Manufacturing AI: From KPIs to 310% ROI

Ed Ballina (00:00)
Hi, I’m Ed Ballina.

Alvaro  Cuba (00:02)
Hello guys, Alvaro Cuba here.

Ed Ballina (00:05)
So welcome back to another fine time with Alvaro and I on the Manufacturing Meetup podcast. Yeah, we have no guests today, so you’re stuck with us for at least, you know, the next couple of episodes, but we’ll keep it interesting. Because you know, this is where we kick back in our downtime. We talk about efficiencies on the shop floor while having great conversations at your local watering hole. Whatever that may happen to be.

Alvaro  Cuba (00:32)
And before we start with the show, what’s about the hat?

Ed Ballina (00:36)
So the hat, this is my bird watching hat. whenever I go bird watching, which I like, I wear this hat because it’s nice and cool. put on my head? Yes, bird watching. I’ve got extra telescopes we can put on. And you have a very, very fashionable hat there, my friend.

Alvaro  Cuba (00:45)
Yeah, look at this. I just came back from Thailand. So as you can imagine, I was enjoying Thai food, the palaces and the temples. Wow. And nature, tropical, very tropical, so great. But  a quick interesting fact related to supply chain manufacturing. Have you ever heard about a car made in Thailand?

Or a truck? They are the ninth in the world producing cars, pickups, and CKDs parts. And this is an excellent example of an offshore production. developed, licensed in Japan, the U.S., or China, send the orders to Thailand.

They take advantage of the free trade, cheap labor,  still good quality of manufacturing. They produce there and then they send it to be assembled or to be sold in Japan, the U.S. So very cool example. We talked a couple of times in the show about offshore or repatriation or local production. Well, this is an excellent example of offshore.

Ed Ballina (02:28)
That sure is, because now that you mention it, I’m a fan of Nikon camera equipment. ⁓

Alvaro  Cuba (02:37)
For bird watch, for the bird watching and picturing. Before that, just a quick reminder to our audience, we’ll do the same at the end of the show, but  hit the subscribe button if you don’t want to miss an episode or the conversation. So  join us, hit the subscribe button. Yes, down there. Great!

Ed Ballina (03:38)
So hey team, we wanted to talk to you about AI and AI transformation today because you can’t swing a dead cat without seeing an article about AI and you know, is it delivering results and all that.  We are all talking about AI transformation, but how do we measure if it’s actually working or not, right? Because we all come into these with great expectations and great ideas in terms of what it can deliver and the promise of it is tremendous.

Sooner or later, you have to pay the piper, right? And your finance person, your CFO is going to be saying, hey, we just spent all these million dollars on this AI and the cloud and where’s my money, right? We have to pay our mortgage. And you’re hearing a lot of that conversation that in general seems to go, we’ve invested a lot in AI. We’re not quite sure that we’re seeing the return. You know, do we need to tap the brakes? And at least part of what I’ve been seeing, I think Alvaro corroborated this earlier.

It’s agentic AI and really focused AI implementation that is leading the pack, right? And, and delivering real results. ⁓ so part of what happens is there’s this disconnect between the C suite that’s making these decisions, right? And maybe the actual implementation on the shop floor. So listen the C suite, if you convince us that this is a good investment, we check for technical rights to success. What are the areas that you’re hitting? Right.

Does this make sense to me does it pass the sniff test and what’s my return? Once you’ve got me kind of convinced, I sign on the dotted line. Well, we all know there’s a startup curve and we all know there’s implementation timelines and all that, but in the minds of those executives, they’re managing a million things. In their mind, I signed that already. When am I going to start seeing the money? And within a few weeks or months, depending, you’re going to start getting the phone call, yo, Ed, where’s my rent? Right? So  there’s a lot of questions being asked. And part of the problem is, the implementation of this stuff is not easy folks.  It takes a while, it takes a lot of training and it takes understanding your culture because at the end of the day, that person on the shop floor has to see value and has to close the latest yard in order to fix the problem at the end. So that disconnect is a bit of a challenge, but we’re seeing more and more examples of very focused AI driving some real huge ROIs. 

