Tom Raftery: 0:00

And the biggest resilience mistake that leaders repeat?

Asim: 0:03

They still believe manual ways can work. They don't realise that the maintenance has not kept up with the assets. Assets have become more digitised. Maintenance has not. It's manual. That is the biggest mistake.

Tom Raftery: 0:32

We spent years digitising warehouses, fulfilment centres, and supply chains. But if the maintenance systems keeping those assets running are still manual, that creates a very different kind of resilience risk, one sitting inside the four walls. Good morning, good afternoon, or good evening, whereever you are in the world. Welcome to episode 138 of Resilient Supply Chain Stories and strategies that Keep business moving. I'm your host, Tom Raftery. My guest today is Asim Akram, CEO of MultiSensor AI, and we're looking at what happens when automation moves faster than maintenance. Conveyors, motors, chillers, electrical panels, the physical infrastructure that keeps high throughput operations alive. And there's a useful twist here. Degradation happens gradually, but failure happens in one shot. For supply chain and operations leaders, this is really about recognising where resilience risk has moved, and why predictive maintenance is becoming a business continuity issue, not simply an engineering one. Let's get into it. Asim, welcome to the podcast. Would you like to introduce yourself?

Asim: 1:47

Sure. Thank you so much, Tom for, having me on this podcast. Really excited. I am currently serving as a CEO of a company called MultiSensor AI, out of Atlanta. Happy to go more, or would you like me to do a deeper introduction of the company as well?

Tom Raftery: 2:04

Yeah, I mean, just tell us a little more about MultiSensor AI.

Asim: 2:09

Absolutely. Absolutely. Here at MultiSensor, what we are focused on is essentially, what we call providing the asset intelligence layer for our customers. Now, what does that mean, right? We are monitoring the critical assets of our customers. These critical assets are enabling the production and the operations of those customers, right? So now you ask, okay, why are you monitoring, why are the OEMs or others not monitoring? Well, this is where our multi-sensor name comes into the picture. We have multiple modalities, right? Whether it's your infrared, whether it's your acoustic or vibrational. We enable our customers through different type of sensors that take that information and brings it into the AI platform.

Tom Raftery: 3:02

Okay.

Asim: 3:02

That AI platform then gives you the insights about the behaviour of those assets. While the others are taking a snapshot at a time, we're providing the continuous monitoring of those assets. So you can look at the degradation, or the behaviour of that asset over time and take proactive actions, right? So we help you with that as well, and that is very relevant in the topic today. That will talk more, right? The resiliency which is enabled through the visibility is what we provide in this case there. So essentially it's all about preventing unplanned downtime.

Tom Raftery: 3:40

So I guess two things to ask out of that. The first is, where are your customers? What are they doing? Is it manufacturing? Is it warehouse? Is it something else entirely? And then also, I mean, you mentioned resilience and we often think when we think of resilience, we think of geopolitical shocks to be a little bit topical or supplier disruption. But your argument is that resilience failures increasingly start inside the building itself. Is that fair?

Asim: 4:07

That's a very good observation. So let me first answer the first question, which isn't good. We focus on e-commerce logistics. Airports and grocery stores, which have warehouses and fulfilment centres. They share the same critical assets across the board. If you look at the warehouses, you've got the conveyor belts. If you look at the fulfilment centres, they have the conveyor belts, the motors, the air handling units, right? The chillers, right? You have the electrical panel monitoring. These are all what I call the high throughput checkpoints through which a lot happens. And any degradation, Tom, in this case can lead to either, obviously shutting down of the operations in certain cases. Think of the electrical panel. You have a degradation of the connections. It can lead to fire. Fire can lead to a lot of fatalities, right? So we don't want that. But also the, there's another aspect of it, which then takes you to the second question. If Tom was to order something today, he's so used to getting that stuff in 24 hours, that he's going to get very upset if it takes two or three days. Look at the patience that we have now, right? What is enabling that? That is that warehouse and the operations. Historically the supply chains that we talk about, they are focused on what the inventory. They're talking about transportation, right? The capacity of the transportation. Ports, labour, planning. That was always the case, right? But the modern supply chains, they're no longer constrained by these constraints, right? What is that, that they're constrained by? Think of the warehouse. It's getting more and more digitised. The more digital transformation that people have gone through. If you go back few years, there was a whole wave of digital transformation. Well, guess what that did? It put a lot of automation in there that somebody now has to make sure that that operational infrastructure layer, the conveyors, the sorters, the motors, all of those are getting monitored. In this case there. So the reliability and the visibility of those assets is very critical. Some of our customers that we serve are global customers, in this case there. They have centres across the world and they cannot afford to have their, downtimes, right? If the, if you look at the warehouse downtimes, they range anywhere between $5,000 to almost a hundred thousand dollars per hour, actually. That's just their cost. Think about how Tom might be getting upset on the other hand, and he's so easily he can transfer or transition to some other customer. So if you take a lot of those things together and the reputational impact, the resiliency of the supply chain, the topic of this podcast has become so relevant in, in today's day and age.

