Michael Prokopis: 13:15
Some digital shelves? We call it digital shelves. What it is is it’s a camera in the ceiling. And that camera has, depending on the technology we decide to use in any given room, up to 32x zoom capabilities if we’re in a larger room. So I may have to scan 25 or 30ft away, and I the first thing I need to do is read a barcode, because that’s how I’m going to map the item to the item master file to get the characteristics of what it is, being measured in that room.
And so if you think about it, a 32 x zoom camera plugged into the ceiling. Seems like a relatively easy thing to do. But then what you have to do, and this is where the challenge becomes, is you have to start differentiating between product types. A flat pack that may be a blood pressure cuff looks very different than a syringe, a band aid, or a roll of gauze or whatever it happens to be in the room. And so we needed to figure out how to build the capabilities to contextualize those items.
So we could teach the AI, the machine learning exactly what it was seeing and how to think about counting it. And I’ll give you a let me just give you a couple of quick examples. So in the healthcare supply chain, my friends will tell you that the lowest unit is important, and here’s why it’s important. The lowest unit of measure doesn’t mean one each. It may mean one carton and one each versus one carton are two very different things because one carton, when it’s open, may have a thousand of something inside of it. How do you count a thousand? Something’s inside a box that you dump into a bin, whether it’s a human person doing it or whether it’s the camera doing it. The camera has the ability. We’ve taught it how, in those instances where it can’t physically see all the items itself, to look at the top layer and count it precisely, and then you guesstimate based on the depth of the bin that it’s in, how much is actually there.
And we have found with the algorithms that we built that we’re well above 98% accuracy. In fact, we’re at 98.9% right now at MD Anderson in the first 25 parse rooms that we’ve installed at. And they just released another 25 rooms for us. And we expect to be at 98% in those rooms by the middle of May. So essentially what it is is a camera, and then there’s an AI that’s running in the background. It scans the room, and the first thing it identifies is the bin size. And then, and then, and, and, and those were the bin sizes because we’ve loaded a CAD drawing. So it knows spatial relationships. And then it starts solving for the different angles, the parallax angle that it sees each of the items at. And that allows it then to understand what it’s looking at, and it maps it to the item master file. The item master file gives its inventory characteristics. What’s my power minimum? Maybe I’m going to have a minimum for my reorder point of eight.
And so when the system sees four, it automatically assumes one more day of burn rate. So let’s just say two more units between now and tomorrow when I’m likely to get restocked. And so it will go out, and it will order eight items so that tomorrow the bins will be full. And now we can start the cycle over again. But the most important part of that is we don’t always consume products the same way. And because we don’t have the analytics, we don’t have the real-time visibility. It’s a science project to figure out what your minimum should be or what your order quantities are. And so we set them. And then two years later, we finally find enough time to go back, human time to go back and do it again. Well, there are a lot of changes that have happened. This system is designed to look at the system and understand velocity. And so if you’re going through, let’s say you got four days on hand, that’s how you’ve set up the inventory. So every one on the second day you’re going to reorder the system now knows, hey, I burned through it faster than I need to send the requisition.
But I’m also going to send an alert that says your velocity has increased. And now, as a practitioner in a hospital supply chain, I can receive those alerts. I can say, well, that’s because they had a one-time volume increase, or I know they had a little burst of activity. I’m not going to change anything. But now we can go have dialogues with our clinicians about the standard of care and have things changed or not changed, because that allows us to stay on top of the problem, because what we don’t want ultimately is stranded inventory. So you’ve got two problems. You’ve got one, I don’t want to spend more than I have to when I have to. And then on the other side of it is, I don’t want to have stuff that just is going to expire, and I’m going to have to throw away. Both of those problems are bad in a healthcare setting where margin matters.
Dr. Jeremy Weisz: 17:31
Software engineers. So you must have a team of software engineers. This seems like a very difficult feat to pull off, just overlaying video with what you needed to do by just visualizing it.
