 Hello Global Supply Chainers and welcome to our third and last live event from SC0X. We are here today live from our cave lab at MIT. I have today with me Chris Keplis, I have today with us Chris Keplis, he is our Executive Director from the Ash Center for Transportation and Logistics and of course you are from SC0X. Thank you very much for joining us today. I'm glad to be here so you might be having an echo now. I'm sorry about that. Glad to be here. So thank you very much for being here. Our agenda for today is a very clear cut. We're going to do a recap of SC0X and Chris is going to provide some context of SC0X and the other courses of all my classes for them come together. And afterwards we're going to spend time on the problems for practice that our learners have supplied with some pretty cool problems that cover a wide range of tools and methods that we have discussed here yet and we're going to dive into some of them and discuss. And finally we do a quick recap of all the courses that are set for the next 60 minutes. So before we start I wanted to start one more course where we want to see which of the methods that we have supplied as our supply chain toolkit in SC0X are the most interesting for you two while we go through our recap just go down there and see where we end up. So Chris, want to start with the recap? Sure. But also for slide out you have the ability to enter questions as well. So enter those questions and try to answer them best thing to get them in early and we'll try to answer them all but don't wait for the last one. Alright SC0X, this is probably the first class you've taken most of you. So take something out of order but hopefully you'll get your first 4A and it's the most intensive course we have. So it's easier from here. We hit you up front with all the math early. Not sure if that's the best but that's the way we do it here. So you have all the tools. When you hit the other courses 1x, 2x, 3x, 4x they're equipped. So there are really 4 main things I hopefully you took away and learned from this course. And the first is one that we didn't talk about. It's kind of like a Trojan bonus. It's a mathematical next area because everything we do in supply today is based on some kind of a model. It might be a simple model. An analytical model might be a very complex optimization model but they all require functions. And so hopefully you got very comfortable with y equals f sub x. y equals some function with some independent variable x. And it might be x squared y plus a plus bx. Something like this. Hopefully you got comfortable manipulating algebra. And it seems trivial if you got it just coming from school. Even working in the industry for the last 10, 15 meters it gets a little rusty. So the real reason why we do all this math up front is to make sure you get that dexterity back. So you're comfortable hopefully with how to use functions, different forms of functions and also approximations. These are some of the big things for mathematical dexterity. Then we went into three buckets of tools. First tool bucket is optimization. It's been the first half of the course pretty much focused on this. We started with calculus, which is really just how you solve an unconstrained optimization problem. You want to find what sounds an extreme solution. It's a minimum or a maximum or an inflection point. Which we don't really cover too much. But generalizing your minimizing cost, maximizing profit or maximizing revenue. That's all you're going to see. And the big takeaway there is the first or second order conditions. Proving your extreme point in the second order tells you whether it's x globally or min globally. So that was the big takeaway. And you'll use that in SC1S because the fundamental inventory model that you'll learn very well and you'll tattoo on somewhere in your body an economic order point in your EOQ. EOQ is simply first order conditions for a maximum or a minimum cost of how much inventory you store. So that's starting off with optimization. We started unconstrained and then we added constraints. So it became constrained optimization. And this is where we went into, we talked a little bit in some extra things about ranging, or really focused on linear programming. And this is where you have a bunch of constraints. You have an objective function and you have to find the maximum or the minimum value subject to these different constraints. So you'll have the form bin, z equals f of x, whatever the different x's are. And you're subject to, maybe you only have a certain amount of supply. You have to be a certain demand. Your x's have to be greater than zero. We spent a lot of time on that. We showed it to you graphically. You should be very comfortable using linear programming. And we spent more time on mixed integer linear programming. That's where some of your variables, instead of being continuous, are integer and more specifically, binary. And you'll see this a lot later on in some of the courses because for supply chains, a lot of your decisions are, do I open this facility or not? And so the mixed integer linear program will milk, allows you to make those decisions in conjunction with how much should I flow? So you'll have both continuous and integer and binary variables all together. And the mixed integer linear program is a great tool. So everything we did in optimization is known as prescriptive. So these are things that will tell you a solution. It'll say what the x should be to maximize or minimize. It tells you what the answer is. That's why you're using those. The next bucket of tools, that will make your mixed probability and statistics. And these are more descriptive. They're describing what is going on. And so we talked about different distributions. The two biggies we hit were normal distribution, which you should understand the normal distribution backwards and forwards. It's used widely throughout supply chain. Maybe it shouldn't be, but it is. It's a general assumption, right? It's a general assumption