 Hello everyone, this is Chris Kaplis and this is the SC2X second live event And I'm here with Steve Ellett who's the VP of supply chain network design or all network design at chain Analytics and we're gonna talk for a little bit about some questions for him about some of the things that he's done in his career He's spent the last 20 years pretty much working on network design So I know no one who's done more of these engagements than Steve and I'll let him introduce himself in a second Let me just give you the lay of the land We're gonna talk for about 15 20 minutes at the same time you should be able to enter in questions using Slido So you should do Slido slido Dot-com it'll take you there and the code for this SC2X And you'll be able to see polling questions that we put out And it's also the way for you to enter in questions that you have so if you have any questions for myself or Steve or Sergio or Arthur or anyone here just enter that in and we'll try to get to those And we'll let you vote on those so as questions come in I'll advance them out So everyone sees them and you can vote on which question you want to see first So we'll try to answer your questions starting in about 20 minutes, but I'll start out asking Steve some questions first Thanks, so first Steve tell us a little bit about yourself. Absolutely. We'll appreciate appreciate the invitation very honored to be here so I lead chain analytics supply chain design practice and I've actually been doing supply chain design work for 25 years as since you're in high school hard to believe. Yeah, hard to believe middle school is a So it's I thought about the other day. I was actually doing this I started out doing this kind of work in the cement industry while I was actually in college undergrad I was in school for industrial engineering at Georgia Tech I started out working while I was in school and just after school for a cement manufacturer And I was doing supply design work gave me the opportunity to build some models and some early tools It was actually a tool called what's best by lindo that you guys might very well. You still use that. Yeah, exactly It's still around. I was that's where I started so it's and from there just moved into some other things But I went from there. I met Some folks at a software company called the intertrans logistic solutions and there was a product that they brought to market called supply chain strategist and I joined that firm and I was in consulting around that product for a while So I was helping companies use it and learn it and doing some training and that sort of thing And then they were based in were they based in Toronto? They were based in Toronto But the development office here was here in Cambridge about two blocks from you. That's right. Right. That's right So it's very very nearby and so I had some good exposure early on matter of fact in my Georgia Tech days You know one of my early mentors Don Ratliff was one of my professors who was a founder of capital districts Who was another early kind of pioneer in the desktop optimization space? So I went into software I was in consulting for a while then I was the product manager of their product called supply chain strategist for a number of years left there and chainelitics started up at the end of 01 I joined in very early 02 and Have been at chainelitics ever since doing doing consulting work for clients We use commercial software to do that work and more recently last five six years I've been leading the supply chain design practice. So that's that's kind of where I came from and now Yeah, so you've done engagements both as a software company where you own the tools and you've also done as a consultancy Right. Is there any differences or what are the differences or similarities between those kind of engagements? You know, I think it's a little bit different. I mean, I think that the software companies certainly have a little bit more been to Training and getting users up and running. I think in a consulting firm our focus is more on the answers Let's get to a recommendation Let's take, you know, the 12 14 16 weeks and take it from raw data all those is a process to get to a recommendation I think most of the software comes even still today I mean the focus of software companies consulting is is getting that company up to speed which we do some of that too Okay, but generally we're in more in the mode of getting to a recommendation and a set of answers Okay, which do you prefer? I think it's actually that more rewarding to see the whole thing end-to-end and help, you know Build it from from data, you know, get the questions that they want answered through data through building models and getting all the way to an answer I think it's although the software When I was product manager, that was pretty rewarding. That was a lot of fun We did some neat design work in this in the tools and so I did find that the software development software design part So let's talk about that for a second. So I don't know if you People watching it know what a product manager does and the software coming Can you describe that because that might be a role that people might be interested in? Yeah, it's it's great Actually, it's a nice Intersection of the business requirements and user requirements and the technology So I wasn't the one writing code But I was the one who understood the space and what the requirements were of the users and the customers And I would we would translate that into the needs of the product. So how do you design it? What do we need to do priorities or prioritize the releases prioritize the next things that we should put in and then We had a team of broad engineers that would then translate that into kind of pseudo code And then developers would actually do the development. Which do you find more? Challenging working with customers to get the requirement. I'm just turning to something or working with developers convincing them of what the customers want Boy, the developers were great. They really were back in those days that it was it was both were a lot of fun Actually, I really enjoyed both sides you got it was a good really good mix of interacting with customers understand the requirements and Going into the hardcore data model and technical side and thinking and also just thinking about the user first How do you make something that's? Intuitive yeah, that's a big part of it too is there's a lot of software out there that is you know less than intuitive and That's that's a big challenge. What would a user expect? What's weird? What would make sense? What's logical whenever when I was doing this I found it almost like you're translating because having a developer talk to a customer was Never worked out, right? Yeah, they speak different languages have different world views So you kind of have to wear a different hat when you talk and bridge between the two absolutely Yeah, but now you're chain elitics. Yep, which is a software agnostic, right? You don't develop software Correct anymore. So tell us a little bit about chain elitics. So chain elitics is a consulting firm We focus I think there may be a slide that has a company Company overview on it, but we focus in a few different practice areas. We're about 200 people We started out focused exclusively actually on supply chain design And we recognize the need to branch out from there and we've gone into transportation inventory planning Operations and warehouse layout and design Packaging optimization right inventory planning several different practice areas now our roots are in supply chain design That's what we started the firm to do and that's still our one of our largest practice areas today we've got about 200 people overall about 40 of us are focused on supply chain design and It's it's been good growth. We've grown from two or three people In at the end of a one early oh two to about 200 overall today And we've got offices around the world three in North America three in Europe and three in the age Pacific That's great. Now is your team you're running the network design team Are they all in Atlanta is your team dispersed as well? Absolutely dispersed. We've got a handful in Europe a handful in the age Pacific We've got a Number in Atlanta. We also have folks remote office and in our other offices around the United States. So it's it's pretty It's pretty global. Yeah so You probably how many engagements do you think you've run or supervised? I mean just personally or say or that you your team has Yeah, that's a good question. I mean we probably at this point are doing