that low hanging fruit of things that are … They kind of fix every day. And I would say good manufacturers, at least Central Pennsylvania is notorious for this. We’ve got guys who get stuff done, but that doesn’t mean they always do it the easiest way. And so they will continue to do things the wrong way or just to deliver on time. And they just keep doing that and banging their head against the wall and they might not slow down to fix that issue or they could fix it and it never be a problem again.

Yes. That’s fascinating to me. So I came from the power construction world into manufacturing. And so that whole process, I don’t know how to do pretty much everything in the shop. It’s why I’m doing half my summer out in the shop, but just to learn some things. But having the ability to not get bogged into the way you’ve always done things and be like, why do we do this? It’s a really stupid question, but I don’t know the answer. And then it just starts unlocking like, “Oh, I don’t know. I don’t know why I walked past that for 15

Years.”That’s where getting that outside perspective is really important and looking for those opportunities. I absolutely commend you for going to the shop to work. I would say one of the, and I just talked about this yesterday, one of the most pivotal things that I did was as an intern. I interned at a union shop. And over the summer I was doing engineering, solving engineering problems for them, but I was also their go- to guy when guys went on vacation.

Nice.

So hey, laser operator’s out for a week. You’re running the laser for a week. And what a great opportunity for me to move around that shop and actually execute and run equipment and see what it’s like day in and day out. And I think everybody in leadership and in engineering really needs to be willing to go spend some time actually cutting parts and doing those things.

Yeah. And I think a lot of manufacturing is probably internal promotion. I know a lot of our team have spent a lot of time on the floor, so I’m the odd person out. So I’m kind of going backwards into that and trying not to screw things up and being completely … What’s so great is when you get out there and you just start really asking questions and people wanting to share their knowledge and understand and show the process and stuff like that, it’s quite humbling

Experience as well. You realize how smart these guys are. Manufacturing is not like you don’t want to go to college, go into manufacturing. It’s this easy, no way. It’s for smart people and you get to use cool science and tech to do hard things. I totally agree. We’ve really been trying to push that message at the manufacturing association. It doesn’t have to be down and dirty. It’s high tech. It’s you want to use science and math. Here is applied science and math to do cool things in the world.

Yeah. I mean, one of our core values is creativity. And I think that people ask themselves why, but it’s artists. Manufacturers, especially metal fabrication like we’re doing, we’re taking raw materials and turning them into final products that could produce food. It’s

Insane to me. That’s funny. One of the projects I worked on was I was helping as an industrial coder really address one of their biggest issues, which was rework.

I mean, they were spending hundreds of thousands of dollars on rework. And so our focus was if we eliminate rework, we don’t have workers anymore. We’ve got craftsmen. And so we implemented a craftsman bonus where we really figured out, “Hey, we’ve eliminated all this rework.” And the great part was we actually offered to split it with the painters, anything that we stopped. So we knew what we were spending and what we cut and have, we split it with them. And so not only did they get a bunch of money in their pocket, because they’re throwing up their hands saying, “Hey, this doesn’t look right. Let’s stop it now.” Our throughput went through the roof because we weren’t reworking stuff all the time. So putting a handle on that and then paying people to be that craftsman, incentivizing them to be that craftsman, critical.

Does it surprise you or have you had enough experience with that incentive side of things that the wrong incentive, the wrong approach to incentives or commissions or bonus structures on the wrong things can completely deteriorate what you’re looking for without even knowing it.

Yeah. Those things, number one, I don’t like to say we’re going to promise this incentive program because, and I would say I’m guilty of this with my own team. Me too. I still haven’t figured out the right incentive program. And I probably try, you have to ask when our team is in sometime, I think I’ve had a new incentive program every single year. I was like, “I think I can make it better. I think I can make- You’re my spirit animal. But I want to do one. I just want to make it the right thing.” And so we just keep tweaking it and just keep tweaking it. And so that you have to be willing, number one, to bite the bullet when you probably did it the wrong way. And then to be willing to, hey, experiment and have a team that trusts you enough to say, “Look, I’m trying to do the right thing.