that you're going to have a bell curve. So you should be really comfortable understanding that throughout, especially when we talk about inventory. But we also talked about the facade, which is more a discrete distribution and you see that in slow moving inventory. And then I also talked about my favorite distribution, the triangle distribution, which is a nice one for approximations, where you know minimum, a maximum, and a mo, the most common value. From that, you can do really quick probability distribution. So into that, then we characterized data and the two big things for characterizing a dataset, which you're all going to have to be experts on, because data is massive and ubiquitous. It's everywhere now. And so you have to think about the central tendency and then the spread, the dispersion. Those are the two key things you want to figure out. What is the average and what is the standard deviation or the range or the inner port idols? So you want to have those two ways of characterizing the data. And then we spend a lot of time on hypothesis testing. We spend way too much time on chi-square tests and details with chi-square tests and all the infinite little details, p-values and things like that. If you're interested in that stuff, it is a whole domain. Same thing with optimization, but statistics is like its own world. So there's a lot of depth you can go into there. I wanted to teach you enough the amount that you'll see in practice, actually. So do many companies do actual hypothesis testing? Not really. But they'll look at the distribution. They'll want to make, if you do a formal test or like investing something, and see does it do an improvement, you need to know how to do that framework. But it is not used, in my opinion, as much as optimization. Because again, those tools are more descriptive than describing what's going on. So we talked about optimization, probability and statistics. In the last or the third set of tools is regression and simulation. Regression is all about its prediction. And so you're trying to predict what the dependent variable y would be given a new set of x's. So we have y equals beta naught plus beta 1x plus beta 2x, so forth. And you have a bunch of data and you're calibrating the betas. You're coming up with an equation that you can then formulate or estimate a new y, a new dependent variable, based on new independent variables. And we use this for a bunch of different things, and this is widely used actually in practice to understand what the effect of something is. I use it a lot for transportation. You can model transportation rates by looking at actual rates and deriving what that function looks like. It's pretty widely used. In fact, it's used a lot to develop what the relations are going to be to feed into a simulation. The simulation is simply where you're simulating future behavior taking one of those probability distributions. So if you assume demand is normally distributed, you can simulate a bunch of days of dance to really see how your system operates. It's really good at understanding a really complex system. We'll talk more about that in later courses, but if you have a system that's simple enough where I can solve it analytically, like that economic order quantity that I talked about, where you want to find the optimal quantity to order to minimize my total cost, then I don't need to simulate. Simulations for those problems where it's too complicated to model exactly analytically. If you have something with many players going on setting up a simulation, even in a spreadsheet, it's sometimes very helpful and it helps you give an answer of how the system will behave. Now, one of the things that people always have a hard time grasping is that a simulation will never give you an answer. It'll tell you, set up, given these policies, this is what happens. It'll never tell you the best. That's what optimization is for. That's a prescriptive set of tools. Regression and simulation are more predictive. Given that you do these things, this is your outcome. It'll never tell you what to do to have the best outcome. So simulation and optimization are different types of tools. In fact, there are some packages, like Lomisol, that are robustness. These are two sets of tools. They're not the same thing. You never use simulation to find an optimal solution. How are you going to use these tools? The rest of the courses will SC1X. You're going to focus in on forecasting demand, managing inventory, and planning transportation. Here, it's all about statistics. You'll use a lot of regression for demand and planning. For managing inventory, to minimize my cost, we're actually going to introduce something called a news vendor problem, where you have a single period. In that, you're actually maximizing the profit based on the probability distribution for the demand. The probability distribution for the demand. So it kind of combines the two. How do I maximize something, optimize something under uncertainty? We'll talk more about that later. For planning, we use a lot of what's known as total cost equations. It's just a function of all the costs for all your activities. So we try to minimize the cost or maximize the profit. So a lot of these, you're just creating very simple mathematical models and even minimizing or maximizing them. In SC2X, it's all about design. You're going to use mixingers or linear programs extensively. You'd use them to set up a network design, for production planning, for commentatorial options for procurement. And then in SC3X, we do a lot of simulations and it's a little more complicated than what we covered here, because we have what's known as a complex system. So we've gotten the system dynamics, but it builds on what we did for simulation. We'll show you a different type of simulation in SC3X. And we also get into risk contracts. And this allows you to use both optimization and probability. The question there is, how do I share risk in a supply chain? So it uses these fundamental