Tom Raftery: 7:14

There's a useful shift in perspective there. We normally look for resilience risk outside the four walls, suppliers, ports, geopolitics, but automation is creating another risk layer inside the facility itself. So are automated operations becoming more fragile, not less.

Asim: 7:34

I wouldn't call fragile, but they're now more prone to some of these shocks that can happen. Now let me explain why. There are certain angles that kick into the picture, right? The more automation you do which is obviously the case, the more you have to now make sure you keep up with the maintenance of those assets that are helping you automate. So the fragility does come in, in a sense that you've got a bunch of assets, but you don't have a consolidated or a continuous way of looking at them. So what do we do? We do manual inspections that happen every time every six months, for example, every three months, right? Well, the problem is, as we all know, this happens in our lives all the time. The air conditioning unit is only gonna go off when it's peak heat. It doesn't go off anytime else right? It, it waits for that perfect time when you have the peak things happening, then it says, I'm gonna give up on you. Same thing with these critical assets. You've got these peak days, and that's exactly when it decides to go off. But that's where we come in to remove that fragility that you just talked about. But let's, let's take that and expand that a little bit more. There's another wave. It's a very interesting question you asked because the fragility can happen not only from inside, but from the outside as well actually. So let's talk about that for a second. a big wave that's happening, which is the data centres. So we talk about digitising the warehouses, more automation comes in. Obviously there's a data element that you have to look at. Well, that's where is that data going? In the data centres

Tom Raftery: 9:20

Hmm.

Asim: 9:20

at this point, right? That was the always the case. Now, that was never the problem, Tom, right? Going to the data centres, they were pretty good. When the AI came in, that changed the equation completely. And I was having a conversation with another news agency and we were talking about the fact that the data centres, you don't realise how often they go down because of the infrastructure that's outside the data centre, the pumps, the cooling equipment, electrical panels, when they go down, guess who gets impacted? The warehouses, That's the top. So it has impact across the supply chain. Lot of your stuff that you're running, your inventory plans, right? Your software that's running all the supply and demand and tracking of the logistics. When that goes down, you're out of sync in your physical inventory and you what's on your computer, and now you're not able to figure out where your product is, and that's where the fragility comes in. So by themselves, they're very strong because they have done the automation, they have all the latest equipment, but the, the maintenance and understanding the intelligence and having the intelligence about those assets is not there as yet. And I think that's where the multi-sensor comes in, the continuous behaviour with that. There's a third element Tom that's getting more and more prevalent nowadays, which is think of all the different assets that exist at this point. Different types of those. If there is a certain level of automation when it comes to understanding the behaviour of these, the way the customers are today doing that, Tom, is they have one set of data for one asset. You have another set of data and monitoring mechanism for another set of assets. You have mechanical, you have electrical. Imagine having a massive amount of data coming at you from different assets. How do you make sense of that? So now the fragility is not just the assets and the things that we just talked about, but also with the analysis of that data. You have the raw data, you have the knowledge now from coming from there. How do you use therefore, decision making becomes even more critical from the resiliency point of view. And that's where again, we provide that tools and the AI platform to help our customers to be able to look at this holistic data in one what we call a pane of glass, right? And then be able to make sense out of that one there. Hopefully that answers the question that we talked about the fragility. It was a good question though.

Tom Raftery: 12:06

Sure, sure. But isn't that what you know, traditional preventative maintenance was supposed to solve?