about 40 large-scale a year And that's probably in growing over time. So gosh We're about 40 a year at this point. You know, you don't keep track of you Don't have a little tick mark on your behind your desk the number of engagements you've run No, I don't you get it now. I need to get a chair No a gold watch. So what do you see as so looking at all these engagements that you've run because they're essentially Kind of doing the same thing, you know, where do you put facilities? How do you flow product? What have you found to be the key drivers of success of an engagement or a failure on the other side? Sure, sure I think the biggest thing to think about in when we talk about success is what's the criteria for success? And I think of it in our team thinks of it as as confidence The whole idea of doing this work is to build confidence in in the answers and the recommendations and to do that You've got to build confidence in the data that goes in the the models themselves the people the team You've got to understand the business because at the end of the day You really have to generate confidence in the client organization and the client executive team and their whole team So that they are willing and able to implement the results because this is a bit different than a lot of say optimization work like think of a Routing application make a UPS driver get to the truck in the morning And he's got his route that was generated by an algorithm right overnight And he gets it and that algorithm wasn't probably checked or even seen by a person Maybe somebody look over it, but generally that is implemented immediately right they go and they do it right then Well, we're talking about doing is is strategic network designs. We're looking at infrastructure kinds of questions How many plants should there be where should they be? What do I make where how many warehouses should there be and just infrastructure kinds of questions and those don't Go writes implementation They have to go through a person a set of executives a set of leaders that have to believe it and buy into it and Become convinced to do it and to spend money so confidence and credibility is the whole thing if you have it even it's even more Complicated because you're dealing in the future. So back to your routing example The data that's fed in is actual data. I just delivery these and they find the best with known data You're doing something what is your typical time frame 5 10 usually looking out 3 to 5 years and the limiting factor on not always Sometimes it's further, but the limiting factor tends to be a company's ability to forecast their demand Okay, and they're a confidence level in their demand if you start modeling out 10 or 15 years And you put some models in front of executives. They don't believe those numbers They they're their horizon is much shorter than that. It tends to be 3 to 5 is about where most of the falls Okay, so but because so do you find that that's one of the biggest things you have to discuss the Having faith in your estimates because everything you're doing you have estimates of cost estimates of volume all those kind of things Yeah, so I think there's really two kinds of data that you got to have confidence in one is historical data So we gather very detailed transaction level data So things like order history and shipment history and production history and so usually the confidence in that is high Sometimes it's too high. Sometimes it's over confidence and though we have great data and sometimes we find out that it's not that great But that data usually has pretty high confidence level in it And so most of what we do is based upon that data and then we can take that and extrapolate it out to the future other data we call it design data, which is Data that didn't exist historically. So what would a new warehouse cost in Topeka? Or what would a new transportation lane cost between their plant and the new potential dc in Topeka or Those kind of costs we call that design data and that's a big part of what we do that's a big part of the confidence building as well and one of the big challenges is Getting that and getting that data right and unbiased So think about think about this example if you built a model of your current network Right and you populated the whole model with your actual transportation costs Let's say you put every lane, you know what every cost is and you put it in there Um, and then you needed to populate the new lanes that you don't ship on today and you just One idea would be to call a carrier and say, hey, what would you charge me to ship on this lane? And you would get Potentially if they would even respond you would get a rack rate You would just get a walk up type rate and you would put that in the model And it wouldn't be on a level playing field with your existing rates And so so it naturally favor it wouldn't favor the lane So does everyone make sure I know what a rack rate is it's kind of the retail price It's the price with no negotiate non Exactly So if you populate those potential new lanes with the non-negotiated rates and you push the optimization button It's simply going to reinforce the current state You're going to get results that say keep doing what you're doing do exactly what you're doing now And everybody gets pat on the back. Hey our network's already great. We're perfect. The problem is that you've not fairly Compared all the new lanes. And so that's a big part of our work is getting that Those estimators, whether it's for freight or for warehousing other costs fair and unbiased so that we're really testing it I'm surprised that go does it typically go that way that the new things are come in as more expensive or they come as Underpriced where people say oh, I will save Something we'll get better deals because then that would favor the new solution I think it's more typical that they would they don't know in freight I think it's more typical they would go out to a carrier and a carrier if they like said if they would even respond They're just going to give the retail or rack rate. And so I think in a freight perspective It's it's likely to be overstated. And so, you know in warehousing it could vary You might pull your warehouse person to hey, what would it what would a warehouse cost over here? And then you don't know then you do you might have bias you might not and you just don't know Um, so we've done a few things to combat that. I mean, I think one of the things you're familiar with we we developed this freight market intelligence consortium We call it the fmi c chris has been instrumental since its beginning But it's a basically it's a consortium of shippers There's about 150 members in north america and they give us their actual transportation spend data multiple times a year And we take all that data shipment by shipment level data of what actually happened And we put it in a big econometric model a big regression model And we can then predict what the transportation costs are actually from Anywhere to anywhere across north america for the north american one We do it in europe and some other regions as well But the point is that then you can actually get an unbiased view of what the transportation costs are For lanes that you don't operate today Now do you when you use that if someone has existing rates on some lanes? Do you use their existing rates and then the fmi c rates on other lanes or do you use fmi c across the board? So what we do is we take a sampling of lanes and we see what their position to market is right So if they're five percent below market for example below what the fmi c market rate is Then we apply that five percent rule everywhere and then we use the fmi c lanes everywhere to make sure that it's unbiased now some companies might buy transportation You know regionally and you might have a different discount in a certain region versus another We take that into account and obviously different modes there might be a different Factor for dry van versus flatbed or refrigerated, but we'll get kind of specific to it But we prefer I think it's best and it's most at least bias to use the fmi c rates across the cross that makes sense because it's apples to apples Exactly. Um, so we talked about a little bit about the technology talked about the data The math underlying everyone who's taken the course now. Hopefully you've gotten very comfortable Using mixed-injury linear programs because at the heart of it That's really at every software. It's still using a mixed-injury linear program. And then there's the people which do you think is the most