tools that you've mastered in a much more complicated setting. And then in SC4X, we kind of take it to a step, a next step. Because when we talk about data in this course, we might give you 100 lines. These are little baby data sets. And we do that so you learn the concepts. In SC4X, we give you hundreds of thousands of data records. Because in a real situation for a company, you don't get baby data sets anymore. You get massive amounts of data. And this can be from point of same information. This might be GPS Pings. If you're doing transportation, it can be the signals from a machine. If you have Internet of Things. So the amount of data is increasing. So we teach you how to do that, how to create databases, and then how to use machine learning. And so what we do in SC4X is take everything we've learned so far and show how to scale. So all these techniques will be used and we'll add more to them. And so that's what we've done in SC0X. And that's kind of how we will use it for the rest of the course. It's still our toolbox for supply chain management. We have that toolbox now with us and can use it to solve specific supply chain management problems. Shall we quickly take a look at our poll on different tools? And I think the learners have clearly voted they get to do two votes each. Yeah, so they get two votes each and I think simulation and constrained optimization are clearly their preferred parts that have them impacted most. I'm too bad that you guys don't like approximations. Approximations are it's a lost tool. Did you have to learn a slide rule? Yes, we were in engineering school. I still have my grandfather's slide rule. It's an 18 inch slide rule in a leather case. And you can move that around. The beautiful thing about it, it forces you to approximate. And so it's a lost skill. So whenever you do one of your problems for something, you should always think what will my answer be in order of magnitude. It's such a good skill to have. And it's kind of interesting that compared to the first round, first life event where we asked essentially the similar question of what you are looking forward to the most, algorithms and approximations were more liked, but still some simulation, constrained optimization were those that they were looking for the most. And we snuck in hypothesis testing. Yes. It's not really interesting. Someone is three percent. All right. So the good news is you'll see a lot of optimization. Beyond a simulation, we do some. We don't use it as much for other parts of the courses. Because it gets for a large problem, it gets very specialized. The software that we use for here, it's very intensive to be honest. So we are more heavy users of constrained optimization and some regression and everything else is pretty even. Okay. We want to look at some questions or let me just quickly cover the one thing that is still ahead of you, which is the final exam. That little thing. Right? We're going to launch the final exam tomorrow at 1500 UTC. It's going to be open for just one week until 4th of July. And it's going to be again a timed exam. So essentially it's the same setup that you have already seen in the midterm. So you have four hours to solve four problems, three mathematical and one checkbox. And it covers all eight weeks, right? And it's going to cover all eight weeks. So from the first day that we have presented content it goes on to what we have seen in the last week. And so would you say the final exam is comparable in format level and difficulty to the problems on the midterm? Yes. So that's our objective, right? So we want to cover what we have seen, we want to test whether you have learned the contents and it's going to be similar to what you've seen in the midterm. One last question that came in, I'm just looking at the question that came in for this. Is there any practice final exam even from a different website or resource we could work through? I don't think we have one. But you know what, here's what we do. We come up every time the exam is new. We come up with new questions but they might you never throw a problem away. They become practice problems. And so look at the practice problems and they're going to be very similar to the types of problems. You should be comfortable with what we've been asking. And you can go through the practice problems again, right? Just work those again if you feel that you have to look at them and you will have a good set of training problems to work through. So one last question about the exam. I'm just hitting the exam questions first. Being our first exposure to a final exam we would like to know if there are any specific insights and what we should be concentrating on for better scores. So that's something you have to take. Do better. I kind of talked about what we'll be doing and you know the areas that you're going to see. It's going to be very similar to the approach of the midterm. Absolutely. And in terms of content I think you have given some good overview of what is important to us. We think about the different concepts of why we think they are important in terms of supply chain toolbox and I think that gives you a good idea of what you can expect we are focusing on in the exam. We're not going to ask some weird side section we're focusing on what's important to us. I can answer some other questions that are along these lines. Is that okay? Sure. So someone asked is our practice problems are our practice problem scores accountable for the MicroMasters credentials? No. No, they're practice problems. We don't care. We don't look at them. You get no points. Hopefully you do them. They are for your practice. The only stuff you get points and credit for is doing the GAs the greatest assignments each week. And do the midterm and that's already your first point collection and then there is the final exam that provides the largest share of points. Right. At the midterm we had two attempts. At the final exam is it one or the same as the midterm? Again, two attempts. We