Bob Rogers: 0:03

We were able to, help them redesign the way they're, packaging materials and holding materials across their supply chain so that with minimal increase in their total operating costs, they're able to actually serve their retail customers more reliably. That has resulted in about a 30% increase in revenue for them And, we think it's gonna be about an 80% reduction in chargebacks

Tom Raftery: 0:33

Good morning, good afternoon, or good evening, wherever you are in the world. This is the digital supply chain podcast the number one podcast focusing on the digitization of supply chain and I'm your host Tom Raftery Hi everyone. Welcome to the Digital Supply Chain podcast. My name is Tom Raftery and with me on the show today, I have my special guest, Bob. Bob, welcome to the podcast. Would you like to introduce yourself?

Bob Rogers: 0:57

Hi, Tom. Thanks. Yeah. So I'm Bob Rogers. I'm co-founder and CEO of O ii.ai. We're an AI powered supply chain design and optimisation company. Company.

Tom Raftery: 1:12

Okay.

Bob Rogers: 1:14

I was, I was gonna say, sorry. I was gonna say, in the past, I have also been expert in residence for AI at University of California San Francisco. I'm a member of the Board of Advisors to the Harvard Institute of Applied Computational Science. Was previously Chief data scientist at Intel where I worked. with, Intel's partner ecosystem in analytics and ai and, um, was co-founder and chief scientist at Op Apio, which is a healthcare AI company that was acquired by Centine in 2020. And of course, the most important thing about my past is that I used to work on black hole computer modeling,

Tom Raftery: 1:52

And when you say black hole computer, modern, you mean actual black holes out there in the universe?

Bob Rogers: 1:58

Yes, Yes. Started my career with a PhD in physics and, my first research was really about using computer models to understand what was happening inside, uh, or just immediately around super massive black holes at the centers of other galaxies. And in, in fact, I'd, I'd say, it was work that amounted to building a digital twin of super mass of supply chains. So, of course, digital twin is something we're hearing, in a lot of applications today. So there's a nice connection there with with my old black hole research.

Tom Raftery: 2:38

Okay. And what's the connection though between Black hole research and AI?

Bob Rogers: 2:46

So the connection for me is actually, if you think about what you wanna do with a digital twin. You're trying to understand how would this system behave if something happened? So it's like a what if scenario. In the black hole case, you've got material swirling down and getting compressed near the event horizon. You've got x-rays and gamma rays popping and crackling and you know, beams of gamma rays. And, and you wanna know, well what if, a thousand stars. All kind of converged in this, on this black hole at once, what would that look like? What if they came in at an oblique angle? And so you, you do this sort of what if, and the simulation tells you what, you would see if that, what if took place? Well, we do the same thing, in supply chain at OII.ai, we actually build a digital twin of the supply chain, and you wanna put in the right scenarios for the what if part of the question. And so what we do is we use AI to predict what are the kinds of challenges that a supply chain's gonna need to face. So for instance, we have a forecast of demand, but we know the forecast isn't gonna be exactly right. So what, might future demand scenarios look like? What might the network performance and disruptions to that network performance look like? And then finally, what might we see from the supplier side? Shortages, delays, inability to get product. All of those things need to be put into a set of future scenarios that you run the, digital twin through. And so what we do is we use AI to predict what are the most likely scenarios that a supply chain should be prepared for.

Tom Raftery: 4:36

Okay. And if I am a supply chain manager of, I don't know, it could be a manufacturing company, could be a whatever, every single supply chain is unique and individual, I gotta think it's got, each one has got its own foibles, its own inconsistencies. So how do you do digital twins of all these different, of every single different supply chain of every single customer or client who comes to you? Because they're all gonna be different. And if you have to do them all individually, I can't imagine it's scaling very well very quickly.

Bob Rogers: 5:18

So, there's a couple things. One is, it's really the creation of the digital twin is based purely on the data that we receive from the customer. So we are asking the customer for information about past demand, past forecasts, the network shape, the costs intrinsic to their supply chains, manufacturing costs, manufacturing and, and distribution constraints. So we use all that data to automatically build up a digital twin of each supply chain. And we're constantly looking for sources of data that will improve the fidelity of that. So, for instance, if you look at what's in the supply chain management software for lead times in your network. They're gonna be coming from the master data, which is not responsive to what's really happening on the ground. So we're, looking at, actual transactions in the e R P. We're looking at monitoring data for distribution networks, you know, actual downstream monitoring of packages to build a very, granular and time sensitive, view of what's happening in each supply chain. And then, all that data basically ends up creating what you would consider a graph of each supply chain in memory. And then the optimization is about figuring out not only what parameters for each portion of the supply chain. So imagine, imagine, we call it a network group. It's a, you know, the, the supply chain that delivers one product out to all of its customers globally. All the network groups are coupled together, and so we're computing the, best arrangement of tradeoffs between those, network groups, as well as figuring out what are the best parameters to run that network group. And it turns out that with all the data that's out there, we can actually get a very good view of what each distribution center, each factory, each, uh, warehouse, each cross docking station, financial nodes, whatever they're doing, we can actually capture that functionality once we have the data.