uh The the the hardest to manage which thing leads to the most difficulty Is it the the data the mathematical optimization the software the people or something else? It's it's the people but it's um You think about work that we do we tend to work in pretty large complex organizations and so What happens is that you need a lot of people Cross-functionally to participate in this process because it's not just something that affects transportation or warehousing or manufacturing It affects finance it affects customer service. It affects sales. It affects the future of the company And so you really need a good cross-section of people So even if it's one business unit in one geography, that's still a handful of people Then you take a complex business that has multiple business units and multiple geographies And there might be a finance person or a free transportation person for each geography and each business unit and all of a sudden the number of people you know actress starts spiking And keeping them all on board and continuing to build that confidence like I talked about right that's that becomes hard That's the challenge is keeping all those people in lockstep right on board and believing what we're doing So how did you learn that because your your background was george tech right? I did that's all math and i'm sure you didn't take any courses on change management So how did you develop that skill because that's something that we don't teach as much we talk about it How did you develop that and is that something you just naturally born with that's a good question That's something that we've evolved over time. I mean, I think you know to be Transparent I mean most of us the chain elitics started out and our core background is optimization and math and things like that but we have some Some folks from you know broader consulting firms that have some project management experience And I think but it sort of has been on the job training over the years And I think when we started out most of the projects that we did tended to be Typically we're on the smaller scale one business unit or two within a company and project management was less critical Now many of our projects are global many of them are multi You know business units and you might have a client that wants to bring 50 people to a kickoff meeting And it's a it's a very different Different situation and you really have to take in you know treat seriously the project management aspects of it so Of those four things over the last 20 years you've been doing this What's changed have the have people gotten smarter not less smart has that been easier the math the data What what has changed over the last 20 years in the engagements that you've run? Yeah, I think and before you respond back I just want to remind everyone you can enter in questions using slide out Yeah, just enter that in and I will moderate and push those out and you can we have a poll out there as well That you can participate and I just wanted to see how much experience everyone Who's in the audience how much they've had in terms of the engagement and it looks about Not much well most people have not which is good. That's why we're here Okay, so lots of things have changed I mean, I think clearly more's law is real like computers have gotten better software has gotten better The optimization engines themselves that underlie the software have gotten better And so we're building bigger and bigger models all the time Is that's the optimization or the ability of the database? Handle it or most is both it's both. I was talking more about the optimization But it really is the data system, too. I was going to mention that as well. It's the data systems are getting better um Everything we did 20 years ago is sort of Microsoft access, you know now we're using sequel and tableau and python and in different, you know Alteryx and data guru and things like that that are much more Savvy in terms of data manipulation So the tools to manipulate data are definitely getting better the systems to house that data and just the quality of Data science in general like the people coming in and joining the team have more data Backgrounds than we did then certainly than I did and that any of us did when we started doing this work But it's good because there's a lot more data, too So we're dealing with larger data sets, right? I think that's some things that have changed some other things that have changed is the frequency um Frequency of frequency of doing this work doing the engagement. I think that when I started out it was really common I think I have a slide on this. Um, I believe that it's a slide It might be slide three if I look at the uh, not that one The next one. Yeah, so there's a um This is kind of what things looked like On the left side of this chart when we first started doing this work. It was common for This work to be done Maybe every two or three years and what you would do is you just recognize I have a need for supply chain design Something's out of alignment. Let's hire a consulting firm. Let's get some software. Let's do a big data collection effort Let's just do this large scale effort And then you get to an answer and then you spend the next two or three years Implementing that answer and kind of living with that network And so you got to the point where you just had to do it again and then two or three years You do it again and keep repeating and it's that cycle that's on that left graph there that That that's what it looked like big spike flat big spike flat So every time what you did that it's a massive data collection. It's a big data collection. It's a big effort It's a big deal and you don't really have any way to deal with questions that come up in the meantime Right, you don't have an ongoing model, you know, and so your network kind of atrophies That's what the bottom part of that the lower left part of that graph shows You're atrophying away from optimal the whole time And 20 years ago or in 15 or 10 like that was the way to do it. Everybody just did it that way Now what we're seeing is the frequency is increasing Like there's a need to do it on a more continuous basis because some of the key inputs are changing right customer service expectations are changing The world is you know fuel costs might change freight costs might change, you know things like How do I not get amazon or you know different customer service expectations are a big driver Mergers and acquisitions may come up So the idea is now much more about how do I build a model once there's that spike, right? But then how do I keep it going more of an inertia kind of some of the decisions aren't you don't change every year You're not going to Suddenly move a manufacturing plant. So do you have a question? Do you find when you do it more frequently? Do you have a constrained set of decision variables and changes that you do? There are certain things we would refer to more as different planning horizons for different kinds of decisions so if you You might have a longer planning horizon for an open close decision on a plan and a shorter planning horizon on a Different a crewing change on a line for example And so when we do this when we set up a model to be done Let's say quarterly right for a year certain decisions You're going to make every quarter and maybe once a year you plan to open things up more broadly and look at open Close decisions or restructuring decisions, but you never know in the middle of that year Merging acquisition target might come up Where you need to reopen everything and so you've got a single model that's supporting Some of those those different kinds of decisions And so at any point you could take some of the constraints off and answer some of the big strategic questions too now So we teach in this course mainly mixing your linear programming to come up with a deterministic solution But we talk a little bit about stochasticity randomness and the demand and other things But we've taught in previous courses se0x simulation So how do you handle or how does chain handle? Variability because you wonder some kind of sensitivity analysis, right? You're not going to be exact on that. Sure. How do you how do you approach that? I think there's a few different things we do one Is pure sensitivity testing meaning just changing some of the variables once you narrow down to a set of scenarios Start to make sense Changing some of the key variables some of the key inputs and seeing if your solution is still robust is one way to do it Right, so let me get to an answer Okay, that answer is