don't change the format. You're trained on a certain format we've tried to educate you on that form we're not going to confuse you with different forms. The only difference will be when you take the CFX the comprehensive final exam and in that one it's a little different. We don't have the nice green check of happiness or the red X of despair we just tell you that we got the answer but for this for the final you have the checks and the X's so you're good. One thing that you should be clear on is when you start the exam because we have that four-hour time window so once you click and start the exam you need to continue and finish it because the timers runs and if you stop working on it if the four-hour timer hits or if the clock hits in four hours your exam is going to be automatically submitted. So you need to make sure that you have that window ready to work the exam. Don't start working on it two hours before it's due. Exactly. That's the second thing I wanted to make sure that you start latest at 1100 UDC because then you have exactly four hours until the final exam totally closes, right? Yes. All right. What do you want to do? Do you want to answer questions or do you want to go into the... So maybe we go to the life... Problem from practice? Problems from practice because I think you want to answer some questions about the future. Any questions? About the future because it fits content once. I'll do HemDeep's one. Do you accept students into the blended program who haven't worked as supply chain analysts? Yes. Prior experience in supply chain is not a requirement. I heard you need to know Python for SC3X or 4X which one? Neither. You don't need to know Python. I'm trying to think if we're starting to... We're finding more people know R or Python and they're doing more coding and to be honest our 10 month residential program and our blended program we're now teaching Python. We're teaching it in January. I don't know if you knew this. Sergio is going to teach the course of Python because we're finding that companies really want to know one people who can handle large pieces of data and that requires some coding language to kind of tie things together. So for our SCX courses we do not need to know Python. It's a good thing to know. It's a good thing to know and we're finding... It's one of the most flexible languages that you can apply to data sciences and I enjoy it very much. I like it. I saw something about the key concept document. Did you kill that? Okay. I think that's it for now. So let's move to our problems from practice then. All right. So we got a lot of problems from practice from our learners. We selected a couple and we'll see how we go through them. So our first problems from practice is submitted by Wisdom G.Dubb. I don't know how you pronounce it. I'll say Dubey. Dubey. It's a French touch and it's called a chicken farm. It's about a family owned chicken farm who sell chickens to their customers at $7 a piece. So it seems to be US based but it's a small entity and they're trying to grow. Currently they are doing 100 chicken in a batch and it takes them six weeks to bring them to the target weight. So that's essentially the production cycle. They buy 100 chickens that are one day old at a total of $105 or 105 per chicken and then they have additional cost like the different feeders that they need at different stages of the chicken's life cycle, antibiotics some electricity transportation cost salaries for the workers that's in the table as the additional cost that we have over that six week cycle to actually grow the chicken to a to the finish to the target weight and then they face currently 100 chickens for two weeks as demand. So 50 chicken week. It seems to be deterministic in this first exposure we'll discuss about that later and what they want is they want to sell their chickens fresh so it's one of their important objectives. So two weeks time frame they want to push out the chickens that have the target weight otherwise they lose their quality or so that is when they are slaughtered or they need to be fed continuously which then incurs additional cost and maybe also some weight problems or something like that that they get too big. I'm not an expert on chicken. So they're interested also in growing expanding their business seems to be working well in the last years as to wisdom's exposure but they are currently having a 600 chicken set limit of how they can produce that's like capacity constraint if you will and in the end they have a capacity constraint on also the on the refrigerator where they can only store for the chicken so it's kind of at the end they need to figure out when to slaughter them so that they don't throw them away but they don't need to feed them so there is some kind of dynamic at the end so the major questions that wisdom posed to us was well how many chickens should we actually produce especially given also the question of how you expand your business how to set the optimal price because currently they're working on a $7 piece question and that connects the pricing question connects to the inventory question because they need to figure out how to hold the inventory and whether to provide discounts and what is with salvage cost and something like that yes so this is kind of a rich problem I think and our already started voting on how to attack the problem with a pretty clear idea of well that seems to be constraint optimization setting yes but only very few of you have answered so far so I want to give you a few seconds because I know we have gosh we have at least 60 or so 60 we had in the first yeah so give you some time to answer those but the problem but what I love about this question is it could go in many different ways and so when I first read this I emailed back to wisdom to understand because the devil's in the details because if I had to solve this right now with someone with a gun to my head to figure it out I would say by 50 chickens a week right because the idea is every week I'm