So speaking of that, Alvaro, what’s your take? What are the KPIs here that really matter?

Alvaro  Cuba (06:11)
Yeah, I think that is critical to start because AI comes second. First, you need to understand how you measure what you are doing. If you measure the right way, whatever you put, it’s going to be difficult to understand the difference, the improvement, and then sell or demonstrate the benefits of what you are implementing.

So just very quick review. First three, critical. We always talk about that. So the base safety, quality, and morale, your people. Now then you can go to performance and in AI, uptime is critical. Throughput is another good one. GE, another. No, that’s performance, but…

All that, besides improving quality and safety, have to impact your cost. So it means it needs to reduce waste, reduce energy utilization, optimize labor. The different components of cost has to be hit. And then  at the end, also very important and most of us sometimes forget, especially in the plants, is about cash. We’ll talk later about how to convince the CFOs and what is important for them. But cash, inventory, less investment. If you make your lines to run much better, then you don’t need to invest in a new line, so you save capital. But also, if your lines run better, your service goes better, and… you need less inventory because you are more predictable. So when you put all these together, the end result is a line running well, smooth. Well, once you have that, then you have an additional benefit, which is Your people is calm, relax, time for innovation, time for thinking, time for improvement, time for growth. So when you think about AI and how to demonstrate, you need to think about first in these KPIs, where are you doing well? Where are your gaps? And look for specific solutions that help in those specific AI.

But the connection needs to be whatever quality or cost or performance those need to impact one of these three, your revenue, your margin, or your cash. Otherwise, it’s not beneficial for the company. Once it’s beneficial for the company, will also be beneficial for the plant and for the guys.

Ed Ballina (09:34)
No, absolutely. You have to pay. You’re taking out a loan. OK, you have to make the payments. So the results have to be there. And one of the big areas that I think sometimes gets missed is the whole impact on waste. Right. Because we all know about uptime and, you know, lines run reliable. But waste sometimes winds up that your efficiency gain is the top of the iceberg. What’s underneath the water is even bigger. So.

One of the problems that we face, right, is our traditional measurements systems, right, which in some cases, frankly, are antiquated. think about, let’s think about vibration analysis. You are, you’re a forward thinker, so you actually have trained somebody in vibration analysis and data collection. And you have this person doing rounds of your, of your plant. By the way, this is a real world situation happening with one of my customers right now. That person is retiring.

And they are trying to figure out who they want to train in their place. And my comment to them was, Hey, you can have somebody check your equipment, you know, once a month or once every couple of weeks and get a reading. Okay. And that’s valuable versus not, but how about if you had some real time analytics, right? Where you were actually monitoring this machine 24/7 and provide, you know, data,  on an ongoing basis. Cause sometimes these failures happen catastrophically and quickly. You could have vibration data, for example, or infrared thermography data from a week ago or a month ago, and you can have that machine self-destruct in 30 seconds, right? So not that it happens all the time, the closer you are to real time, it happens. Also, a lot of the KPIs that we use are lagging indicators.

They tell me what happened. So when I look at a waste number, cents per case, right, I already committed the sin, folks. By the time I see it there, I’ve already incurred the cost. I’ve had to throw the junk away that I made poorly in the first place versus predictive metrics. When you use infrared thermography or oil analysis, you’re reading what the line is telling you. The line doesn’t have to blow up to tell you it’s a problem. You have to be listening. So you need to have the right ears.

Predictive metrics win the day. And let’s face it, we are in a major shift when it comes to  knowledge and experience on the shop floor, right? A huge wave of retirements, COVID, right, have driven a significant reduction in the  levels of experience on the shop floor. And those are not easily gotten to, right? They’re hard won. So, couple of thoughts for you to think about. But so let’s say we do know what to measure, right? Can AI actually move these needles? Let’s look at the proof and hint. Alvaro and I both say, heck yes. So tell them all about this Forrester report.