eight warehouses right now How robust is that if I crank fuel up by 10 or 50 percent? What if I put transportation costs down 20 percent? What about labor? What about demand? And I can test it and the idea of course is to find a network that's robust under a wide variety of those conditions And usually that's possible with that sensitivity testing Sometimes we do get into simulation simulation We can take the result of an optimization or resulting network and put it in a simulation And see how like acid test it see how well it performs Exactly exactly you can set some initial conditions opening inventory levels that you do it requires a lot more data Because you've got to now know not just what the discrete number is But you've got to know probability probability distributions around some of the key variables like demand like lead times and things like that So it takes more time and more effort to do it But you can do a simulation that shows how robust a particular answer is And we do that from time to time. It's not what we do every time and it's kind of the exception But it it is it is a way to do it I think the other thing that answers some of the some of the uncertainty is doing it more frequently It's like if you're if you're going to make a plan that has to exist for three years That inherently has more risk than a plan that you are able to refresh every quarter or every month or at some more frequent Because you can respond now you can be flexible And some ways companies are doing that is building internal teams to do that on an ongoing basis And we help companies build internal teams in some ways some companies have chosen to outsource it to us We have an offering we call managed analytical services where We actually become The outsourcing partner to continue to do that kind of work because that's one of the big problems I know from companies perspective these big bang every three year kind of engagements They couldn't do them in-house because you can't keep someone there that you only use Every three years and if someone likes to do that stuff and they only get to do it Every three years they kind of go away. Absolutely. So have you what have you seen as the shift? Have you seen more going in-house or more going so there's three categories, right pure in-house Pure, you know outsourced and then this kind of mixed model that you see there's a mixed model I think I think what happens is Very very large companies that can justify and and invest significantly in a large team That they can take it in-house successfully And I think we've seen just the data over the years that you you really need when you design a team to think about Its longevity and its sustainability over time and some of the things that affect that are How many people do you have do you have critical mass because those people that you pick to do that work? They tend to be pretty good. They tend to be good at a lot of things Well versus different things and they get pulled in a lot of directions And so if you've got two smaller teams, you've got three or four people, let's say, right Let's say that's on the small scale and two of those people get promoted or they leave or something Well, then you've got two left and that's really fragile. And then if it's hard to backfill, right So teams of internal teams of less than three or four or so people Tend to be really fragile. Those people are going to potentially leave or get promoted They'll do good work maybe for a year or two But teams larger than that and some of the teams we work with it big customers might have 15 people You know, then they've got career path and mentoring And and and true sustainability like I don't mean sustainability from a from a greenhouse gas perspective perspective But I just mean from a sustaining the team and those can continue to do good work And even when those exist though We still are asked to come in from time to time when something Different happens so a large merger and acquisition comes in and and is different than what that team normally looks at or something That's different like reversal just something they don't typically deal with every day So we still are have relationships with these internal firms internal teams at large firms Um, but I'd say that you you ask the question about what I see more often I think it's more common still for companies to do it internally the big companies the big companies Versus outsourcing it on a on a continuous basis. However, we're seeing a big shift from that We've Launched this managed analytics service that we offer about two years ago And we've seen some significant adoption of that. So I think companies are realizing. Hey, this is something that I need to have I need to do on an ongoing basis, but keeping it internally is hard right and Retaining those experts is hard and if we can put that on a firm like chain analytics to retain it Then the combined team works pretty well. Let's talk about your team though It isn't because this has changed from being just a you know using what's best and throwing numbers in and pressing a button It's do you have one role or do you find that you specialize people on your team? Is someone really good at running the project? Someone's really good at the math. Someone's good at the data Do you specialize or do you expect someone to be good at all things? I see how do you do you specialize? You do you try to have the I think depending on the level I think it really dictates the role So we would have you know people that have been doing this for 15 or 20 years in a senior advisory role that look After the whole project right and then there's a project manager usually a project manager slash lead analyst that that That designs the models, you know executes the models takes care of the overall engagement and then One or more analysts underneath that tend to be more focused on data tend to be more focused on results reporting visualization those kinds of things And the team can grow in different dimensions from there But certain people and team might have more expertise in a particular industry that might help You specialize that way or I mean are there are there more similarities between industries than differences or the other way around There's more similarities in when we're doing supply chain design The reality is that if you're doing a supply chain design model It's pretty similar from one institute to the next But like we talked about before building that credibility that confidence It certainly does help to have a decent experience and talk their language and use their terminology and have Have be well versed So if we do have the option of putting somebody in a project that has Experience in oil and gas or experience in cpg or that's better Right right from the customer's perspective from everybody's perspective with the credibility. Yeah, exactly. It's better Certainly from the customer's perspective, but if the goal is to build credibility and have confidence It's better that somebody has that background But not you ever find the case where someone from a different industry comes in and they have an insight Because sometimes what we've done is can cross industries get get siloed and they think a certain way because we do a certain thing And sometimes they do something different they from different industries Yeah, we very rarely have people who focus only on an industry. There are people who have more oil and gas or more pharma or more cpg or retail But no one really it's not exclusive. Got it. And so no, I think the cross pollination is important. Definitely Yeah, so we have a question and I encourage you guys to keep putting questions in And I'll get to those as they as they come in VJ asks if there's any advice for people starting a career in supply chain management For example, which type of positions will be better for building a foundation than others? Well, what's your recommendation? You know, you mentor a lot of new buyers, you know, it's funny. I actually think that Having an operations background is critical I actually interviewed a person a couple of years ago and there was a gap in the resume and I said What happened during this gap and the person said well, I just worked in a warehouse I said just crazy like that's the That should be front and center the fact that you have operation experience You can talk that language that you've lived that is critical. That's important. Like let's not You know, let's not, you know downgrade that that's huge. So any kind of operations experience, whether it's in a warehouse or You know managing I