Michael Prokopis: 17:45
It is extremely complex. But believe it or not, we’re doing it with six engineers. Garvace and his incantation when it started, you know, through the Airbus and the USPS in particular, they went up to 140 offshore engineers. And when we we right size the company, and then we focus it on US health care. We found out that the tooling that we had built really enables the AI to learn at a much faster clip. And so we’re able to, to it’s able now to ask them for things and we’re able to respond to what they ask. And so we’ve created these tools that say, hey, your variances are, and it knows what its guardrails are and says, my variances are off. So we go in, and we take a look, and we’ll make whatever modifications are necessary. What’s really cool about the technology as well is that once it’s seen an item.
It decides how it’s going to, it’s going to measure and how it’s going to count it. Whenever it sees that I am going forward and knows what it is, and doesn’t matter what room it’s in, it’s going to treat it the same way. And so your speed to implementation, as you see more and more items, is just increased. And really, the longest tent in the pole becomes installation time for cameras because you have to get a third party involved. They have to pull cat5 cat6 wire. You’ve got to plug them in. We’ve got to make sure everything works. That takes time, but the actual eye itself takes a minimal amount of time. In fact, when the moment you turn a camera on, you’re probably at 90% accuracy within two weeks. By the time you get to four weeks, you’re probably at 98% plus. And we expect that to continue to compress as we teach the AI more and more products.
Dr. Jeremy Weisz: 19:19
So, is it mostly hospitals or who’s using DARVIS?
Michael Prokopis: 19:25
Yeah, we are exclusively right now focused on US healthcare, ID, and systems. However, we have been asked by a large retailer to take a look at janitorial supplies for them in a vendor-managed inventory context, because they drive to these locations to make sure 2 or 3 times a week to make sure products are on the shelf, and what they find is that half the drives are not fruitful, right? They’re not needed. They don’t need to bring anything. They’re just going to go in and count again. They’ve already counted.
They don’t. Right. So it’s kind of a waste of a trip in some respects. And so they’ve asked us, can you put a camera in there and help us do that? And the only reason we’re doing that is that health care has those same kinds of closets, and they have large supply rooms where cleaning supplies are housed, but they also have these cleaning closets throughout the entire facility, and we’re going to help them. And so it was a natural to say yes in that retail opportunity, because it directly feeds back to what we’re doing in health care.
Dr. Jeremy Weisz: 20:18
Yeah. Talk about that for a second, which is healthcare. And then how much do you look at expansion when the healthcare is such a big market?
Michael Prokopis: 20:29
Yeah, so very carefully and selectively. And we’re just starting to have those kinds of conversations now because I don’t want to dilute what we’re trying to do, which is to be an effective partner in health care. But there are opportunities that have come up, and we have to consider them one at a time. Are they valued? Are they additive to what we’re doing, or do they divert our attention? If they’re additive to what we’re doing, then it’s easy to say yes, because it shouldn’t be too much of a deviation from where we are. But in those instances where it’s not, then you have to decide, am I going to stand up a whole other team potentially, to focus on that industry? And, you know, I think I described you before, Jeremy. I gave an executive session for the Zen Terraces executive MBA class at Texas A and M, and I talked about my history and my background, and how I came to DARVIS and how DARVIS originated. And my original idea in 2008, when I was done, several people approached me from other industries and said, how do I get it? How do I do it? And the answer.
Dr. Jeremy Weisz: 21:26
What industries were interested?
Michael Prokopis: 21:28
Yeah, I saw one is large. Polyphony company, I have large construction sites all over the country, and so they want us to put cameras in their construction yards to help them manage their inventory. We talk about retail. There are a lot of instances in retail where our technology could work. And in fact, in many instances would even be simpler than what we’re doing, whether it’s a retail store or a grocery store or other things like that. The way we talk about it, though, especially in the vendor-managed inventory world, is the Walmart model. Walmart says, Okay, vendor, you get ten feet of shelf space, and I don’t care how you manage it, but you don’t ever run out, don’t ever miss a sale.