Tom Raftery: 7:37

Okay, and what happens next?

Bob Rogers: 7:41

You know, the, there's a bit of an iterative process because we're, we're taking in data. We wanna reflect that data back to our customers. There's a couple real benefits to that. One is if there are, and there quite often are errors in the data or places where something's been mis apprehended in the, source data systems. That gives our customers an opportunity to review that. Quite often there are patterns in those challenges which help them actually figure out how to improve their processes. At, at times we work with consultancies who will go in and do some process change management and some architecture to help do that. But what we're doing is shining a light on where, where are the places that the most, important, challenges to data are? So the data gets to a certain point. If there are places where they don't have data, we can impute it because we're working with data from many, many different, products and supply chains and, and regions. So we can often impute something that's not immediately available and then determine how important it is. So anyway, that whole process results in, a digital twin model, which is this graph in memory. and then we also, use the, the historical data along with what we're seeing across all of our different customers to make these predictions about what they should prepare for. So we can't, predict that a particular ship is gonna get stuck in the Sues canal on a particular day at a, in a particular location, right? I mean, it would be nice if we could, and people often think about AI as, wow, it's this magical forecasting ability. But what we can do is say, okay, what kinds of disruptions should you be prepared for? What if a, major delay in a particular lane is observed, or what if there are, a number of different challenges within your distribution network that kind of happen together and cause some compounding effects? Those are the kinds of things. So we build that. Then we run, all of that through the, current supply chain design and create what we call a health check. So the idea is we're showing exactly how your current supply chain is performing. And where the risks are. So these products are likely to have service risks and, these products, there's a risk of discards in this, part of this network for these, items. Where is there excess inventory that is not really serving a purpose to reduce risk in the supply chain? You know, there's quite often inventory needs to be moved around to be, to be most effective And you know, interestingly, when you think about service risk, if you ask any CEO what their on time in full is, it's a hundred percent of course. And that's, that's what's being reported to CEOs. But the reality is you can get to a hundred percent on time in full, the hard way, or the easy way and the hard way, which is what we usually see is massive amounts of expedites, factory interruptions, putting things on planes. I mean, I was, I was talking to a logistics company the other day and I was telling them about how wonderful our software is and it, reduces the need for these, air freight expedites. And the guy said, Well, don't talk to any of my customers because that's how I make my money, is charging them huge dollars to put stuff on planes. So, yeah, you might be able to get it there, but you're gonna get it there at a high cost and, and there's gonna be a lot of disruption. So the, the benefit of the health check is it really tells you where you're configured, given the realities of supply chain on the ground. Not just your master data, but the realities, what kind of service you should expect. And so that's, the first point. And then what we do is we actually run, uh, an optimization of the whole thing. So this is a direct, multi parameter, optimization of the system to figure out what are the, best, most cost effective ways to achieve service levels or to make the trade offs between service and cost, across the supply chain, and especially given that there are capacity constraints and and things like that. So we do that. And that's called the action boards in our software. The action boards tell supply chain designers or planners, exactly what steps to take to mitigate those risks. So it's, it's not good enough to say, Oh, well, here's a dashboard of all the places things might go wrong. We want to give them an action board that says, do this. And that way they can, take the action that they need to take.

Tom Raftery: 12:23

Okay, and wouldn't a seasoned supply chain executive be able to do that by themselves without the, the requirement for an AI?