managed a rail fleet early in my career and you know, some of those skills every day, don't you? Well, you know, it helps because you you have a good sense of just what you know When somebody tells me something is 100 tons I go. Oh, that's a real car. You know, I have this kind of background That's interesting and I have a kind of a frame of reference of just how things actually are and how they work That you can't just get from doing models And I you know, I would encourage even when we do a project now We still always go out and see a facility Tour a site get a good handle on what things really are in the real world You know before we come back to the ivory tower and you know build the models, right? It's critical to get that hands-on experience. So I would say one thing would be to get operations experience um I think you know, certain roles are going to have more breath than others I mean, I think consulting is a good way to get breath across a lot of industries um But I I think the those of us that have sort of different kinds of operations background That's probably one of the things that might A student might least expect the answer to be okay because I know it's it's easier to To learn your math and your operation and well the math stuff early and eventually learn the people's skills But the practical experience is actually easier to do that that front line experience earlier in your career Right, and I know for a lot of companies We have a lot of students who graduate and they go straight to consulting And if they want to come back to a company and take a like a senior VP role If they don't have that operations experience if they don't know how to Talk to people in a warehouse or on the front line They're at a disadvantage and so getting that real experience. I think is something really helpful. Absolutely. Yeah um Hey, we have another discussion another question This is from carolina in one of my classes at the university. We were discussing cooperation between firms in the same market Is it more about increasing profits or exchanging information? I don't know if I understand the question So in the same market meeting competitors um Yes between firms in the same market So have you ever done an engagement with two competitors that do a shared facility? Because I know in my transportation experience. I've done this where you have two competitors But they do a joint bid so let me transportation So some things that happen in certain industries are things like swaps and in cements or chemicals or things like that Uh, if you have a competitor of yours makes a product near one of your customers And you make a product near one of their customers you can do a swap and avoid the transportation So there are those kind of deals and there it's up and up. I mean you Those are things that we do model and can model Um, you know, there's a fine line when you start talking about modeling competitors together You get to collusion and things like that. So most most of the time we don't I think there's certain things that come close to that Would be Modeling a whole end-to-end network where you model a customer or you model a supplier like that. We do pretty Oh, you go up and down. Yeah up and down the supply chain. That's pretty common The other thing that we do That's surprising to people sometimes is that we'll build models of your competitors Ah, I think but without their involvement, right, right? So we take what a company knows about their competitors already through whatever information that they have at their disposal or public information on the internet or something No industrial espionage. No, we don't do any dumpster diving or anything like that But if our client can tell us like right here's where our competitors locations are or here's where google says that they are We can build a model of the competitors and then we can use that model to say well Here's what we think the customer service that they're providing is to these markets or what their landing cost likely is with our freight estimates And you can actually build models and you can even ask models. Well, who would be what's the lowest cost? Who would be the lowest cost vendor into these markets and then where would I need to put a facility? In order to be the low cost. That's really interesting. I hadn't thought of that. Um So you've done all these hundreds not quite thousands but hundreds of these engagements and oftentimes We learn from the bad experiences. So out of all of your engagements Can you name or talk about the most the biggest disaster that you had the worst engagement that you had? You know, I think some of the real challenging ones are when um When there's just a large organization or a merger and acquisition for example We're where there's just so many stakeholders and they're not all on board with what we're doing Right, you can imagine a very large merger and acquisition if you've just gone through a merger and acquisition There's a lot of things going on there. There's fear. There's concern about the future people are worried about their jobs You don't know who's going to be in charge Sometimes we're involved in those prior to announcement as well So we may be in a clean room environment that we to to be the broker of data between the two sides That aren't allowed to see each other's data yet. Sometimes we're involved just after you know announcements Um, but in those situations, there's a lot going on besides just building a model There's a lot of people and a lot of different factions and people Who have different opinions about the how it should integrate how it should not integrate? And so that whole people dynamic is Is huge and getting ahead of that and sometimes particularly if it's if we're only talking to a small subset of executives at the beginning Right, you go in with one of you know thought about how this is going to go And all of a sudden you get into some political Scenarios where you know entire divisions don't want to take part and so that that's probably the most challenging environment We get into it has nothing to do with modeling and it has all to do with just people Is that is that You're talking from experience that I am this has happened What what happens? Do you do you have engagements where you just have to pick up the pieces and move forward? Or do you walk away to something? So, I mean in those situations, uh, I don't I don't think we've walked away I think we've just we just pushed through it and uh, and what you know, I think the best way around it is to Try to just build up that credibility And just say look, you know, we're on your side. We want to make the best solution for the whole thing We're trying to Put together this optimization program to create the best and strongest company at the you know at the end of the day And you know, but it just it's just more work and effort and a different kind of work and effort than you were expecting Right go in to do what about the flip side Can you name an engagement or talk about one where the Result or the conclusion was counterintuitive or something you just really didn't expect because a lot of the times You go in you kind of know what the answer is going to be at this point Don't you kind of know because you've done so many hundreds of these, you know, it's funny I think the most fun ones are when when you are wrong or when you don't know and I've kind of stopped guessing because it's It's it's hard. I mean sometimes if you stay within one industry another industry might look similar, but you know, there's a There's one I think about a lot. I was actually at a grocery store with my kids and we were having this conversation about the difference between Different kinds of products that appear to be the same from a consumer standpoint and and the example that I we talked about was You know detergent or a laundry product or bleach or something like that that sits on a shelf and then something like soft drinks so it's soft drinks sitting on a shelf and a Bottle of detergent or bleach sitting on a shelf from a consumer standpoint looks the same and it's a plastic container About the liquid in it. It's on a similar price point might be a little bit different but not remarkably different and So I was asking so what do you think and I think we have a question we have a polling question Go ahead. Yeah, so so we're asking if you have a manufacturer of laundry detergent Yeah, uh a png a coal gate a unilever and then you have a manufacturer of beverages So a Pepsi a Coke any of those which would have more dcs Do you think the laundry