And you. But you’re going to get it every time a unit goes out the door, we’re going to send you a point of use scan information. So you know when it was sold, make sure it never runs out. And so that’s how we’re thinking about vendor-managed inventory in a lot of instances, especially in primary care clinics. And you know, some of these smaller doctors’ offices where they just have closets full of stuff because that was the easiest way to order when in fact, you know, that ends up being stranded inventory or in some cases, wasted inventory. And so there’s a way to do all of that much more efficiently and effectively. And so those are the kinds of things we’ll think about as add-ons as we continue to grow.
Dr. Jeremy Weisz: 22:43
Our medical supply companies are there. Because it’s like, it’s interesting because the hospitals touch a lot. Like you said, a lot of different things. But I mean, I visualize these huge whatever Medline you line, those are all probably feeding into some of these companies anyway.
Michael Prokopis: 23:01
Yeah, absolutely. And they do have an interest. I mean, you know, if I’m, if I’m a mainline distributor, I may say, hey, you know what, I can completely change the economics of the game. And I’ve had these conversations with big distributors about you owning the inventory all the way to the shelf. I only get charged for it when I take it off the shelf. What was missing, though, was that true point of view, real-time data saying it just got consumed. And without that, it’s awfully hard to manage a program like that. And let me talk about consignment for a minute. We still do consignment in the healthcare industry. And so you sign an agreement with your partner, and you say, all right, I want ten larges, and I want 12 mediums, and I want six smalls.
Well, it takes so much energy and effort to figure out if that product volume mix is right. You never really go back and change it, even though it’s never right. And so you end up with more product on the shelf or less product on the shelf, or you’re constantly chasing the wrong sizes. And it takes 30, 60, 90 days to pull all the data together to have an effective conversation with that vendor. And so here we’ll know exactly what sizes are being consumed and when. And so now you can update a consignment agreement literally at the stroke of a pen and an amendment rather than, you know, the science project that we were doing before. And that’s exactly where DARVIS fits in. We want to eliminate as much manual work as possible on the things that should just really, quite frankly, be automatic.
Dr. Jeremy Weisz: 24:20
Talk about selling into, I imagine there’s a lot of red tape when dealing with health care and hospitals. What’s the. What have you found works from selling into and getting in front of these people, that you have a solution that will help them, because they’re probably used to their old ways, I imagine.
Michael Prokopis: 24:38
Yeah. Patience. So there are early adopters, and they’re interested, and they want to talk to you. But the reality is, you know, cyber risk is real. And so, you know, a lot of health care systems have been burned in the last number of years. You can see them in the Wall Street Journal. And so many of them are really careful with, you know, cyber risk and making sure they understand what’s being exposed. They’re also starting to understand the value of their own data. So, how is your data going to be consumed not only by you now, but in the future? And they want to know that. And I think they should. I was a proponent when I was at one of my own supply chains in health care, that our data should be monetized. It’s literally one of the last assets we have that we truly own, independent of everybody else.
And believe it or not, it is a wealth of information that can be used. And so really, that becomes the question. And so for me, as I started thinking about it, I know a lot of people who sat in my chair. And so I understand how busy they are. And I understand how valuable, you know, if I can get 45 minutes with somebody, the first thing I do at the end of that session is thank them for giving me 45 minutes, because I know that there’s probably ten other things that they probably could have gotten done in that 45 minutes, and I prevented them from doing that. So I’m always very grateful. But you’ve got to be in it for the long run. You’ve got to be willing to get in, do a proof of concept for very little money, maybe just cover your costs at best. And then you’ve got to be willing to just let the process work itself through.
Dr. Jeremy Weisz: 26:02
Because on average, Michael, like these facilities, how many average cameras are they do they need because there’s a lot of storage, a lot of closets, there’s a lot of stuff. Right?