Bob Rogers: 12:37

Yeah. So, I think in theory, given enough time and the right computational tools, the answer would be yes. The challenge is these are highly interconnected systems with lots of coupling between them, lots of complexity, and so in practice, what I see quite often is. The same safety stock, same minimum order quantity, same reorder frequency, applied peanut buttered across an entire supply chain. And that's the design you know, if you think about what those managers are doing, one of the things that they're thinking about is segmentation. How do I divide up my products to work out what's the best service level to target for each of them? So, you know, the classic thing, at least in pharma is to do a nine box where you have A,B, and C class items, which are the, you know, highest, middle and lowest revenue producing products. And then you have X, Y, Z, which are the lowest and highest, variability respectively, often measured as coefficient of variants. And then you, you have some sort of strategy for what service level you put for each of those? Well, what we've seen on a number of occasions is that despite the manager, having some scheme for, for how to assign those service levels. The reality on the ground doesn't match it at all. So there's, both a lack of optimization at that higher level in terms of the design, but also, a lack of transparency around execution. So they can't see easily that the, the strategies that they're putting in place are not being, followed. So it turns out that the, the result is that in practice, very, very few organizations have their systems, optimized. And then here's the kicker. In three months, everything changes. Network, network times change, network variability has changed, demand variability and supply variability change. And are, are you gonna go back to the drawing board and do all that manual calculation again? And then try and keep up with that all the time. Or do you want a tooling that will just tell you, here's the thing you need to change. Now here's where you're in the red zone, where you're going to be in the red zone if you don't make changes. So it, it ends up the fact that the world is changing so much, and this is not just the pandemic and war and all that it was happening beforehand. It's just that the, the scale was not as large. All that change makes it virtually impossible for someone to do this by hand. And then I, I'll add in organizations where planners are setting parameters, which is a highly, concerning practice anyway. You've got a lot of turnover in planners. So planners are trying to catch up. They've got maybe less experience on average. They're trying to keep up with, you know, suboptimal tooling. This kind of a AI augmented intelligence really helps them stay on, track because they know what information to focus on.

Tom Raftery: 16:24

Okay. And again, if I'm a supply chain manager and I have your system rolled out and I've got these action boards telling me something is likely to happen and I should take these actions, how do I know to trust that?

Bob Rogers: 16:39

It's a great question. So, first of all, the, the starting point is that you gotta be able to see why is it that the, the system is, making this recommendation. So, you can actually do what if scenarios for each recommendation and try different things and see what the impacts are. So immediately you start to see, it's often a question of do I increase the safety stock or do I decrease the safety stock? Or what is the impact on, on, you know, of reorder frequency on something upstream, say, at the factory. So you, you give them this ability to turn the knobs and see what the impact is and see what the underlying factors are that are driving that. And that ends up giving them confidence that they're making the right decision, or quite often what it does is it gives them confidence to move in that direction, but maybe not all the way. So we often see, suppose you have a, a recommendation to go from 30 day safety stock to 20. Most, network designers are gonna be very uncomfortable with that jump and so first, they can see what, why that trade off makes sense. Then they can put in the value 25 and stick with that for a transitional period. We're tracking the compliance of different, folks to the recommendations, but not in a punitive way, just in an informational way. And so then over time you see, oh, well it really does make sense to move to this other, this different configuration. and so they can, they can see how it pans out.

Tom Raftery: 18:12

Okay, and do you have any referenceable customer success stories you can talk to?

Bob Rogers: 18:21

Yeah. So, we've done some work with a very, large pharma company where we took all of the products from two of their, packaging factories and the distribution of those products all the way out globally. So all over, all over the world. And, we, we optimized that whole system and found, a, a couple of interesting things. One was they were actually not using their supply chain management software. So the, the software was running. But everywhere you looked at, what the inventory would be based on the parameters, they had, uh, a large, multiple more inventory. And so this is a typical thing you see in pharma because you don't want to ever have a, a service issue, right? A stock out, with a critical pharmaceutical is a big problem. But here's the challenge, when everyone is overriding the system first of all, you have no operational control and no transparency. But furthermore, there's this really interesting challenge that pops up in pharma, which is that these products have expiration dates, and so if you put too much material into your supply chain thinking that you're gonna have high service. By the time the product gets to the end, it's close to its expiration date, and you end up getting service disruptions as a result of write offs. And so actually doing the, computational exercise of working out exactly where that best balance point is, is actually much easier to do with, with analytical tools. So that was a, really interesting, informative, project where we're still working with them to develop a plan for going to full production. But it's something that's an ongoing discussion. On the other side of the spectrum, a, consumer, packaged goods company, they are actually providing white label vitamins and other products to a number of retail outlets. And their challenge is that they have very little visibility to demand from their customers. They have a lot of variability in, in their suppliers, and so we were able to, help them redesign the way they're, packaging materials and holding materials across their supply chain so that with minimal increase in their total operating costs, they're able to actually serve their retail customers more reliably. That has resulted in about a 30% increase in revenue for them And, we think it's gonna be about an 80% reduction in chargebacks, just on the first deployment of the tooling. So, cuz you know, the chargebacks happen when they get a stock out. But also if they are off by even a few hours. on their estimated delivery. So increasing the control over exactly how the product works through the entire supply chain has a huge impact for them on chargebacks.