Alvaro  Cuba (12:45)
Yeah, we wanted to start with KPIs because that’s the only way to demonstrate things. Now going into the next section, is, and Ed said it, AI is working. And we’ll talk about in this part, how to validate those results. So first.

Couple of things about the actual state of AI in manufacturing. Let me give you just very quick examples on very specific areas where AI is very active in the plants, in some of your plants or in other plants. Data insights big with AI because AI manage sheer amount of data and the speed is unbelievable.

The other is safety with robots, behaviors. They can spot the behaviors they can tell you immediately. So you can correct it. Quality, the inspections. Ed mentioned this in a couple of episodes ago. Once you have AI, nothing passes. Nothing wrong passes these controls. Then you have machine health and production health and waste monitoring. You also have training, AI in training, AI in communication. So there is very different parts where AI, it’s already active in manufacturing. But, there is a gap in many cases because we are putting that, but in theory, we are not seeing the results in the big indicators of the company.

Because at the end has to impact the business otherwise. There is a promise, it has to be a reality. We have to be able to demonstrate it and to see it in the numbers. There is where nowadays one thing that is starting to work is external validation, a third party that comes and validates the results, helps you to see based on their experience with others, they help you to see where is the benefit. And not only that, help you to calculate that, which it’s very useful. No, and it’s very useful. And that will also give you an idea of how high is high. Right. 

If you continue, what is the size of the price?

Ed Ballina (15:48)
Very important.  Getting this alignment ahead of time is so critical. So if I’m a typical manufacturer, right, somebody comes to me and tries to sell me on any kind of a program, AI specifically, right, I’m going to have a couple of questions. First is, as I mentioned,  the technical sniff test, okay? I’m going to dig into your technology just enough that I can satisfy my intellectual curiosity and so that I can puzzle out whether I think this thing is going to work or not. Now you’ll be able to fool me. No question, because I am not an AI expert. But if right from the stop, you know, 50,000 foot view, the numbers don’t line up, you’ve got a lot of convincing to do. So. What’s that? Yes, the sniff test. Yeah. How does it smell? So then I would say, hey, so What about the payback? Right? Like how long is this going to take to pay back? And also tell me where the savings are coming from. Don’t tell me you’re going to ingest all this data and make my total supply chain 5% more effective. That’s really nice. Dig down to where I’m going to see that 5%. Maybe I’ll see 2% in waste. How did you come up with that number? We’ve had some pilots where we’ve seen this, right? So convince me that this ROI that I’m going to get a payback, right?

And another real pure question is, is this the first time you’re doing this? Because most of us have had in manufacturing an experience running a serial number 001 piece of equipment. And let me tell you, none of us cherish those experiences. We all want to be early adopters. But if you’re stuff, if I’m going to be your guinea pig and your R&D project, that’s a non-starter for me. I don’t have time for that. I may be willing to incubate you offline, but not as my priority goal. 

And really at the end of the day,  you know, what’s the size of the price for the investment? You know, this has got the potential to really touch almost every part of your business that is competing for mind time and creates disruption in the business. So are you, how are you really going to reduce my costs or is this a way to shift it to another area which just doesn’t interest me very much? So if you’re going to sell these, these are the questions that you need to ask. Do I think it’s going to work?

How quickly am I gonna get payback and can you give me specificity where these savings are coming from? Because if you can’t, it is not a proposal, it is a pipe dream.

Alvaro  Cuba (18:21)
And guys, is new, relatively new. So it’s okay. Sometimes it’s difficult to calculate and to answer all the questions that Ed is doing. But you are also hearing from your competitors, from your peers, from everyone, AI is there and it’s working. So how you reconcile these two?