Michael Prokopis: 26:14
So yeah, so MD Anderson, they said they want to get to 250 PA locations right now. The way the installations have gone were it’s about 1.1 cameras per room. Now that’s greatly skewed by a core where we installed 12 cameras. So one room with 12 cameras versus the usual. So, most, most room, most PA locations, you can get by with one camera. And then occasionally you run into PA locations that need to. But as you get into these bigger rooms that have a much larger inventory inside of them, you do need more visibility.
Dr. Jeremy Weisz: 26:45
I feel like, you know, selling is selling no matter what industry. And when we first talked about it, it went into having a very clear value proposition, right? And like you mentioned, making it as frictionless as possible for people to try it and then do more, right at a lower price point. But talk a little bit about how you came up with the value proposition so that they listen to what you say. You’re not just spouting out features, but like how it will benefit them.
Michael Prokopis: 27:15
Yeah. So when we started, when I, when I first started interacting with DARVIS, and we said, we’re going to, we’re going to build this thing together, I gave them six problems I needed to solve. One of them was labor. One of them was vendor management of inventory and information flow to that community. One of them missed revenue opportunities because of scanning issues, or somebody was just too busy to even pick up the scanner in some cases. And a couple of others. But, but so those became the six tenets of how we were going to build DARVIS. And then at the very end, I said, but the most important thing is, this product has to pay back within 12 months, you know, like 12 months. That’s insane. I said, well, here’s the reason why in supply chain and health care, when I go to the CFO and say, Hey, I got this great idea, and it’s going to pay back in 4.5 years, the CFO is going to say, No, thank you 999 times out of a thousand.
And the reason is, they can buy an MRI that will pay for itself in that period of time, and they’re getting revenue in the door. And so, you know, it becomes a very hard sell in the supply chain to ask for technology dollars when you can just throw more people at it. It’s probably cheaper in some cases, but you don’t get the benefits of it. And that’s the distinction that we make. And so we talk about, yeah, there are some human upside, some human labor upside that you’re going to get. And you can decide how you want to redeploy those people into other tasks because there are always other tasks. You’re always understaffed. It’s the supply chain. But more importantly was we had to get to that payback with an ROI. And I talk all the time about 19- 20% ROI within 12 months. MD Anderson will probably realize that within ten months.
Dr. Jeremy Weisz: 28:50
It’s really interesting. Michael. First of all, I have one last question before I ask it. I want to point people. If you’re watching the video, you can see Darvis.com. Check out what they have. It’s pretty amazing, the technology. It’s interesting. I had a company, and my last question, I’ll tell you, but I’ll tell you a quick story on this. My last question is about some of your mentors. And this could be distant mentors, meaning like books, you know, people that you’ve read about, companies you’ve read about, and/or actually just mentors in the industry that have helped you or colleagues. But I had a company that was AI, it was called AiLert, and their camera that detects firearms.
So they put them in public places so they could immediately hook up. And if it’s on the camera again, it’s probably just a technology that’s overlaid on video, just like yours. Overlay on video just for a different use case can then go, what’s the, what’s, where’s the location of the person? What’s the type of gun and alerts the, you know, emergency staff, police, or whatever right away. So people, the responders can get there within, you know, minutes instead of hours. So that people can check out that episode, just like a very cool use case of video and AI technology, right? So yeah, so for you talk about mentors, either maybe resources books or, and, or colleagues and people in the industry.
Michael Prokopis: 30:28
Yeah. So, you know, in the industry, and it’s interesting, I found the healthcare industry to be very different than anything else I’d ever worked in before. And, I think it’s because we care for patients, therefore we care more for each other. And, we don’t feel the need to compete as much. We’re willing to share ideas and concepts. And so I’m going to talk about four people, quite frankly, who really kind of not only mentored me, they became part of my inner circle. We started talking about long-term demand planning. How do you start thinking about data 12 or 18 months into the future, and how do you make real predictions so you can help not only change what you’re buying today, but help your manufacturers figure out what they should really be manufacturing for you? And so that conversation took place with Dan Hurry at Bon Secours at Hitchcock when he was at Trinity.