Asim: 12:12

It was, but it hasn't kept up. On one hand, you're absolutely right. I mean, the preventive maintenance. Well, let's take a very simple example. Let's take a very simple example. Your car tells you it's time for oil change. That's when you go and do the preventive maintenance. It does not tell you that over time, the other components of your car are degrading. You only show up when it says, Hey, it's time. Maybe the issue with that happened way before, right? So now it's come to our world, which is the supply chain, the warehouses. How do we do that today? I have someone who is manual of being a person, shows up, he's got a handheld device. He walks through massive amount of warehouses, right? In this case there. So what's the issue with that? Number one, labour shortages. Believe it or not when I talk to my customers, their number one issue is getting the right people. It's not just, you can pick any person and say, go start taking scans of these assets. That doesn't help, right? In this case there. So, you've gotta make sure you have the right resources, which are not easily available. So that's one. The second is you now have to go and manually scan at a certain point in time. Well, the, the degradation that happens, it happens over time, but the failure happens in one shot. So you need to be able to look at that behaviour continuously, not a snap in time, That's how we have not kept up with it. So the traditional way of, let me just take a look at it every three to six months has not worked out clearly. Hasn't kept up with the demands. But then if you look at the volume of the information, if the volume of goods that are passing through, they're not slowing down either. They're still increasing. The demand is there, right? That's never where the issue here. So what happens with the assets? They're, they're getting used 24 by seven. Some of our customers Tom, in the automotive industry, their per hour downtime is more than millions. More than millions because they cannot afford to have the line shut down. So imagine somebody shows up in January, everything looks perfect. He shows up in June, everything looks perfect. But what does not know is yes that I guess threshold is there, but is degrading over time. You're still in that threshold. That's all good, but the degradation is happening. You don't get to see that. But if I can tell that to my customer, he can make sure he can be preemptive and can take, preemptive actions.'cause then they have planned down times. They can go and fix that. Now there's no downtime, unplanned downtime, right? So, no, the, the answer is that the, traditional ways have not kept up. It's manual. It's the handheld devices. It is the snapshots are taken every three or six months, just not enough in today's world. The digital transformation, digitised assets as well as the, the demand that we see.

Tom Raftery: 15:54

And I gotta think, if you're monitoring all these different devices made by all these different manufacturers in all these different locations, different temperatures, different humidity, different whatever, different longitude and latitude, et cetera, et cetera, et cetera, et cetera, et cetera, how do you set those thresholds? Maybe it's vibration, maybe it's temperature, maybe it's whatever. How do you set those thresholds for that many different types of devices in those many different locations and settings and know when something is going to raise an issue?

Asim: 16:35

What are the true thresholds? Because the infra red in a threshold is the heat, I guess the thermal activity that we talk about. The thresholds are different than vibrational, but more importantly, not every asset needs multiple type of devices. Certain cases you need more of vibrational sensors. In certain cases you need IR. Then the next step is, okay, what is the business tolerance? So it's always the business that drives, the thresholds. Because what are we talking about here is business continuity. At the end of the day, it is the business continuity. So we have folks who show up who understand the business continuity and the thresholds for that. They get translated into the thresholds that we can set up in our product. It's very configurable based on the customer needs. And then that allows us to then actually provide the, the alerts provide the warnings, the early warning indicators are all set up in our product that enables the customers to learn way in advance.

Tom Raftery: 17:39

Okay. And how much earlier can problems realistically be detected?

Asim: 17:45

Well, when we talk about problems, they can be obviously vibrational related, right? So as I, I always say this, I can see things faster than you can realistically, because I'm looking at the thermal footprint. I can feel faster than you can because I have the vibrational and I can actually hear stuff faster than you can because I have the acoustic sensors.

Tom Raftery: 18:09

And when deploying these systems, what assumptions can turn out to be wrong?

Asim: 18:16

One of the assumption obviously is what we call the field of view, right? A region of interest, which is synonymous to ROI, but that is a region of interest. We are just making sure that we have the right assets covered, and sometimes those assumptions could be wrong, meaning we have captured the wrong area as such, right? And then obviously we fix that. That's, that's not an issue. We have the ability to go back, work with the customer. As I talked about. We have the commissioning plan, which allows us to go fix the, the things. So that's not an issue there. Rarely it has happened ever that we have used the wrong sensor because we know in certain cases the threshold for the customers to be able to get the information faster is based on certain sensors. The most important part is making sure that we have the right sensors. But I would say typically is the adjustments, the assumptions that we made in the beginning. So we do site audits, right? So before we even get to the customer and give a solution, we go to the site, we make an audit, and we make some assumptions that we have six different sensors that we need. Well, it turns out in most cases that they are correct, but some cases the customer may say, I need a little bit more aggressive monitoring of these. So whether you call it wrong assumption, or whether you call it change in scope, or whether you call it, Hey, I need more sensors, those kind of scenarios happen, Where you have certain assumptions made as part of the site audit that get altered based on when you deploy them, right? The reality then kicks in there. And I would say that's where majority of the changes happen. But the assumptions of the, the sensors that we need for certain equipment I think we've been pretty good with our customers on that.

Tom Raftery: 20:01

Up to this point we've been talking about whether the system can see a failure coming, but detection only matters if people trust the signal enough to act on it. And do the operations teams trust the AI recommendations immediately or is there still cultural resistance?