I’m trying to incentivize you the right way.”

Yes. I’d love to have a deeper conversation around that because that’s probably one of my biggest failures at SWF is creating this incentive program that I think I was smart enough in math to figure out and I clearly wasn’t. And then a month into it realizing this isn’t going to work and having to pivot back and say, “Here, we’re going to do a bonus for now until…” It took me in just two years, not in just two years, but two years later, I finally had the eureka moment and spent an entire morning in ChatGPT for four hours just so peeing, right? And then I sent that to a really good friend who does financial modeling and I said, “Can you turn this into something that’s actually doable?” And that’s what he’s doing right now.

Interesting. So I would say one of the biggest hurdles there is the data collection. So if I’m trying to do an incentive plan, I have to have a good way to collect data behind it. And then looking at how I’m collecting that data, how granular and how accurate that data is because you’ll get people who … And I’ve never had this on my team where people are gaming the system or something.

Yes, I hear you.

But people certainly could do that. So you have to be cognizant of how do you avoid that part of it? I think once you start giving it to a pretty high level, and I don’t like doing individual incentives, I really like doing team ones. That’s

What we’re doing. Yep. There’s one equation that we all work through.

Yeah. And I think it really then helps to supplement that teamwork aspect. So I would say that you got to focus on something that they can see too. And that’s been a struggle for us is I’ve tried to tie it to dollars or days, and it’s been a tough thing for my team to see. And this year we finally have been able to deliver on essentially hours and say, “Here is the hour goal. Get to this, right? Get to this weekly, get to this monthly, get to this quarterly more happy.” And I think that’s been a big deal to kind of set that stage and set the expectation for the team.

Yeah, that’s really good. I struggle with the same thing because it needs to be simplistic and needs to be understandable because people will play the game as long as they understand the rules, but you have to be able to take something. I want six data metrics that are important to me, but that’s complex, but then you have to scale that back into simplicity of collecting the right data, being able to present it, and then understanding the rules against that. So you’re right, that is the hard part, especially if you want to something that everyone can affect because then it’s really a game and you’re like, you really focus on the little things. So I’d be glad to have a longer conversation. I’ll send you my SOP and you can be like, “This is horrible or not. ” So I’m totally cool with that.

I love using the ChatGPT though to help work through that thought process. I mean, it’s a partner in that and that’s a great place to start. Yeah.

If anybody takes anything from this entire conversation, I would say treat it before you go talk to any expert, just start with it and have a conversation like a friend that’s really smart at anything. Because even if it doesn’t give you the solution, your SOP or your SOW, your scope of work’s going to be so much tighter, which means you’re not going to have a back and forth with an expert, which means you’re not going to waste all that money on discovery and you can go right to, can you do this?

So it’s a great example. I’ve been more willing than ever to do projects that I never would’ve done before because I have a partner that doesn’t make fun of me when I ask a specific question. 1,000%, dude. So I will go … I mean, my wife and I did a kitchen backsplash. I am not the most handy engineer on the planet. And my wife is an art teacher and so she’s going to pick a complicated tile. And so we did it in a weekend. We’ve been pushing this off since 2016 when we moved into the house.

And

We just did it this year in a weekend and I had two tiles left at the end. And it was all because I had the confidence to do it. I was much happier with my math because I had the backup and I could ask questions like, “This isn’t going well. What do I do?

” Yeah.

So that was amazing. And then I just posted a video, I don’t know if you caught it. I learned to weld in college. I am a poor welder, but I would love to spend more time learning. And so I bought a welder because I needed to do something with one of my tractors at home. And I said, “You know what? This is a great opportunity to learn.” Well, ChatGPT helped me figure out what to buy, all the equipment. And then I welded and it looks horrible, but it’s holding. I call it a farm weld. And so I said, “You know what? I should take a picture of it and see what it says.” And it inspected my weld and told me to immediately rework it. Right.