Let me give you a very concrete example about this verification that I was telling you. Augury, the expert on machine and process health,  commissioned a study, the Forrester Total Economic Impact Study.

These guys went and found four customers, different customers, talked to them, and then went deep into the operation. And at the end, the ROI, as Ed said, it is the measure. So what they did in these four customers of Augury, they went and measured first benefits.

No? And based on the experience they already have, they went and dig in several places. And this is only for machine health. They were and digged on unplanned  downtime. What was the reduction?  Maintenance cost, energy reduction, bottlenecks that were solved with this inefficiencies that were solved, labor productivity, training, sustainability impacts, safety, quality impacts. So they went in this 10 or 12, each one measuring what you had before, what you have now, and they calculated for each one of these, the money benefits. Then they went into the costs.

So license fees, implementation costs, ongoing maintenance of the system, and they calculated the cost. Then the time, ROI is based on time, no? I get this amount per year in three years. In this case, they are taking three years and putting everything on net present value. 

Really, the results were impactful. Ed, do you want to share some of the results?

Ed Ballina (21:17)
The numbers are maybe staggering to some but Alvaro and I have been around this for a little while and folks these numbers are real. That’s the value of having an independent outfit go out and not give you feedback from your own echo chamber but sit outside and be objective. Say what is this doing? So here, how about a 310% ROI folks? Less than a six month payback.

And I can tell you these numbers are real because I’ve experienced personally these numbers in some trials I ran in my previous life. And we installed monitoring on 12 pieces of equipment. Within three months, we had saved about 24 hours of total plant downtime that paid for that installation in three months.  All of the downtime that’s avoided, right? The spend on R&M, the better you run, as Alvaro has said, The less energy you use, the less you consume resources, less goes out to the landfill. And there’s a lot here, right? $20 million net value over three years. But even the great work that was done by Forrester, they hint at the other piece of the iceberg the ones that’s under the water, okay, it’s employee morale. It is less product out of stocks, which brings you customer delight, right?

It’s better quality products, lower inventories as Alvaro said. How about retention? Alvaro and I talk ad nauseum about the labor shortage that we experience on our shop floor. How about if you get to retain your best mechanic that is getting ready to maybe move into a technical manager role because they’re doing something that’s actually fun and gives them control over their life.

In manufacturing, that’s one of the things we don’t always have, especially if you’re running a line that is inefficient and unpredictable. Folks, you don’t know when you’re going home. You walk into that plant at eight o’clock in the morning and you think you might leave by five or six. Maybe not today, Roger. We’ll see you in the morning.

Alvaro  Cuba (23:35)
Yeah. think also this way,  why this matters. We have been many, many years, especially in food, dealing with TPM and process improvements and equipment and great progress has been made, but with huge effort and time and well,

Once you have the right process and the right strategy, the right mentality, this is the time that you put AI. And then AI is going to accelerate all that. So what the Japanese tells you, and the line is running and stops every 10 minutes and should stop every four hours, five hours, and that stop because the machine stops, stops because you…decide to stop the machine to lubricate or do whatever. Then it becomes predictable. Then it becomes all the benefits. So finally, I would say, and we were talking the other day, we wish we would be now in the plants. Because now finally we have a way to make this continuous improvement in a way that doesn’t require huge amount of effort and time and years to get it done. 

Ed Ballina (25:13)
Yeah. And the paperwork, I mean, Alvaro, one of the things I hugely disliked about TPM is the amount of paperwork and administrative work it generates. And frankly, I had some TPM managers that I wasn’t happy with because they were really good paper hangers. I’d walk into their office and there would be charts and TPM forms and checklists and all that. But did I see you on the floor executing some of these fixes?

That to me is where the rubber meets the road. So anyway, I will get off my rant. So now I’m going to talk about why most projects fail. And these are not necessarily in ranked order, but 

Alvaro  Cuba (25:53)
or in the positive, no? Why they fail, so what are the lessons learned to make them successful?