Asim: 20:19

I, I think that's a fantastic question. I, I think there are pockets, Tom, of, of customers. There are customers that are very much ahead of the curve. They believe the AI has it right? And then they take our AI recommendations. In that scenario, obviously the recommendations over time get better and better. As we get more data, the recommendations get better, the confidence level goes up. But there are certain customers who are okay with that, that, hey, I'm willing to do that. Then there are other customers that prefer what we call human in the loop, which is, I know AI is telling me that, but I still want a human to double check that, In this case there. So there's a human in the loop there. Now, in these cases where you have human in the loop, the applications are very sensitive. What that means is you are talking about fire, you are talking about thermal runaways, things like that, that do require double checking. And we are fortunate that, our products allow you to go back and forth between visual and infrared, in this case, the one that I'm talking about. So the human in the loop can very quickly see that in real life what's happening. If, for example, I see a lot of redness behind you in my infrared, I can switch to visual to see, ah, that's a coffee mug. Nothing to worry about there in this case there. So the customers don't want unnecessary alerts that can then obviously disrupt their operations because if any of those alerts are truly missed, that has a lot of impact. So some of these very sensitive and critical applications, Tom, we offer the, the solution of human in the loop. We say it's almost 60 40. 60% of the people do believe the AI does work. And yes, over the course of time, things will get better. I would say 40%. And those are purely application specific.

Tom Raftery: 22:19

Would it be fair to say that executives often treat maintenance as a cost centre instead of a continuity function?

Asim: 22:26

Yes. That's a short answer, but, but, but this is very interesting. Again, a great question. When I talk to my customers, there are few very interesting trends that are coming up. Why is that cost now directly related to the production impact? Well, one of our customers wants to make sure that they bring fresh products to their customers as quickly as possible, because otherwise, they, they're perishables, they can go bad. So the product goes from the farmers straight to their warehouse and straight to the stores, and they cannot afford to have the line down because if the line is down, the product goes bad. So on one hand you're talking about the cost of maintenance to make sure that the assets are up, but what is it doing though? It's enabling the top line of the P&L, right? So that cost then becomes an enabler in that case there. Where it becomes tricky or where it has become tricky, where again, we believe that we have a value proposition is that those costs were very variable. Meaning when you go back, Tom to the manual inspection days where you don't pick up stuff right at the right time. I'll go back to our favourite example of air conditioning unit. When you find that out, that some that it's broken and peak heat, and when you call someone, he's gonna charge you the maximum rate because he knows that he, you, you need him, right? Doesn't change in anything anywhere else. The same thing. Your whole conveyor belt goes down, and at that point you need someone to come and fix. You don't care about the expense at that point. So the cost becomes variable, which the executives don't like. They prefer to make sure that the costs are down, the variable costs are down. The cost in addition to that, the problem, as I talked about, the labour shortages, then you end up getting expensive labour too, if they're available. Those are the factors. Then that switches the mindset. So when I talk to my customers, the operating officers and the chief supply chain officers or the safety folks, they're starting to realise bulk of majority of them, the fact that this, if done properly, this cost actually enables the top line in this case. It's not just that, oh, I'm maintaining something. Come on, why am I doing it? It doesn't break down. When they talk about the implications of that breakdown most of them realise that. And I've talked to many, many of our customers in the IT segments. I've talk, as I mentioned to the supply chain folks. And I think supply chain folks have realised that I think the significance of the resiliency is very important to their customers. Because again, I was talking to one of the grocery store customers, they specifically gave this example where you have a person who shows up, he's got many, many tasks outside getting groceries, but when he shows up at the grocery store, he wants to get that product from that aisle. Well, darn it, the production line is down. We couldn't get the product. The aisle is empty. Guess what the customer does? I'm gonna go to the other grocery store and I'm never gonna come back. This is truly what the executives have realised that the customer service, right? So if you see through the lens of the business continuity, if you see through the lens of customer service, if you see through the lens of production, the behaviour start to change. Whatever was a cost is now like, oh darn. We gotta make sure our supply chain is resilient.

Tom Raftery: 26:10

That's the bigger business shift. Maintenance stops simply being a cost to control and becomes part of protecting throughput, service and revenue. The next question is how much of that resilience can eventually become autonomous? And do you think we're heading towards self-healing facilities, or just better early warning systems?