going to have a demand for 50 so if I produce if I bring more in right then I'm going to have to do something to them so the real basic choice when I first looked at it is do I want to buy 100 chickens every two weeks or 50 chickens every week and I can figure out what the difference there and that's what an economic order quantity that we teach in value but what we don't have here is a fixed cost so let's say I had a fixed cost that it cost me ten dollars plus a dollar and five cents per chicken right so if there was a fixed cost to make the buy then I would probably go to buying every two weeks to be honest 100 every two weeks instead of 50 every week the thing about doing 100 every two weeks at the end I've got 100 chickens that I've got to sell within two weeks and so some of them will get a little old right and so that's a challenge just looking at that way so first time I looked at this I said well it's either 50 a week or 100 a week you can think of it as a constrained optimization you know your objective function would be something along the lines of maximizing right seven which is a revenue the seven a piece minus and the cost I think it is a buck five plus five dollars and 19 cents I think if you add this up all the costs yeah so then so you're maximizing some number and that number gives you a profit margin of about 60 cents something like that because that's 6.25 75 cents roughly a chicken so if you have something if you're unconstrained you're gonna do as many chickens as possible because you're making 75 cents per so what's gonna constrain you so you could say well I'm constrained to 600 because that's their capacity but that's not really what their capacity is because they can't sell that many so they would lose those so your capacity as I first look at it that limits it is your demand the demand 50 a week that dictates everything so when I first looked at it it was okay do I order 50 every week or 100 every two weeks because then I'd be steady state yeah does that make sense and the only thing that is that to add here is probably well if we're growing our business and demand picks up we are probably going to increase the number of chickens that we have per week and it follows your rationale if it's 75 demand per week then we produce 75 yeah but the other thing that I thought was interesting is you know we say we have two weeks but I'd have to ask because now like I said the devil's in the detail when you do these things can I actually buy chickens every week or do I have to buy them all at once is it every six weeks so that's one it looks a little bit like he can only buy batches of 100 one day chickens right because otherwise we would have given been given one or five chicken yeah that's true so it has to do that then he's going to be forced to do every two weeks and if you buy but you wouldn't want to buy 200 because then what would that mean you've got four weeks of chicken right there and so but the other thing I was going to ring let's say I have the chicken after six weeks let's say I have 100 so I have 100 chickens now um after like a week and a half or two weeks the ones that aren't sold can I put them in the refrigerator and get two more weeks or is there two weeks clock start right you see so something as simple as that when does the clock start going to be two weeks live and then two weeks refrigerated that'll change your strategy the other thing is we don't know how much they're fed we need to know a cost there because there's also keeping them alive it's that if that's the case you want to keep them alive a little longer because then you always have two extra weeks and so maybe you want to minimize that that 40 chickens you only kill the old ones the ones that are hitting their 13th day that's what you throw in there first in first out yeah yeah and you try and sell those first right you try to sell your old ones we did a project for Hershey's chocolate and it's very similar to this problem strangely they produce big massive blocks of chocolate ahead of time and you wouldn't know this but when they make candy bars the shelf life really matters and so when they sell it to a store different different candy will come from different runs of the base chocolate so they'll have different shelf life and the question they asked us to try to solve was do I give my oldest chocolate which has the shortest shelf life to my best customers or my worst customers and the best customers are defined as those that turn really quick versus the slow ones and the answer is generally you give your oldest chocolate to your customer that turns the fastest and because it won't they don't need the longer shelf life which is kind of counterintuitive you think you would give those guys your longest shelf life but it actually doesn't work that way and in that discussion then the refrigerator capacity becomes relevant right because then you need to decide to slaughter the chicken and put it into the refrigerator and how much can I store there or if there is overflow then I need to feed them continuously so that there is going to be a trade-off with the refrigerator. So then the next thing if we want to consider an emailed wisdom about this I said it's not really 50 every week it's 50 plus or minus right we know there's going to be a distribution so what if it's 50 plus or minus 10 so if you assume a normal distribution with a mean of 50 and standard deviation of 10 now your problem gets a little more different right and so now you want to figure out how much do I want to stock and so this gets into a problem that we look at in very much in depth in SE1X called the news vendor or a single period problem and what we'll show you then I'm not going to go into too much detail you're going to make a trade-off because that's what SE1X is all about making trade-offs and there's two costs cost of having too much and cost of having too little and so the ratio between those is actually cost of shortage over the cost of shortage plus the cost of excess