Grind that baby out. I was like, no. And so I said, “Wow, that’s pretty interesting.” So I just posted a video probably last week. I built a weld inspection tool. So you can take pictures, give it context like your weld procedure, what type of welding material you’re using, drop in pictures, and let it look at it and give you advice. So imagine that you’re maybe a manufacturer and you don’t have weld experts in house, but you send something out for weld and you’re a little concerned or a little … You can take a picture, upload it and get advice. For me, it was telling me, “Hey, you really need to up your speed.” And so it was giving me that back and forth. And I think it’s just really cool to have that, number one, instantly, but two, be able to ask that question and I didn’t have to call an expert.

I have that expert in my pocket.

Dude, that’s fascinating. So our minds must think very similar because I just started building a root cause analysis with the same concept for safety. So you did the same thing. So I’m taking incidents, all the narratives, pictures, plugging it in because I want a root cause, which is really hard to get to as human beings because our fears get in the way, whatever. And then suggestions to help it never happen again.

So I actually start with having a company sort of let somebody brain dump in. Here’s everything that happened.

Beautiful.

And then I make them upload pictures or anything they’ve got. They’ve got any document that you can give me more context. And then I make them walk through a very simple fishbone, right? So like, “Hey, where man could have been the issue?” And I want them to put things in like they would put things in. And if you’ve got a manufacturer, they’re going to say, “Well, Jeff on the CNC machine doesn’t know

What

He’s doing.” And so, okay, put that in. That’s fine. We’ll let the AI tool interpret that and then help us drive the solution. So the first thing I do is send it to make a really good problem statement. And so make that problem statement. And then I actually send that to say, “Okay, use that plus all the context to deliver a 5Y for me.

And so I have it do that and then I take that to that next step of what’s give me some suggested corrective actions. It is beautiful for that. And I tell you, your customers will appreciate that so much when you’re delivering corrective actions that are actually effective and actually it’s not, “Hey, we’re going to retrain Jeff.” It can’t be that. And so you can get to that sort of deeper level with

It. I may have some questions for you around that then because I’m down that process and I’m just like, I mean a million. I’m with you. My mind just doesn’t sleep now because I’m writing myself notes. I have a daily task advisor for the shit that’s coming out of my head that I’m just like, “I take this down because I don’t know if it’s going to leave my brain and then I go do something else.”

Yeah. I’m trying to build what I would say is like one tool per week because it lets me slow down a little bit, and I really have to think of, okay, what is my end goal? What do I want my output to look like? Cool. What are my inputs going to be and then build the tool in between. And by the way, the biggest trick here is you use ChatGPT, you use the AI tools to build it. So it’s going to coach you through all that and help you, did you consider this? That has been sort of really a game changer for us where I’m able to think through those processes and identify things that are good use cases because that’s been a problem across the board when I talk to people about AI is number one, they don’t think about a use case. They can’t find one because they don’t have enough experience with it yet.

Or I think they’re thinking a little too big to start

And

I really want them to do that crawl, walk, run where let’s slow down, let’s work on some little problems first and get really good at this and really comfortable with it before we dive into these big ones. And then if their data is no good, it’s only going to be as good as that. So I think we have an opportunity now to say, “Here’s the problem we want to solve and we’re going to solve a year from now because we’re going to spend the next year collecting data on it.

” Yeah. Do you ever send your ideas out to a third party to build something larger than you have the ability? Do you have the ability to code solutions and everything?

So I would say I have two partners that I work with that can build if it gets out of sort of what I’m able to do.

The

First question we get whenever I introduce these is, cool, can I connect it to my ERP system and have it start doing stuff? It’s like, well, number one, what’s the stuff you want it to do?

Right. Do we need to?