detergent manufacturer the about the same or the the beverage right more? So there's a poll for you to take to see what you think would have more And this is interesting. We probably should have asked it to like what degree difference. Yeah, but that's uh So it is interesting. So some people it's coming in about about Wow, there people coming in they're below people are changing their votes. You see though. We've only had to Oh, there's only people aren't voting. You need to vote from 20 to zero Um, wow, someone changed their vote. Come and change their vote. That was Sergio Right That's good. Okay. So it looks like We're getting more in now But it's saying that most people are saying that the detergent manufacturer is one of dcs. Yeah. Yeah, it's 80 20 So so that's actually right and it's and it was but what you might not think about is just the severity of the difference the so a company in the Let's say bleach or or laundry might have typically probably less than 10 in north america, let's say um Someone a large softening company would often have more than 400 And that's a very very different answer and both of those are mathematically right Like if you built both of those models and ran and you would get an answer that was close to those And so that's I find it fascinating that the numbers drive that and there's a lot of things that are going into that One is the the product velocity itself Just think about how many bottles of coke you buy versus how many bottles of of detergent I go through a lot of bleach steve So velocity is part of it Part of it is also just the model itself, right? Uh, softening companies typically are a direct store delivery model dsd. They're delivering directly to the store So they have to be close and Liquid in a bottle is a very expensive thing to move around right? It's mostly water It's heavy. It's expensive And so that has the effect of pushing you closer to the customer location And so if the customer location is everywhere Right everywhere. There's a retail store. Naturally, you're going to need to have more so you don't drive that And some of it is also well and just the counter to that is someone that's making laundry products or bleach or something Is typically not doing direct store delivery. They're shipping into The distribution centers of the retailer in typically full truckload quantities And so that's a cheaper motor transportation. You you can be further away. So the the numbers really do add up Uh, you know, if you look even just things like the fixed costs of the operation So a bottling plant for soft drink is is just a less expensive proposition than a A plant that makes detergent or bleach like a large kind of chemical type plant that makes these things So there are more bottlers from that standpoint as well So what's fascinating is if you put all those numbers together and build models You'll come out with very different solutions for things that appear, you know to be surface to be the same Yeah, so I think that's a neat little example that kind of highlights it But it's I think that's the fun part of doing the work is that Is you let the numbers lead let the data lead and the math leads you to something that you may not have thought was intuitive Let me ask you something a little more technical and then people are putting in questions And I'll get to those in a second But as you see the questions there you guys can upvote them or not So as you decide if you see a question on there that you'd like Um, just upvoted and I'll start from the most popular ones and work my way down So we'll get to that but let me ask a technical question One of the things the points that I make in the lectures that we do is when we do a supply chain network design It really doesn't look at inventory safety stock and all this and so How do you address that because a lot of people they use, you know, the square root rule I mean if you cut down my dc's i'm gonna have less inventory and it kind of Favors going just reducing the number of facilities. How do you deal with that? Yeah, it's a good question I don't think I would word it the same way that you did that we don't consider it I think almost every project that we do considers it. It's just how we consider it and and how much detail right consider it And safety stock in particular is a hard thing to model Um, most of these as you mentioned earlier, we're doing linear and mixed integer optimization and a safety stock behavior doesn't really lend itself to A linear relationship or or even one that's easily to represent in a mixed integer world And so it really comes down to how important is the inventory to the questions that we're trying to answer So there's a whole spectrum of things that we can do on on the easy end of the spectrum would be You know as you mentioned just ignore it right if something doesn't matter if what we're testing isn't going to affect inventory You could ignore that's almost never the case, but it's possible to ignore it That would be the extreme end of the spectrum The second step in would be do some just pure You know textbook calculations for safety stock Let's take you know, let's run the math and let's take the demand variability and the other variables that go into Safety stock calculation just purely calculated in a spreadsheet or something The middle of the road is what you described Which is a power curve or the subset of power curve of a square root function I mean when we do that we're a little more advanced than square root. We actually take actual data and Do a nonlinear regression to try to figure out what the curve looks like rather than the square root function So we talk about that and so I show how to do that because I follow the same kind of thing Where you look at what they actually do in performance and adjust it great and and that Is not the end all right You could take one step further and and actually run a true inventory optimization We do that often too. Okay, and so then you're so you do it like sequential you do the network iterative Right, right. So it really just depends on how important the the inventory component is to the question that you're answering And so it is important when even if we're talking about a simple Number of distribution centers kind of question. How many should you have? Safety stock is important, right? If I stock inventory in one location I'm going to hold less in general than if I have to spread it out into 10 locations And so that's a true that's true almost in every case And we need to be able to represent that now Do we have to represent it down and make recommendations for this particular skew and what the inventory policies should be for that skew? Maybe not if the answer is just generally how many where should I have and they're all going to be full stocking In that sense a square root or or maybe a bit better than square root of a more general out of power curve Might be sufficient. Okay, but in other cases, it's not if I'm moving around You know high tech or high value products or you know Something that we're inventory really really matters and that's the driving force And I really do want to know what my stocking policy ought to be At this location and it's different than that location Then and and I want to meet this exact service level. I want to hit 98 percent You know service and I want to set the policy for this Well, then inventory optimization is a great way to do that So there's a whole spectrum of choices and depending on what we're trying to accomplish We're trying to find the right one how detailed you go into different different forms of analysis Yeah, all right. Let me ask some questions from you guys I'm going to start with a question from me zheng and I'm starting with may because she submitted A problem from practice for se 0x or se 1x if I'm not mistaken From her work at 10 buck 2 which makes handbags and naps x So her question is we're just a team of four people in our operations department at tim buck 2 I'm surprised how big the teams you mentioned Do you think it's short sud for us to open a dc in a new country just based on the cost savings? Or do we need to analyze more? So it's kind of two questions in there Yes, okay So the first question is you're surprised by that by the number of people that are doing supply chain design work Well, when I was using the number of 10 to 15, that's for a very very large In a multinational corporation that's doing this around the world and and is basically kind of Mini-chain analytics, you know inside that company that there are, you know, many of those I