Why Manufacturing's Most Valuable Data Isn't in Any System — with Anand Gnanamoorthy of Ingersoll Rand

[Music] Welcome everyone to the Emerge AI and Business Podcast. Today's guest is Anand Nanamurthy, Director of Corporate Strategy and AI at Ingersoll Rand. Anand walks us through why a significant share of manufacturing knowledge still lives inside experienced workers and why capturing it has become urgent as long to need employees, retire and operations shift toward AI-enabled ways of working. He separates the three data layers manufacturers already have, structured operational data, decades of unstructured archives and the tribal knowledge held by frontline staff and explained why those unstructured archives sitting in employee drives and folders are the biggest untapped source of value in most companies overlook. Today's episode is sponsored by POCA. Please note the opinions and views shared by an aunt in this episode are his own and do not reflect those of Ingersoll Rand or its leadership. Position your brand alongside the Fortune 500 leaders defining the Enterprise AI roadmap. For the opportunity to showcase your solution to the executives currently funding and scaling global initiatives partner with Emerge to reach the decision makers holding the strategic mandate. Secure your partnership at go.emerge.com/partner. That's g-o.em-e-rj.com/pa-t-n-e-r. Now the conversation with Anand. Anand, thank you for joining me for an interesting discussion today. Thank you. Happy to be here with you. Absolutely. I'm sure you'll agree with me that a significant share of manufacturing knowledge does not live in systems anymore. It lives in people. In the heads of experienced workers who know which machine runs hot, then which process needs a workaround which shortcut is safe, which one is not. And as those workers retire and operations push towards AI-enabled ways of working, the question of how to capture and transfer the knowledge is becoming one that needs some urgency in the industry. And that's why I want this to start today. So from your work in manufacturing strategy and AI adoption, where do you see the biggest challenge when companies try to prepare frontline teams for more digital ways of working? Sure. I think it's a multi-dimensional problem. You cannot, you know, classiva. Hey, this is one single problem and you can answer it pretty simple way. You have to look at several different, often conflicting set of ideas. For example, one dimension, obviously, as you mentioned, right, the knowledge resides in the people. If you look at our current systems, the current systems have data in them, the knowledge resides in the people, and it's typically the managers who would be making the decision. So that's how our current systems are defined. When you're moving from our current digital systems to AI systems, wherein, you know, AI is not that great with data, but it's very good with insights and, you know, and also who has to make the decision. So if you look at it as a data insights and a decision problem, you can see it in a in a different angle. And if you look at all the three dimensions, so for example, if you take data, currently data resides in so many different places. And so companies have this extensive data stack to collect data, collect data into a single space, derive insights from that, share it within their network to make the decision. So it has such a high levels of, you know, automation involved in that. There are so many different layers in that. Now, if you're going from that system to another system with AI, which can look large volumes of data, can deliver, can develop insights immediately. And then you can make decisions based on top of that, you know, the workflow needs to be adopted for that. What I mean by that is currently decisions and insights are developed by the humans. Now, when people are moving towards AI, they are kind of confused between, hey, should I let all, should I let the AI make the decision? Which of those decisions should I keep? Which of those decisions should AI make? So that is the biggest challenge right now that I see in the market. And if you're not able to define that properly, that's where the pilots fail. So a lot of people are doing pilots. They would do one example is, you know, the simplest example of AI being use right now is email automation. Where you have an email coming in, the AI system would automatically develop a response to it. So what it is doing is it is looking at data based on the data it is developing the insight to, hey, this is what the user is asking for as a response to that email. That's that's pretty simple, but should you send that should AI send the email out or should an human have it review the data or review the email draft and send it out. So that's where the biggest challenge right now is, you know, which are decisions that AI can take, which are decisions that humans should take. And many a time we have seen several examples where it was done poorly and, and you know, people have got it into issues. For example, companies use AI for their chatbots and we have at several examples where an user would be asking it questions. It is it is promising them things it should not saying, hey, I will give you a discount right. I will offer you this car for a dollar. Those are not decisions that AI should be making. AI should be just giving it giving information on insights, but it should be the human in the loop that needs to decide. So to me, what I say is that is the challenge that I see in not just in manufacturing space, but in general deciding where humans, where AI should stop and where humans should be taking it. And when we say knowledge still lives in people, how big of a risk is that actually a company is really a way of what they stand to lose if an employee with lots of knowledge and experience is going to retire or something like that. That is true. The large portion of knowledge stays within the user. There is a risk towards