Ed Ballina (25:55)
Yes, in the positive. Agreed. Go ahead. I, you know what, one of the first ones that I think we’ve all experienced and not the only one is pilot purgatory trap. Okay. So usually it goes like this. You get started, everybody’s excited. This is the new shiny toy in the facility, right? Or we’re going to do a trial. We got to do a pilot first. So we go in, we do a pilot. There’s excitement.  And then at some point in time, you know, there’s a new shiny object or you wind up you know, blowing up a line and now you got to figure out how to get that up and running. People lose focus and then gathering this data is not easy. That is the biggest challenge really is getting the information to prove the impact because the impact is there. So then people lose a little bit of interest and then before you know it. Okay, it’s doing great in this plant, but you know, how do we scale folks? This stuff, this is no longer proof of concept. It works.

Okay. And the longer you wait to execute, One, you lose momentum. Two, you’re losing a ton of cash. If you really believe half of these numbers and we can’t even give you a less than one year payback, I mean, I don’t know about you, but tell me so I can invest my money in that, because that’s certainly not what’s happening in the stock market. Although I can’t complain, it’s doing well recently. Anyway, purgatory trap is one. Secondly, don’t try to do everything, right? Don’t try to boil the ocean.

Alvaro and I’ve  talked about number of…times now that what’s really, really grabbing and holding on and delivering great results is agentic. It’s very, very focused  implementation of AI. And then  you got to start at the beginning. And the beginning is about your machine health. And that is the biggest impact right from the get-go. Some of these other improvements that Alvaro and I are talking about are systemic and accumulate over time, right?

You’re not going to have happy, cheery employees the moment you turn on that sensor, right? It’s going to take time for them to recognize that work is different and work is better and more predictable before you have them with you. But once you have them with you, they’re really unstoppable. So how do you take these and shift into the implementation mentality?

Alvaro  Cuba (28:19)
And a couple of things on what Ed was saying. Starting point is you need to understand. We talk about the KPIs. You need to understand where are you in each one of them, in each part of your plant, so you can decide where is the right way to start.

You need to think on projects and the implementations that give you the result. Let me give you an example. You choose one equipment in a line that is giving you a lot of headaches and you say, I’m going to solve this. I’m going to solve the problem. If you don’t take it the entire line and you only fix one piece or one equipment on the entire line, you will not get the results at the end of the line, which is what matters. No. So that’s for instance, an example. Don’t, don’t take one equipment here and other there, took an entire line and make it happen so you can demonstrate the results. Benchmark is another thing. Today there is, we were talking about ChatGPT. ChatGPT gives you huge amount of benchmark that you can just go in and use it  to understand where is better to start, how to start. We were talking about Forrester in these studies to demonstrate the benefit, but you can also use those studies to understand where and how to implement.

All this just to say Jump to it, jump small, then roll out, but go for it. Start crawling, then walk, then run, but start.

Ed Ballina (30:32)
Agreed. Sometimes you just need to put energy into the mix folks and just  do something right.  Just do something and start showing the real promise of this.  So you know I’ve talked about you know picking your bad actors getting that focus on the KPIs first and being very very prescriptive about here’s where I think I’m gonna get the savings and then you got to document this. I know nobody likes to do that folks I get it right.

And this is where ChatGPT, right? And AI can really help you to take the drudgery of trying to get this information out. Not to go too far, but I was out in a plant and I had to read all of these reports, right? To figure out what was wrong with a piece of equipment. Cause I don’t trust ChatGPT yet. I read them and I fed into ChatGPT and yes, I had to do a little polishing and a little more data addition, but my goodness, it probably saved me an hour and a half’s worth of time.

Right. And that’s what it’s all about. Use technology to make your time more efficient. So we’re kind of moving on now  to the wrap up.