mean, I think that that is probably the right size team for what they're doing It doesn't mean that every team needs to be 10 or 15 people But I think once you get smaller than a handful once you get smaller in that three to five range It just simply becomes difficult to sustain it Yeah, so no matter what size the company is right. It's just because you're just then exposed to people's career objectives and You know, I think in general if you're if you're inside a company and you're working on the same model all The time I do think there's a tendency for people to start looking around think wow I don't want to be pigeonholed into this role And so there's also that natural force that pushes people to look outside of that role Whereas in a consulting firm, we don't have that right? We're able to give a variety of project work and so keep people interested for 15 or something You can progress them through different stages of exactly stages and just a variety of kinds of models And in a in a medium-sized company It's just harder to do that because it's harder to keep the people interested in that They have a career aspirations as well and often those involve You know, if they're moving up in organization, then they're moving out of that out of that modeling role Whereas at a place like us or at a large team You can actually have a career path within that team, right? And that's just something that doesn't exist in a team of you know, less than five, let's say The second part of that question was Is it worth should they have more than one dc in a low cost? If they find low cost in another country, I'd be happy to put together a proposal and Come in and help me. There you go. I mean the reality is like we just talked about it It's very difficult to know of because You do really need to study it and look at it and That's a question we look at all the time is I'm entering a new market Should I distribute from my prior market or should I enter? You know should have a dc there and a lot of things go into that You know, we answer the question all the time about you know, border crossing Do I have a facility on the other side of the border to ship to it and have stock of inventory there So that my customers are not exposed to the uncertainty of the border crossing those kinds of questions are Are difficult to answer, you know from a high level description like that, but they're important ones to answer Right, right and in fc 3x we spend more time on global supply chains And what that means border crossings and different tariffs tax rates and things like that Yeah, um, let me ask another question from arthur. How important are your communications materials for conveying concepts to your clients? How do you translate the complex design models to shareable models? They're Critical there. I can't overstate how important they are. I mean, that's the whole game is when we talk about building confidence and credibility with the clients doing that Visualization is critical and I think I mentioned earlier that Visualization tools have gotten a lot better. We're using things like tableau and other, you know Uh, you know bi type tools and other data validation tools. I mean, I think, you know, we're seeing some Pretty impressive ones behind us here that might be a little above what we typically do Um, but it's critical to convey this complex concept of what's going on To the executives that then have enough confidence in it to implement it is Is is the whole game and so if I look at some of the things that have changed, I mean that that has gotten a lot better I mean, I think the tools that we use now are significantly Easier to use and produce better better than what's best better than what's best in my, you know lotus for docks Yes Yeah, um So here's a question from francisco. How do you identify a bad network design? Hmm. That's interesting. I mean that that's something that can you look at a network? And someone says here's my network. Can you do you have a sense you can say? Oh man, that sucks There certainly are some key metrics you can look at right I mean some of them are are you shipping a lot of long distance small shipment sizes, right? I mean think about you probably have covered this in the class But just think about how supply chain works You generally want to push things as far as you can in the largest shipment sizes and then at the last You know as late as possible, right make them smaller shipment sizes, right? Zone skipping and pipelines and ships and things like that all work on that principle And so if you look at a network, that's one of the things you can look at is um, you know We call them light and long are there a lot of light long shipments? Are you shifting lto 3000 miles across europe or across the country? If you are it's probably a better way to do that. That's a low mean first probably there's probably a good indication. You know, I think um Some companies too you just sort of know it They've grown through acquisition over the years have never really taken a holistic look at the network And you can just look at a map and say yeah, that's there's some definitely some opportunity there And just knowing from experience But there's some key metrics and I think there's some just intuition from doing it a lot too Okay, here's a question from chris. Um, what is a typical engagement like if there is such a thing? How clear the business is the business problem the math and or process problem the state of the data buying from internal Stakeholder stakeholders. Could you give a general sense? So what's a typical project looks like sure? I think in terms of time, um Usually the low end is somewhere around 10 or 12 weeks and the high end is somewhere 1820 and plus and we can easily fall outside of that range too with very very large projects or very very small projects But I think that that 10 to 12 to 1820 it's it's typically in that range Um, and what it looks like is there's data collect there's understanding the business Right, we need to come in and understand the business tour facilities see what's going on In parallel to that we're collecting data, you know the Shipment history and order history and production history and those kinds of data sets We're taking it through a pretty rigorous validation process to make sure it's right and make sure that we're understanding it And then from that we build a baseline model of how the network works today What it's what it's doing right now and we're able to then compare that to actual data so total transportation costs total warehousing costs we should tie out pretty well and then We decide on what scenarios we're going to run and we develop the design data that I talked about Which is the new lanes and newer houses and then we run a series of optimization runs Maybe it's 50 or 100 or so, you know different optimizations and sensitivities and then the final step is to Interpret the results and make recommendations and all of that is done hand in hand with the client organization This is not us disappearing, you know with the data and coming back with the magic answer It's very collaborative process and that's what it typically looks like and the the size of the team Uh, if that was part of the question the lowest that it would typically would be from our side would be about two and a half or so people Um, maybe somebody who's less than full time in a senior advisory role and then two full-time people in a Lead analyst slash project manager in analyst mode and it can size up from there I mean some of the really large ones that are on the other end of the curve Would be things like global if you think about some of the things that make a project bigger Uh number of business units, right number of regions things that are global Um, it really is helpful to be smart about what's in scope as well And if you think about There are problems that are truly global and there are also supply chains that seem to be global that are not really global Right. Um, I'll pull soft drinks example again Uh, you can go anywhere in the world and get a particular soft drink that's based in Atlanta, right? You can have it anywhere That doesn't mean it's a global supply chain It kind of seems like a global supply chain because it's everywhere, but it's really not It's a multinational supply chain, right? They're not containers of soft drinks You know 12 packs of soft drinks in ocean containers being shipped around the world. They don't do that They make it regionally for some of the same reasons we just talked about and so That's not a global supply chain at least