capturing all those tribal knowledge from the employees and experts and capturing that via AI systems. Yes, that's a big challenge, but it's the challenge not just exist at that level, but also at even at operational level. What I mean by that is if you look at any manufacturing organization, they do have, I would put in three buckets of data. One bucket is the structured data, which is all your operational data, which is coming from your shop floor, coming from your finance systems, even coming from your HR system. How many employees you have, how many workers did they work every week? What is their salary? You have this structured data, which is nice and visible across the organization. The biggest source of data that is kind of untapped is unstructured data. So your emails, your phone calls, your transcripts, all the files people have developed over 40 or 50 years. If you are in a manufacturing organization that has been there for at least for 50 years, you will have data in so many different versions and so many different storage formats. So 50 years back, people used to write letters. There are a bunch of letters being stored. When the first way of digitalization came in, people stored in word documents and files and those are in your floppy disk and sometimes those are in hard drives. Then as you move more and more online, there are huge amount of online data points. So there's a huge amount of unstructured data in organization that in a way kind of mimics what the user has learned over his let's say 30 years of career. So yes, huge volume of document that he has while it's not a one is to one, but it does have a good collection of knowledge that that the person has one of the major challenges when that person leaves or retires that huge volume of data which is pretty easily accessible to the company to a digital system is kind of left or you know kind of deleted because it's a huge volume of data going through that current volume of data is impossible. So at the biggest the biggest ban for bug or the biggest ROI for an organization would be used to look into that particular set of data, how do they how do they capture that and one big challenges, you might know or you have experience yourself, you'll have the same version of document, you know 10 different versions of the same document. So yes, there is a lot of duplicate data, but how do you how do you reduce the duplicate data? How do you find out what is the most relevant data? And once you have that, you have systems in place and AI is pretty good in that in the sense it can understand and find out patterns in the data. So if you let AI know that hey, this is a messy data. I don't know which is the latest version, look into this and find the latest version and delete unnecessary data. It is very good in doing that. Unfortunately, most of the companies are not doing that in the sense they are using manpower to clean the data and then feed a cleaned data to AI. AI is actually very good in handling messy data. And the third part is the tribal knowledge that sits within the user. There are several different challenges within capturing that one. You know, it's one is a psychological challenge in the sense, if you go and ask any employee, we want to capture your tribal knowledge, your expertise into a system. They're going to be very resistant in the sense, hey, are you going to get, are you going to fire me, right? In the sense there's a natural tendency of employees to resist those kinds of things. I have seen examples wherein when you look at manufacturing industry wherein it's more related to physical movements or some companies use vision guidance. systems or robotics to take close to the worker and see how he is doing it and capturing that. But the challenges in manufacturing industry, any form of recording and these kinds of intrusive technologies are highly, highly regulated. You know, you also have the issue of labor union. So not many people do that. So the last portion of where the tribal knowledge sits actually within the user, it's going to be, it's going to take up some time for us to capture that. But the easiest one to capture is the unstructured data that's within within the employees, laptop within the employees, you know, personal drive, you know, or they're papers and folders. That's where the biggest set of user knowledge is in and that's where I feel companies are missing out and not capturing them really well. And I guess when we look at frontline workers, they might not be the type of workers that work with computers or systems or emails or word files. So getting them to take their paper-based SOPs and digitize that is also going to be crucial if we think about factory workers. Like I said, the guys that understand the machines don't know when something is about to break or when something is about to be overloaded. Getting that to digitize is going to be a new challenge just because people don't, maybe they might feel threatened or intimidated if someone had to ask them, let's digitize everything that you have, your skills, your mind. And with that, the technology does it allow for us to digitize paper-based SOPs as the challenge mostly technical, getting the right tools in place or is it to actually get the workers to use them once they become available? My personal experience has been is that the country per se is not the challenge, it's more related to, you know, I would say two aspects of it. One aspect is how do you defining your workflow in the sense how is it being currently done? There is the, if you go into any manufacturing plant, there is a textbook definition of how a work needs to be done. And in reality something else gets done because hey, you need to get work done, right? You cannot always follow, follow the standard procedures because it's supposed to be the industry-based practices, but not always it works. So there's a deviation between those two things. So how do you capture that workflow? That's number