Asim: 26:33

I, I think we're still in the early warning at, at this point. It's not just technology, it's the human cultural and, the mindset change at this point, People are still thinking about manual ways of doing it. If it's not broken, don't have to worry about it, right? There are still a big group of folks there that's doing that. We also believe that the mindsets are changing, The self-healing, I'm sure the technology is there in certain cases, right? But we are far from there where we can actually in, in the data centre world, yes, we are in the lights off kind of a domain right now. They are making sure that the outside of the data centre is lights off as well, so they're getting there, but the self-healing and the ability to be lights off currently in the warehouses, we're not there as yet. I think that's where the resiliency will start to come in. Once we have these solutions implemented, the confidence in those solutions are there. And there's a history. A lot of these models, you know, this Tom, that when we talk about AI, the models work on data. The more data you have, the better predictions we can make. And that takes time. Sometimes a year, right? Some easily, because you wanna make sure you learn from those and then continuously update the models. So we are a little far away from there but the trend has already started. And I'm pretty sure, as I see it with my customers, more and more of our customers are now more educated. They're starting to get more. So that tells me that the wave is getting bigger and bigger in this case.

Tom Raftery: 28:07

And what do you think resilience looks like in a highly autonomous supply chain 5, 10 years from now?

Asim: 28:16

I think there is a continuous visibility, first of all, in your, in your assets. That's the first thing. Let's talk about it. It's a good question again. Resiliency focus means what? There's the operational infrastructure uptime. That's a component of that. There's the energy reliability. You gotta make sure you have the energy, right? We talked about that. There is the continuous visibility. When we talk about the supply chain resilience, there's predictive maintenance, there's operational intelligence, there's AI enabled detection. All of these are there. And then of course you have the, the transportation, the ports, all of those there, right? In this case there. But I think all of these factors have to kick in to be able to make them resilient. but that's essentially where I see it going.

Tom Raftery: 29:01

Time now Asim for the lightning rounds. So, I'm gonna throw several questions at you and you've got one sentence for the answer. Okay?

Asim: 29:10

I'm ready, man. I have to fix myself. I'm ready now.

Tom Raftery: 29:15

Okay, so faster delivery or more operational redundancy?

Asim: 29:21

Faster delivery.

Tom Raftery: 29:23

Better people or better automation?

Asim: 29:25

Better people.

Tom Raftery: 29:27

Good. What fails more often than executives realise?

Asim: 29:32

Motors. Believe it or not, they fail a lot and the people don't realise it.

Tom Raftery: 29:38

Okay. Okay, interesting. And the biggest resilience mistake that leaders repeat.

Asim: 29:43

They still believe manual ways can work. They don't realise that the maintenance has not kept up with the assets. Assets have become more digitised. Maintenance has not. It's manual. That is the biggest mistake.

Tom Raftery: 30:02

And if listeners, this is no longer lightning around, so you can go a little longer on this one. If listeners want to improve operational resilience tomorrow morning, where should they start first?

Asim: 30:13

I think they should first understand the, asset landscape first. Believe it or not, you think I'm joking, but a lot of the people don't even know the amount of assets that they have. When they fail, they realise there's an asset there, right? You may think this is not possible, but it does happen. And understand the behaviour of those assets, right? In this case, everything else, digitisation will happen later, I mean, even people like us, MSAI will come at some point. But first thing, first, understand the behaviour, the failure mode of those assets, Have the manual processes figured out the processes right? I'm not talking about the maintenance process, the manual process. If something was to go down, what do you do at that point right? If we came in, we gave you all the information, what are you gonna do with that? So make sure you have warehouse operations figured out. And when the automation comes in later, it actually just amplifies the, the productivity versus the other way around. That's how I would put it.

Tom Raftery: 31:16

Okay, and a left field question for you now, Asim. If you could have any person or character, alive or dead, real or fictional as a champion for operational resilience, who would it be and why?

Asim: 31:36

A very, very, very good friend of mine who recently, unfortunately passed away, Mattias Kerchmer, I'll put his name out there. He was known in the industry. He always believed in process led transformation and he and I worked in Accenture and predictive maintenance was our key focus area back then. This, I'm talking about 10, 15 years ago. And he was passionate about the fact that we gotta understand the full landscape, the processes, then you digitally transform them. Then you bring in all the IoT sensors on top of it. He worked very closely with the Dr. Sheer as part of the broader industry. He was a great influence on my career in terms of the knowledge, and God bless him. He was a phenomenal person as well.

Tom Raftery: 32:22

Sorry of your loss. If people would like to know more Asim about yourself or any of the things we discussed on the podcast today, where would you have me direct them?

Asim: 32:31

Well, they can obviously get hold of me directly in, in this case, if executives are looking at it and they're trying to understand the importance of the maintenance, right? So obviously we are available, but if you go to our website, we have a dedicated team who looks at all the information that's coming from the website.

Tom Raftery: 32:48

Great. Asim, that's been really interesting. Thanks a million for coming on the podcast today.

Asim: 32:52

It was an absolute pleasure. I learned a lot too.