Amina Mettler, when she was at UPMC, Jay Kirkpatrick at Lifepoint. And we met every other Monday for two years. Now think about this. This is some pretty high-powered people in our industry running really, really large organizations. And they took the time for us to get together every month or every other Monday. And what we did is we talked about experiments that we were running or trials that we were thinking about, or ideas that we had, or how we were trying to do something else. And so that collectively, that brain trust was amazing for me because it continued to reinforce the things that I thought, the things that I learned. You know, I have two master’s degrees. I went, I went to Dartmouth College, Tuck School of Business, and I got my MBA in the mid 90s. I learned the language of business. What’s interesting is that my entire career, though, had been around distribution, logistics, and warehousing. And it wasn’t until, you know, I think after 2000, that we actually named it supply chain. But I ended up getting a master’s degree from MIT in supply chain management. And that opened my eyes to the language of supply chain. And so I learned things. I probably knew these concepts, but never really understood the math behind them. Now I understand the math behind it. And then given the fact that I grew up in technology throughout, you know.
I was very fortunate early in my career to work in boardrooms with some of, you know, the biggest titans in business. You know, Bill Gates at Microsoft, Jack Welch at GE. And to watch how they think about problems, and how they ask questions, and how they narrow in precisely to the weakness of your analysis. I was always amazed by that. And so those kinds of people, even though they’re not maybe direct mentors, they’re teaching you things. If you’re just aware and you pay attention. And so I really kind of thought of myself as a sponge my entire life. And I’ve always been willing to ask the what-if question, you know, kind of like probably like Elon Musk does too many times a day. And I’ve been called the mad scientist most of my career because I have these ideas, and people are like, Where did that even come from? I was like, I don’t know, it’s just something I’ve been noodling on. And it just felt like it made sense to me. And, you know, from a supply chain perspective, there’s a great book. I’ve got the fourth edition sitting here now called Inventory and Production Management in Supply Chains, and it’s the de facto book on supply chain. Who wrote that?
Michael Prokopis: 33:23
You want to understand the economic order and quantity. You want to understand how the parse or perpetual inventory locations differ from each other. If you want to understand how logistic networks work and how you expand them, that’s the book for you. And it’s been in production for a long time. And, you know, I always look at it and say, Hey, if you know professors at major universities or have it on their bookshelf, then that’s a book I want on my bookshelf as well.
Dr. Jeremy Weisz: 33:45
Who wrote that one?
Michael Prokopis: 33:48
I have to look it up.
Dr. Jeremy Weisz: 33:48
This is a.
Michael Prokopis: 33:49
Three guys, Silver, Pyke,and Thomas.
Dr. Jeremy Weisz: 33:52
Okay, I’ll check it out. Yeah, obviously it’s not for everyone, but for the right person. It’s the perfect book. I mean, it reminds me of one of my favorites, The Goal by Eliyahu Goldratt. I forgot how to pronounce his last name, but it reminds me a lot of a little bit of that type of book, right? Which is like on operational efficiency.
Michael Prokopis: 34:15
Yeah. And for my world, I gave it to my technology people because I wanted them to, when I said economic order, quantity. They’re the ones you want to know exactly how the equation is built. And so they want the math behind it. That math is in there. But for my CFO and some of my other folks, I gave them the, you know, the simpler version of supply chain so they could get the gist of it without having to comprehend and, you know, supply chain for dummies. I mean, I’ve got one of those sitting on my shelf as well. And so I was willing to read it because I wanted to understand how, how, you know, that book presented supply chain versus how other things did. So it was kind of both of them are interesting reads.
Dr. Jeremy Weisz: 34:48
I love it, Michael. Only the first one to thank you. Everyone. Check out Darvis.com or episodes of the podcast, and we’ll see everyone next time. Michael, thanks so much.
Michael Prokopis: 34:58
Thanks, Jeremy. I appreciate your time and your audience.
Dr. Jeremy Weisz: 35:00
Thanks.