one. Number two is, as you pointed out, the element of worker pushing back, hey, those data being collected. So if you look at the first one, work study is something that's always done in manufacturing industry and depending on how good operation, how good that particular factory is being run, you know, they do have standard workflows, they do have, you know, if they are doing, you know, any kind of certification, you know, ISO certification, things like that, they need to do a periodic testing of, hey, this is our workflow. This is how it needs to be done. How good are we in achieving our standard, you know, standard workflows that we have designed, right? So they do periodic studies. So there is, there is some amount of data being captured on how the work is being done. It may not be the perfect way, but it is a good proxy of how it is being done. So all your several different forms, all your, all your data on work in progress, hey, how does the every time a material moves from one workstation to another workstation, it's some form of data gets captured so that, you know, you know, exactly where, where the, where the part is or how many at each workstation, you would need a different set of materials. So when it moves from workstation, one to workstation, two, a different set of materials are removed from the ERP system or from their store. So you have proxy for your workflow automation. So the more you understand, hey, this is how it is being done. And you can kind of digitize those process. Now what, what I have seen the biggest challenges, people believe they need to provide AI the perfect data. And once you provide the perfect data, you can answer a question on top of that. But AI really loves messy data. You can, you can give a rough data. You tell AI that hey, this is not the perfect data. This is a messy data. There are problems in the data. Once you let it know that hey, this is my workflow. Here is, you know, here is how the inventory is drawn every single time. Here is the, here is, you know, the work in progress, staying at each station, the time, you know, if you have the timestamps at each one of them, you are able to develop a map of how the workflows happening. So that's, that's easy to capture. That's the easiest option. Now, when it comes to the other end, when you have to capture the data directly from the employees themselves, that's where you will have challenge. And there are, there are some companies that have done wherein they use co-bots wherein you have automated robots that that work along with the humans. And it is, you know, humans using the co-bots help either as their assistant or them doing one portion of the work and then leaving the arrest of the work to the co-bots. So some kind of data gets captured in the co-bots. So that might be a one. But again, these are, this portion is extremely costly. Now, you have to have a, you know, any time you get into robotics, obviously, safety is the, it's the big one. So whenever safety is involved, the cost of investment is going to be pretty high. So investing a large amount of money to capture that is going to be really tough. But if you, if you can capture the first portion of it, which is how was the current workflow being done there? You can, you know, you can, you can generate a lot of data points. You can generate a lot of insights from that. And that would help you in, you know, a lot of your organization AI calls. And make sense. And like I said, with anything robotics, the cost is going to go up a lot. And also if you have a big factory with hundreds of workers, how many robots will you need? Or how long is it going to take to capture everything? So if this actually realizes and, and we are able to, to capture the tribal knowledge, when you picture a manufacturing operation where AI is genuinely helping frontline workers, what does that look like in practice? What changes do you see happening in the industry? If that is actually being implemented and being used correctly? For frontline workers, there are the biggest help. Again, I could, I could see that in a couple of different ways. One of them would be is, if you go to any factory today, the biggest challenge would be is, do you have the right set of parts, the right set of time coming in? So the biggest challenge that most of the manufacturing organization is having is our own schedule, work scheduling. And that is based on which components are coming in when so providing that visibility helps you reduce the complexity in, in, in work scheduling and that kind of makes it easier for, you know, not just, not just the frontline employees, but also there, you know, the supervises, the managers there to make sure the operations run smoothly. So that, and that's not just looking at the frontline, but it's looking at data across the organization all the way from your supply chain to, you know, your finances, your engineering data, all of that. So it's, it's going to be a little more complex than that. If you were to purely look at frontline workers on AI related ones, it, again, it depends on, between a new employee and an existing employee and an existing employee has larger mode of tribal knowledge and he might not need as much help from AI, but if you are training a new employee and if it's, again, if it's a new product or a new system, they need help with accessing data. That's where it can, it can help one, one example I could think of is, if you look at a complex in, in, in aircraft manufacturing, for example, when you're wiring the aircraft, it's a huge complex beast and it doesn't matter how many years of experience you have every aircraft is different. So a frontline worker would have to access huge amount of drawings and knowledge to find out how to do that. So they'd be moving back and forth referring to drawings, looking at which colors, you know, which wire needs to be connected to where any small error has such a big impact there. So it's, it's a pretty stressful