Tom Raftery: 21:32

Interesting, Interesting. I gotta think as well that this kind of making supply chains more efficient is a sustainability win, even if it's not being sold as such.

Bob Rogers: 21:48

Mm-hmm.

Tom Raftery: 21:49

Uh, how many organizations are thinking in that way because if they're engineering, if they're using something like this to engineer waste out of the system and become more sustainable, it should be something that they're measuring and reporting on in terms of their sustainability reporting.

Bob Rogers: 22:05

Yeah, I'm really glad you asked that question because something we've put some investment in. One of the things that we report on for each potential way of designing the supply chain is what is the carbon, output of that design. So the annual carbon, and we express that both in terms of kilograms, but also in terms of cost. So there are some, you know, cost assumptions you can put in. In terms of what people are asking for, we worked with an online retailer of, organic cosmetics. They were very driven by reducing their carbon impact. And so actually, if you think about the way you optimize, a system with lots of variables, what you do is you say, I'm gonna create a function that's sort of like the effective cost. And that I'm going to reduce the effective cost in a way that gives me what my service and other, corporate objectives are, but reducing that cost. So what they did is they said, All right, we want to double the cost of carbon impacts so that they have twice the impact of anything else. And we just tune that knob when they optimize automatically. It's now putting carbon in the driver's seat for some cost tradeoffs around the design, which is a really nice way to just automate that capability. The other thing that I imagine, and we haven't seen this, really come up yet in terms of practical implementations, but I'll tell you that this is an important part of, my vision for, supply chain. Suppose you have two vendors who are competing for a particular piece of work and, logistics in a supply chain. Well, suppose their cost and their performance are, on an equal footing, but one has a completely different carbon, footprint than the other.

Tom Raftery: 24:04

Sure.

Bob Rogers: 24:04

Well, right. You have to be able to expose that information as part of the decision criteria. So because we're modeling those capabilities and those, those impacts in principle, designers can choose vendors who are actually, actually driving toward more sustainability. And in my experience, that's the only way to get real change to happen is you've gotta put the knobs in the place where people are making decisions. So we're excited about seeing that come to fruition and, you know, anyone who wants to, uh, instrument that I certainly would encourage them to reach out.

Tom Raftery: 24:43

Sure, Sure. Cool. I interviewed this morning a guy from a company called Earthly, and it's for my other podcast, The Climate 21 Podcast where, you know, I, I highlight successful emissions reduction stories and what his company does, what Earthly does is they provide a carbon footprint, platform for SMB's. And it's a sector of the market that is hugely underserved by tools like this. And that's why they, they're going after it. And one of the reasons is they're now working with the larger enterprises who are requiring their supply chain to report emissions associated with, purchases. And so now Earthly can go in and help these smaller companies to report their emissions to the larger enterprises. So it's, it's a fascinating use case.

Bob Rogers: 25:38

It is. And, you know, we're, we're really in some ways a data company. So taking data from these platforms that are collecting information and then actioning it really is, it's an exciting area where we can kind of advance things.

Tom Raftery: 25:54

And where to next for you guys?