Alvaro  Cuba (31:40)
Before you move just one small tidbit, we talk about CFOs. Just summarizing into all this, because CFOs are the ones that have the money, the ones that are at the end are going to report the benefits and are great allies if we convince them of the value. And for that, we talk about the business case. We talk about the study.

That can help. We talk about ROI of demonstration. With these things, you have the CFOs at your corner.

Ed Ballina (32:31)
You get their attention, that’s for sure. So as we think about this,  what’s the evolution, right? So the maintenance effectiveness files are pretty predictable evolution curve. You start with breakdown maintenance and then you decide, okay, that’s not good. I’m going to do preventive. So time-based maintenance, right? And that’s good, but doesn’t get all your failures and you waste a lot of money. Then you start predictive, which is where we are, where you’re servicing equipment as needed because it’s telling you and then eventually you design out. Right now we have to focus on going from reactive to predictive to prescriptive, right? And you know to piggyback on one of Alvaro’s terms, right? It’s about control. And I think I’ve said this a couple of times in this podcast, but who controls your line? Do you run your line or does your line run you? It’s a simple question, right?

Trust me, our lives are all much better if you run the line instead of letting the line choose when it’s going to break down. Usually in the middle of your kids little league game on a Saturday and you got to run it. And AI doesn’t just stand still, it learns, right? That machine learning is just amazing. The power, right, of what that can develop to in several years. When I first started working with Augury, I think we used to say we had 20,000 machines, right, in you know, in the AI cloud, right? We must have over 100,000, right? That’s the value of compounding, and it gets smarter and smarter every time. Each failure that you prevent makes a system smarter. So, Alvaro, as you would say, let’s go for it.

Alvaro  Cuba (34:17)
Yeah, at the end, we want just to finish this  episode with a call to action. No,  as Ed said, it’s proven. So don’t expect perfect, just start going.  The competitors are moving and others are moving. I was sharing with Ed  before starting the podcast.

I was reading the news about plumbers, not one, but hundreds of plumbers using ChatGPT when they go to work because they can invoice or work orders super fast. When they go and find a problem, they just take a photo and ChatGPT tells them what’s the recommendations. If they need spare parts, they ask, give you the cheapest in the best place. So there is a lot of things. And if it’s beneficial for them in their trade, in their day to day, imagine the benefits that you can get out of it in the plant. So ROI, you can use third parties or you can read the studies. In this case, the Forrester study’s available. You can read it, you can apply it yourself with a finance guy. Is cheap to implement. It’s different than our times that had a lot of capital and a lot of approvals. In these cases, services as a software. So you pay just by month. Ed was mentioning six months ROI. So you invest from your maintenance budget.

And in six months, you get paid before the end of the year. So, that’s little bit of hope. We hope this is helpful. Sorry, you were going to say.

Ed Ballina (36:18)
Yes, a bit of a thing. No, I was just going to say, and you did mention this happened in Milwaukee, right? Makes sense because, you know, Milwaukee is known for one thing and that’s brewing beer. So here again, here’s a connection between AI and some adult beverages.  So yes. ⁓

Alvaro  Cuba (36:31)
one of the places. finishing to finish at the bar with a with a nice beer. Perfect. So thank you friends for joining. Please ⁓ share with us your comments, your ideas. We had very good inputs and comments in the last two episodes. So please keep doing that. And Subscribe, like us, leave us reviews. More important, let your ⁓ pals know and join, join the podcast and let’s get the conversation going.

Ed Ballina (37:28)
Awesome. Hey, join the conversation. Let’s keep it going. ⁓ You can email us at mmu@augury.com We also have a link down in the notes. ⁓ So you can look at the Forrester Economic Impact report. It is really thorough. I highly recommend you spend some time or just dump into ChatGPT and read the 30-second Reader’s Digest version. So anyway, with that, ⁓ great being with you again and see you next time at MMU.

Alvaro  Cuba (37:58)
See you guys, bye.