Do we need to? And two, are we sure you want to start there? Can we start with some little thing that’s wasting your guys’ time and solve that problem? And I think leadership is spending a little too much time saying they’re going to come up with what the solutions are and instead they need to give the tools to their team because leadership doesn’t know what all these small problems are, the things that are costing their people time. One of our tools that’s the biggest hit is just a tool to review material certs when they come in. And all we’re doing is checking that cert, which if you’ve ever looked at one, they’re like the worst document ever, right? It’s like a copy of a copy.

Yeah. Yeah, you can

Barely read it. Yeah. And either you’re expecting your receiving inspection person who may not really be qualified to look at those properties or you’re bringing your engineer to come do it and wasting his time or worse, you’re not looking at them at all. And so if we can run it through a tool, instantly check it versus the spec and get a go, no go, that’s a great alternative. Now look, if I’m building nuclear parts, I still might have somebody look at that thing, but if I could check it the second it hits our dock versus three months later when my engineer has time, we can address it right away. So I like this idea of speeding those kind of processes up and looking for opportunities while we’re doing that. Contract review, I think is a huge one. I mean, it’s perfect for that, right?

Yep. Did that a year ago.

And so what I like to do is build your quality system into the back end of this tool. So it’s comparing their supplier quality requirements, their terms and conditions all directly to your quality system and looking for the holes.

That’s brilliant.

They’re going to have a thing in there like, “Hey, we have a 10-year record retention policy and what do you have? ” So it’ll check that instantly and give you a heads up. We’re trying to deliver a list of what are the final deliverables here because you know it’s not just the part, right? It’s the part, it’s the certs, it’s the- The warranty, yep. All that kind of stuff. And then what are your flow downs? Because I want to catch that upfront so we make sure we do flow that down right away to make sure all your vendors- No surprises. Yeah. No surprises. And then hopefully we can … And I’m not sure how much you guys are getting into that sort of PPAP world and first article, like really deep first articles, probably not quite yet. No. But when we’re looking at those, you’re talking about data packages that are tens of hundreds of pages long that we have to deliver with parts.

Well, if we can prep that on the front end, we’re not waiting on the back end to get this thing done. And that’s one of the things that holds production up, right? We got to wait for the customer to approve this. And if we can have it ready at the end, that’s a big deal. And by the way, I want you to pass it the first time. You’ll get customers that push back for, “Hey, you have a problem on page three. Okay, go fix that. ” Great. They won’t look past page three. So now they push it back and you’re just in this back and forth for like a month of trying to get this part to them and you’re not getting paid in the meantime. So how do we fix that upfront?

Yeah. I’ve actually been racking my head around a schedule solution in that space. From what I heard, I actually interviewed an AI kind of implementer two weeks ago and she said Claude’s getting very digging their claws into the construction world.

So I’m working on a scheduling tool. And this is the biggest problem I’ve worked on so far because I would say scheduling is my customer’s biggest issue. And biggest issue in terms of potential risk because they all have, I like to say is a scheduling wizard that works for them. And it’s like that one guy who knows where everything’s at and how we should later-

That used to be me.

And number one, that’s a risk. You’re right. Number two, I want to give him help and I want to be able to stress test that schedule. And so I’ve been working on some tools for that and I started really big. I wanted to see how big I could go and like Crash, ChatGPT-5, Gem, I’m crashing them all. So cool. Now I know I’m probably too big. And so I got to start pairing that back to say, okay, let’s look at a little bit smaller scale. Let’s look at a one month schedule instead of a three month schedule. I’m trying to include things like who’s available to do what machine, who’s on vacation, when is the machine down for me? I want to get all of that. It’s a big logic problem that people are solving right now. And your ERP systems are terrible at it.

Oh, you can’t even find one.

We’re

A Primavera scheduler where I came from in the nuclear and fossil world. So I don’t know what program you’re used to running, whether it’s projects or Excel or projects.

We’re all over the place, right? And so, I mean, my team, we use some software called Scoro that’s been really, really effective for us on the backend, just tracking projects. And so I’m starting to use it for my team of consultants. Look, this isn’t just for manufacturers, right? I mean, I’m using this all over.