job. So in those applications, AI can be really useful, wherein rather than providing them the huge volume of data, it can just provide them a, this instructional set, which is, which is, you know, kind of here to here rather than dumping all the knowledge and the worker has to pass through individual piece of data for his particular work step. The AI is able to pass that and provide him only that particular data point so that he's focused only on doing that. Not just that AI given that, you know, as, as it can learn over a period of time, it can say, hey, it knows people make mistakes at this particular point of time. So it can make sure our ask the employee to double check, hey, have you, have you connected here? This is this places where they make most mistakes or you can have a camera or a vision system attached to it and to make sure that every instruction that's being provided to the worker is being done. And if it's not and if you, if the system is directing, hey, there's a mistake there, it can alert the worker immediately saying, hey, go fix that right now, which is much easier rather than doing it a QA, which is, you know, too much down the line when all the wires have been done and it's going to be much more costly. So those are areas wherein it can directly help the frontline workers making, making, you know, removing the three different components. Right. So you have the data, you have the insights and you have the decision making. So if AI can take some of that, which is it is focused on data, it is focused on insights and the human is only making position, only doing the decision making part. So that kind of eases the mental stress on him and so kind of makes his life easier. Absolutely. So capturing that tribal knowledge is also going to help us to prevent past mistakes from happening again because like you said, we'll be able to identify. The AI will be able to identify where these mistakes normally happen and might not come naturally for us when we compile the ASOP but when you're actually on the frontline in the manufacturing plant, you have that experience of, "Ah, this is normally where someone presses the wrong button or they use the wrong part or that so that absolutely makes sense." And as we see this being implemented, how do we get our staff members on board with a new AI, new workflows, new integration, if they've seen so many things come and go before and they still just trust their own instincts. How do we get them to adopt to, I think this is now what we're going to use. I think this is more gets into change management and psychology and organizational behavior. Anytime there's a new technology that is obviously a pushback, if you look at the way previous technology adoptions have been, the same thing was said about computers. If you look at both industry magazines or research articles from 30 years back, the same challenges that are being posed about AI were the same thing that we're talked about in 30 years back when computers came first. It's like computers were slow, they were clunky and people like, "Hey, computer takes so much time, I could do faster." But today computers are so much faster, nobody's going to say, "Hey, I can do faster than a computer." Again, look back 40 years ago when the first PC came in, they were extremely slow on people who were saying, "Hey, why are you using this? Not sure how many of the listeners or yourself would have experienced that, but I have experience in that space where people said, "Hey, this is slow and clunky. Let manual process so much faster." As things go by, it gets faster and faster, the resistance to change decreases. Every time we have seen adoption of a new technology, it's been a kind of ebony flows in the sense. The promise of a technology is a lot more, so there's a huge hype around our technology. Companies invest a lot of money and there's a lot of expectations. Then there is the disillusionment where people find, "No, this is not as big as it's promised but then people find out, "This is the application where this particular technology is suited for, once you know the space where they are particularly suited, people start using it." There is going to be a lot of pushback against that technology, but what I've personally seen is in every organization, there is this build curve, so there are innovators, there are earlier opt-ers who want to use this technology and companies need to focus on those skins of employees and provide the tools necessary for them to experiment that, find out what are applications that this particular technology is well suited for, and once people start, the others are in the organization, see the innovators and the earlier opt-ers using a particular technology, being productive, able to achieve their goals, are able to use technology to reduce their work, others in the organization will start implementing it. So that's how I have personally seen, and that is kind of the best practices wherein, rather than forcing a technology wherein you are going to get a lot of pushback, you find pockets of enthusiastic employees who are willing to adopt the technology, have those as your champions and use them as use cases to expand your adoption of this technology. And that sounds like a very practical way to start, without overthinking it, we're getting into interesting things there. Our conversation is reaching a conclusion, and I wish we could speak about this for hours and hours, but before we get into the conclusion, I'm thinking about when we're tasting, who are we tasting with? Are we going to taste this with operators rather than leadership? Why does it matter who we taste this with and what will happen if we taste it with the wrong group of people? I think it's, I wouldn't say it's a wrong group of people, I would say it's making sure you're having the right set of application for a different, for each set of people. So leaders will have a different set of AI application that can help them. It's not