the finished good aspect of it and the surre part might be but the Other end of the spectrum would be truly global, right something like high tech, you know a smartphone or a laptop Uh chemicals also it's very common for a chemical company to make You know sub product one in germany sub product two in singapore bring them together and use them blend them and make one Product and ship it around the world. That's a truly global supply chain You do need to model the whole thing Have you tried to take a company that does have global supply chains as you defined it and tried to convince them That maybe they could shift to a local So it's almost changed to their focus or do do they tend to not change Stripes I think they tend to not change because in those two examples I gave, right I mean with very rare exception smartphones are made In shenzhen, right? I mean, I think there's some exceptions in brazil and for some local laws and things But but generally we're not going to change that the economics are such that It's just where it's going to be done because what i'm thinking is automotive because I know there's some automotive Where they have focused factories and the fact all transmissions are made here All you know that some of these are made here versus You shift and have more General because one thing I also talked about in the lectures is chaining, you know So you make sure that a plant is a product or an assembly is made in at least two plants So you have redundancy and you can right right, so i'm curious if that comes out much automotive is interesting because that one You know we did some work. It's been many many years ago But you know if you look at some of the decisions that they make They that they have to make around where to produce a particular car, right? You'll notice that many of the larger vehicles like SUVs tend to be made in the united states even Even though the parent company is is not it's not a u.s. Company There's a lot of forces for that one. I think the demand tends to be greater for the markets here the market is here Um, but transportation costs are also higher for those larger vehicles. You're going to get fewer of them on a ship I mean there are some transportation components as well and tooling up multiple lines to make the same Product is a very expensive thing to do So those are definitely things that we could model and certainly those decisions are made with these types of models Makes sense So one of the things you mentioned earlier is that the speed of optimization has gotten better You and then and you can do more runs faster It's almost a double-edged sword Do you find a because initially if it took a day to do a run you do five runs and you say, okay We make a decision. Do you find the number of different runs scenarios? Is it just exploding? Or because everyone wants to do a what if is that right? So how how has that changed over time? That's interesting. I think the Yeah, you say it's gotten faster, but we also just make bigger more complex models It's a kind of so so has the runtime stayed constant I think uh, I think in general yes, but obviously if you solve the same problem now If you should then it would be remarkably faster, but then you were making more concessions You were making you were giving up more and now you don't have to give up as much you still have to give up some There's still room for improvement. I mean software people out there don't stop, right? It's it's important like we still need to make more progress in it But we made a lot if you think about just the shape of morse law that kind of thing the software is getting better Hardware is getting better. Right. I mean we're taking advantage of but we are building more complex models So we're doing things today that we wouldn't have thought of doing, you know years and years ago When we first I mean you think about the beginnings of supply chain design. It was called network Modeling or network optimization. It almost always referred just to how many warehouses should I have right? That's still a question we answer but but more often today the question is how many manufacturing lines should I have? Where should they be and what should they make right? I'm using the same kind of modeling tools Same kind of underlying math, but I'm answering a far more complex question Um But I'm also answering that that warehouse question at the same time at the same problem So the the models have gotten bigger because we're doing more in them So I don't I wouldn't say that the general 10 trend is that now we can do 100 runs were before we only do 10 There's some of that for the similar problem, but more often than not I think we're just taking on more you can solve a bigger problem. You can solve a more complex problem Yeah, that's generally what we do and that's probably not what everyone does, but that's probably our tendency It's just to put more in there any we're just about out of time We have three or four more minutes any last words of advice you want to give Our young learners our students out there. Yeah, I mean I think what we talked about before of having operations background In this is critical. I think you know don't downplay that. I think that is Is very important I mean a lot of us that have been doing this long time have a background in one industry or another or working actually You know in In transportation or in warehousing or having something like that and and that's critical to build that credibility up and confidence And being able to talk about it. I I wouldn't shortcut that I think that's that's definitely an important one Um, and I I think some of the things you brought up too that it is a people business You can't sadly you can't just do your models do it in the basement, right? I mean, you know the modeler in the corner model doesn't work that well or somebody's sitting in the basement with a model They don't have the credibility the executives don't know that they're there the executives don't have confidence in them And it just doesn't work You have to be able to convey what you do and it could be through visualizations like this It could be just through standing up and talking But you have to be able to communicate what's going on because of the executives She doesn't know what's happening in in the model, right? You want to get to the end of one of these where the executive at the end of the table Looks at her team around that table and looks at transportation and and warehousing and And finance and says are we on board and and they want to look We want them to look back at her and not and say yeah, we're on board So it's that credibility building and it's you know a lot of that credibility is certainly technical But an awful lot of it is also Right the people and the process so she doesn't ask you about constraints specific non-linear Sometimes you got to be ready for anything All right. So um, thanks steve. I appreciate it. This was great We have more questions that we couldn't get to answer. We'll try to respond to those after the fact But I have a couple quick announcements from Sergio Remember the due date for week four great assignment is may 9th 1500 utc everything's done on wednesday at 1500 utc And remember this is different This is a proctored exam and we reason why we're making you do this as proctored Is because we want you to get used to using a proctored exam because this is what we use for the cfx When you come through and finish the micro masters So this is just one way to get you comfortable and familiar with it in a low risk environment Because the homework each week is only worth 2.5 of your total grade You have to use the soft the proctoring software the instructions are in there You have two hours to do it So it's a relatively a simpler assignment you get it done in two hours But the real reason we're doing it is to make sure you know How do you use the proctoring software and it works on your computer and you understand how it all works Or you're not surprised when you take the cfx and it's all being proctored Second announcement a midterm exam opens may 9th again at 1500 utc It's available for one week. Once you start you have four hours It's pretty standard of you guys this should be your third course So hopefully you're used to this format by now and this will cover everything in weeks one two three and four But what we've discussed today will not be covered. You won't be quizzed on steve's knowledge That's too bad. So with that I'll close it out again. Thank you very much. Thank you. Appreciate it here. Appreciate you. Thanks. Bye