It’s a construction problem.

Construction problem. It’s service companies, right? I mean, I’m out. I want to make sure how do we optimize our schedule? I just used it. I haven’t shown this to my team yet, but my team tends to do full days and half days at customers and half days can really hurt us. And so I just used it to optimize half days. I looked at every half day we’ve done in the past year and I have the address of every one of our customers and said, “Hey, you did two half days in this area in the same month, you could have mixed those two. Let’s optimize those.” And so we’re looking for those opportunities and that’s been really fun to kind of dive into.

Love that. I love this conversation already. I didn’t even get into my first question. Yeah, I want to talk more globally because I think that’s going to help some people so we don’t nerd out the whole time. Obviously spreadsheets, manual tracking are huge in manufacturing. What was the pivotal moment that made you say we need secure AI tools that actually work now? Did you have like a major Eureka?

I would say about a year ago, I spoke at FabTech about it.

Nice.

And so I got to do a session and we were really trying to connect potential use cases for AI. And we went into it and it still felt like it was a little early, maybe a year ago. Things weren’t quite working the way they are

Today. Lots of hallucinations.

Yeah. And so I was showing people like, okay, here’s some cool stuff we’re doing, but it still wasn’t quite there. And over the past nine months I’ve been looking, I got really pushed in that event to like, okay, deliver something that works, show me something. And so that has pushed us past the line. I can tell you, my customer’s trying to still use Excel or geez, paper. I mean, it is my mission to get our customers to hopefully industry 4.0. And with the first step of that is just stop using Excel and get into databases that we can access.

And so then I can use the AI tools to kind of drive that. And so we’re trying to get them past that. A couple years ago, we started an app company called 8020 Quality and we’re trying to capture error data and just make it easy. Put it on the guy’s phone in his pocket that he can take a picture and capture that problem really quickly so we can number one, solve it. That makes ISO happy. But number two, collect that data of the why, the department, that’s the problem, right, the reason code behind it, and then be able to trace that back, look for that 80 20. So since then, now we can take that data set and drop it in. But I would say the last six months we’ve been looking for that solution that is secure because that was the pushback I got with everybody.

I’m like, “Hey, I have some stuff that’s working.” Yeah, but we can’t put anything in because we don’t trust it. And so we found a partner that we work with now and we build on top of. And so our platform is ledge.hats.ai. And with that tool, we’re able to provide a secure AI solution. And when I say secure, it’s SOC two. It’s got that login requirement, but it’s also all the learning is turned off across the board.

Nothing’s going out.

Nothing’s going out. It’s not learning. And the trick is, and this is, and you just mentioned Claude, is we give access to multiple models because we’ve learned … So there’s 48 models in ours. We’ve learned for different projects, different models work better. And so now we’re able to sort of deploy the right model at the right time and start to use those. You fall into the ChatGPT five, you fall into that trap, you’re almost falling into … I just attributed … ChatGPT is a little bit like America Online right now, right? If you get back to the internet, where if you’re in their bubble, you’re kind of in their Shell and you just learn there and you don’t get outside of that. And so I like the idea of, I want to see what the new Google is and that may be Claude, right? I don’t know, but I want to have access to them so we can figure out which one’s the best one.

Yeah. I was just talking to a person last week about that and they said the same thing. They said they’re all going to find their niche and you’re going to know sooner or later we’re all going to know where to go for who, but nobody’s really established a niche yet because they’re not sure which direction. So it’s kind of the Wild West.

It’s the Wild West and I see them as there’s a lot of developers who are good at this. There’s a lot of computer guys, but I don’t know that they’re … Where we come in is we’re the connector between understanding business systems and being able to use the AI and putting it and looking for those use cases and developing the solution. So I think that’s what’s sort of missing from it right now. Agreed. As a communicator,

Yeah. Yeah. It’s like I’ll make you … They don’t know. They don’t know manufacturing. They don’t know this kind of business and so they don’t know what solutions need to be made. We’re early internet and we’re all just looking at websites. And so now we’re all trying to figure out, okay, let’s start showing what those solutions are. My goal is always … We build a lot of tools and it’s not because I’m trying to build the perfect tool for anybody, I just want to plant seeds of possibilities and then let people say, “Oh, well, here’s how that would work for us.”