going to be one size fits all. So for frontline workers, you know, understanding their pain points and how AI can help their lives better. If we were to reorient our outcomes, our objective for implementation of how do I make this particular work or employee more productive? How do I reduce his stress level? How do I increase, you know, his productivity level? So if you can, if you can turn around from turn it around to anchoring the AI use case to a particular worker, then it becomes much more easier. What I currently see is mostly it's anchored around particular application or a particular, you know, particular workflow in the sense, hey, how do I make this process faster now? If you're going to ask that you are, you are taking the human out of that workflow and you're trying purely to optimize the process and in that, in that activity, a human might get the needs and the wants of the worker is not prioritized. And obviously they are going to be pushing back. But if you were to reorient it towards a, I know this particular workstation is handled by this person, how do I make him more effective? Now, if you put him as the sole focus, it's going to be easier for everybody involved. The worker is going to be much more happier because you are providing him tools and technology to make his work easier, not the other way around. So that would be my solution. Absolutely. For our conclusion today, we want you to predict the future. I don't know if that is part of your skill set. But for the manufacturing leaders in our audience today that want to start moving in that direction, what should they be focusing on in the coming months? My, so this is a larger topic, but at least in a smaller way, AI is transforming every single day. So what we saw three months before, six months before, it's not AI today. We started with AI chatbot. When we went to AI agents, now we have AI workflows. We will soon have another AI technology. So the biggest challenge right now is AI technology is changing so fast. The biggest thing would be it's not to get locked down to one particular technology or one particular solution, but rather be open to using the best of breed. I'm assuming that any AI project that you are doing today, you will not be doing two years from now. Going with that mindset, completely understand that, hey, this is a pilot project, even though you might be rolling out to the entire enterprise, be, be well advised that two years from now, the market landscape would be completely different. So you would need to reengineer that. So if you have that as your goal and always understand that, hey, there is a definite end date to this particular project. You are much more agile to capitalize on whatever new technology is coming in. The biggest challenge right now I see is people are getting locked down to the current AI systems or current AI workflows that is going to hamper them when a next next way of AI technology is coming in. So at least for the short term, it's about being nimble at the same time. If you're on the sidelines and say, hey, let me, let me wait till the AI system ensures that everybody is using it by the time you have created so much technical debt that you're never going to catch up on and you are going to be one of those companies that is continuously trying to catch up your competition that has been in the forefront of AI is able to leap forward and you are going to get acquired by that company because that company has knowledge and expertise of using AI and they are going to be able to immediately acquire your company and bring it up to speed. If you are not willing to take that AI pilots with the complete assumption that it will fail, it will be changed in two years. You are setting up yourself for failure. And you've given me so much to go and think about after our conversation today, I've made notes during our conversation and I think the three things that I'll be thinking about mostly is that there is a way that we can prevent past mistakes by capturing tribal knowledge in AI. The next thing that you really left me with was that we need to focus on that innovative employee, the one that will embrace the changes and development and see the value in it. And then what you just said is use AI as, see AI as permanently in pilot mode because the changes are just so intense and so quick that you should be viewing everything as pilots even after you've ruled out just because of the changes that's happening. Thank you so much for your time today. I've really enjoyed this discussion and I am excited to hear what our audience has to say about this. Thank you. Rapping out today's episode is time for our three key takeaways from our conversation with Anand. First, the largest untapped source of manufacturing knowledge sits in decades of unstructured data, already held across employee drives, fullers and archives. And AI is uniquely capable of making sense of that messy duplicate heavy material if leaders stop trying to clean it all before it gets there. Second, adoption accelerates when AI is anchored to the worker rather than the process. Identify the innovators and early adopters, make their jobs easier first and let them become the internal champions the race of the organization follows. Finally, treat every AI project as permanently in pilot mode. The company is prilling a head or the ones moving now with the agility to evolve as a technology does, not the ones waiting on the sidelines and building technical data they can never catch up on. For the opportunity to showcase your solution to the executives currently funding and scaling global initiatives, partner with the merge to reach the decision makers holding the strategic mandate. Secure your partnership at go.emerge.com/partnam. That's go.embrj.com/partnER. For further executive level analysis and to join our network of leaders delivering workflow impact with AI, visit emerge.com. On behalf of the team at emerge, we'll see you on the next episode. [Music]