Bob Rogers: 25:57

So there's two, two areas of, growth for us. So there are a couple of interesting optimization use cases that we're interested in pursuing further. So, one of them is, and we've done some work on this, one of them is, is optimizing under constrained supply. So we did work with a company called Trigen in India during the pandemic, they were the company that was tracking vaccine distribution and people signing up for vaccine. And physically, where is the vaccine, but india didn't have a good way to connect the dots between what the policy was. Things like, okay, for every region we wanna prioritize over 70 with comorbidities or, you know, regions where there's a high population density and you wanna slow down, um, spread. So what would happen is a batch of vaccine would be released and each administrator along that entire multi echelon supply chain was making sort of independent decisions about where downstream, those, things should go. Well that's, so that's a local optimization based on basically no information except, maybe, you know, opinions and who shouts the loudest. So, and, and there was a policy in place. So we built an, a global optimization that optimized across all the echelons to allow them to, drive the right vaccine distribution through all the different links so that it would get to the right place at the, at the end of the supply chain so that those over 70 comorbidity folks were getting the vaccine first. And, that worked extremely well. It's a really great example of a global multi echelon optimization being way better than a local optimization that's, you know, made in, in the absence of these this broader view. And so I think that's an area where pharmaceutical companies, not just in vaccines, but in other, other types of biologics where the exact output of the process, it takes a long time to make these biologics in a vessel. And then, the exact mix of product and quality that comes out the other end is not, known until very shortly before the batch is done. So there's this sort of adaptive piece. We're very interested in, in increasing the use of O II to manage the, the factory constraints, the, the sort of, uh, sequencing on the factory floor and connecting the dots there because sequencing, we see huge planning time fences driving massive costs. And a large planning time fence is basically a proxy for not having, good ways to sequence and optimize sequencing, as it connects to supply chain. And then, just in general, the more places we connect to real time data and really create that, that live view of what's happening now. So, think of when you're monitoring parcels and they're moving through the supply chain, we've got a measurement of the last five, the time it took for the last five shipments, Right? Well, What you don't necessarily want the average or even the average and some probability distribution. What you want is a prediction of the next five, so that you can use those to very explicitly plan how you're gonna drive everything. And that works in the optimization, it works in the planning. And so we really are pushing to connect the dots on all that very granular data so that the system is very, very responsive.

Tom Raftery: 29:47

So we're, we're coming towards the end of the podcast now, Bob, is there any question that I haven't asked you that you wish I had or any aspect of this we've not touched on that you think it's important for people to be aware of?

Bob Rogers: 29:58

Yeah. So, I think I'd be, interested to, talk a little bit about people's concerns about AI in general. Both is it safe and also is my job safe? And, so, in the question of is it safe today? AI is safe unless it's specifically attached to some sort of weapons system I think, uh, there's a lot of stuff going on in the war in Ukraine right now, which is completely outside the scope of this discussion. But generally speaking, we're not, imminently in a Skynet scenario. And so, AI is creating way more, positives than negative in my experience. But the other more tangible question is, what is AI doing in terms of people's jobs, is AI taking away jobs? And, and, how should I think about that and my experience? Again, I, my view of AI is that it really should stand for augmented intelligence as opposed to artificial intelligence. The best way to instrument most AI systems and OII is no exception, is to use AI to do the grunt work. To get the right information in front of people so that they can use their brains and their experience to make the right actions faster and a and in a more informed way. So you don't wanna automate away people's ability to make decisions, But what you do wanna automate way is things like, Okay, go into this database and, look for the ones that are, you know, two and a half standard deviations outside the norm. You know, go into this database and look for, you know, there's a lot of things that people have to do to configure and tune systems that really aren't things people are good at, but once you've got that information surfaced in the right context, it's very easy to make a good decision. And so our goal is always to provide that powerful information as quickly as possible and put it in the right context so that it can be acted on easily and confidently. So really we're making it, making people's jobs less, less painful in many cases. And then, you know, in, in the context of a labor shortage, When you have people coming into an organization who have less experience, you're not depending on tribal knowledge and post-it notes on the walls of cubicles and, secret Excel files, that only the incumbents know about. And so you're really creating a much more, agile environment for your, workforce.

Tom Raftery: 32:39

Nice. Nice. Super. Bob, if people wanted to know more about yourself or oii or any of the topics we discussed in the podcast today, where would you have me direct them?

Bob Rogers: 32:52

Yeah. So, of course they can email me and I'd love to talk to them about anything supply chain, AI or black holes, uh, . if that moves them. I'm at Bob at oii.ai. We have a website, which is O ii.ai, and they can certainly come to the website and request a demo or a meeting there. And, also on our LinkedIn homepage, we have, use cases, case studies, white papers, and some, information from customers talking about their experiences. So any, any of those, those means would be, highly encouraged, reach out and, ping us.

Tom Raftery: 33:34

Excellent. Bob, that's been fantastic. Thanks a million for coming on the podcast today.

Bob Rogers: 33:39

Well, thank you for having me. It's really been a great opportunity to talk to you and great fun. Thank you. And encourage anyone to come out or reach out and, uh, communicate with us here. Find out more. Okay, we've come to the end of the show. Thanks everyone for listening. If you'd like to know more about digital supply chains, simply drop me an email to Tom raftery@outlook.com. If you'd like to show, please, don't forget to click follow on it in your podcast. Application of choice, to be sure to get new episodes. As soon as they're published. Also, please don't forget to rate and review the podcast. It really does help new people to find the show. Thanks. Catch you all next time.