Yeah, no, you’re dead on. It was so interesting when I was talking to that individual, which is really in the computer phase. We’re talking just like this where I was talking about practical places we’re trying to use it and they were like, our brains aren’t even there in your headspace of like, we’re worried about design and everything and you guys are applying it to real world things. And I was like, “Well, come over to this side. The pool’s warm. We need help.”

Yeah, exactly. And the interest is through the root, right? So I gave that speech at FabTech and then I’ve probably done it like 30 times since then and I’ve never given the same speech twice because you can’t, right? Every week you got to update. Dig, you must be online all the time researching. Yeah. Well, you know what? There’s some really good … There’s one called the AI Daily Brief that is awesome, like 20 minutes a day. And I tell you, that has really done a great job of keeping me up to date, but I’m just messing with it. And I don’t know about you, but the way I learn software is just messing with it. And so I got to get in and play with it. And that’s where I say, “Okay, I want to build this new tool.What’s my game plan?” And then I build it.

I’m like, “I didn’t like how that works. Let’s see how I fix it and go a different route.” And so I’ve been doing all that testing so we can then deploy it to

People. Do you have a specific process or formula that you pretty much always use no matter the problem?

Ooh, no. I would say your comment about Claude is dead on. I really appreciate how it writes.

And

So I would say- Very

Legalese.

Yeah. I don’t like how ChatGPT five writes, to be honest with you. Very

Conversational.

Yeah. And so I’ve been focusing a little more on using Claude and I’ve been really focusing on cleaning up that prompt to be very specific. And then at the end of my prompts, asking where it made stuff up. So on my welding prompt, I’m really focused on saying, “Give me a percent confidence you are. Come up with your confidence because I want … Say I deliver that to your team. If you just read what AI tells you, you think it’s 100% confident, right? But you get to the bottom of that and it says it’s 65%.” Okay, I’m the engineer. I can’t go with a 65% answer. But then what I do is I say, “Tell me what would give you more confidence in this? ” So it says, “Give me more pictures, give me a weld procedure, give me those kind of things.” Cool. If I can do that, I can get it.

Build that

In. Yeah. And now we’re starting to build that confidence and have better answers coming out. I would say one of the things people are asking a lot for is delivering me a finished document. And I don’t know, the documents coming out of ChatGPT look nice the first time, but good luck if you try to put a space in somewhere. It’s like a Word document that just explodes. And so I’m trying to say, “You know what? I think you’re better off getting the content and pulling it out and putting in your own templates and things like that. ” Totally agree. But I think it also forces the review. Yeah, true. Because I want to slow … I don’t want people to just say like, “Oh, it’s done, download, send.” That worries me. So I don’t mind a little bit of forced review for your team to be able to do that.

And then have you guys really put in an AI policy yet? What can people do? What tools do they have access to?

Yeah. It’s really funny you say that because I was reading an article about that’s one place to start, get in your handbook, close the loop, all of those things. There’s probably only … The people who are really digging in and maybe using data is maybe three or four of us, but then we’re providing solutions, but I just sent out a top 15 ways to use Microsoft 365 and Copilot to make your J easier, right? Okay. So that just starts the conversation of I’m going to take an email, which then that gets into a problem, so then we got out of policy. So no, have not yet, but it is on my docket, which I’ll just put it-

They say the number is 50% of employees are using unapproved AI tools. Okay. And if that’s the case, I feel like we got to give them … I don’t want to stop them, but I want to put guardrails around it. So I want to give them a tool that we’re comfortable with so they’re not giving away the farm. They understand what we’re trying to do, but I also want to encourage them to go use it.

se it.