tells you what the critical ratio is and the critical ratio is how much of the probability distribution I cover and that will optimize or maximize by profit so it's really an interesting problem when we talk about it in week five weeks six over I think right after the midterm we get into this it's a great model and it's again doing the trade-off of having too much and too little and that's where we get into the optimal price for right now my optimal price given the information I have is a million dollars right because if I assume chickens get sold charge what you can we need to know some more information about what an optimal price would be we need to know it's an unconstrained optimization right so if I have my maximize and my 7 minus whatever instead of the 7 it's now a variable X right my optimization will set X to infinity I have to have something that keeps it constrained and so we have to see what the market can bear so that question you could go to regression to look at comparable products I don't know that's something beyond the scope so you need to kind of figure out how demand reacts to price to give an answer to that right but maybe we can also talk about the discount question because it's kind of the end of that answer that we gave with the constrained optimization the question of when to slaughter the chicken and put it into the refrigerator kind of interacts with whether to use discounts a lot right because I could come up with the idea well if this chicken is slaughtered and sits there for I don't know 10 to 12 days already I may come up with well can I call for that my inventory decision so when we go over the news vendor problem essentially what you're describing is the added cost cost of excess so think of it this in general terms I've got two costs cost of shortage when I don't have enough and what do I what is my cost if I don't have enough chickens for the demand I lose the potential sales potential margin and so this is what 75 cents chicken I don't have enough chickens I'm giving up that much profitability excess well the idea is if I have too many chickens I'm going to have to throw them away right so I lose everything so my cost but if what Alex described is I can sell them for you know not seven dollars I can sell some for five dollars then what that's going to do your cost of excess goes down and so you're going to essentially excess goes down you're going to get a little more you're going to say instead of 50 chickens I'd have 60 chickens so what you'll see is as you change these levers it's going to change whether you order more or less based on your risk so if you're chicken if you're selling it for 20 dollars a piece right 20 dollars a chicken then you would naturally want to have more because then you don't want to miss out on that huge margin if your cost of having too much was more than just losing the chicken you had to do something else decontaminate the farm or do whatever because you have too many chickens running around and that would drive you the other way so again it's just a tradeoff based on the tradeoff of these different costs cost too much costs are too little so it's a very rich problem and one of the cool things is what we're going to do is we're going to connect that logic to service right because essentially what what you're describing is do we want to serve our customers and it's determined by the economics behind our problem right the more we are hurt by having too much so we're not so much interested in service but reducing our inventory the more we're interested in making the profit because the profit compared to what we lose is not so interesting the more service we want to bring out so we're going to connect that economic logic with service levels essentially that's a great point so that probability distribution of demand I said how much we cover that's a metric of level of service that will spend a lot of time in SE1X what percentage of the time will I stock out so it's called the cycle service level and we'll spend a lot of time on it but what is the probability that I won't stock out that I'll have enough chickens and so that's known as cycle service level we can also look at fill rate all these other metrics but they all tie to the same thing looking at that demand distribution so there's a lot more we could dive into this but hopefully what you're seeing is how we think about it and so depending you could use all these different tools to come up with something some people said queuing theory and you could think of a queue because essentially number of chickens in you can think of the processing time and then you could also put an alternate queue for the freezer so you could do something interesting here but again it's the devil's in the details so you need to know more details will help you model this all right very good do we have questions that go directly to what we've discussed here perhaps I don't see any, no not on this all right let's go to the second problem from practice yeah but the second problem I think it interacts pretty well with what we've discussed in the beginning about constraint optimization so this is given by Gino Beltran Beltran and it's about large-scale network design problems so Gino faces in his work different large-scale scheduling problems so in one example we gave it abstractly which is totally fine is that they have 20,000 different products or SKUs stop keeping units over 80 production plans with 10 to 20 ingredients or parameters for product to set with 1,500 total number of ingredients across all the products and 5,000 points of sale so just with these numbers I can just envision how large that graph would look like I would draw it out worldwide network of production plans points of sales that you have to model big company life so you'll never have a model a tool that solves this whole problem you're either going to solve the inbound so that's getting from where you source things in using a bill of materials into production plan or the distribution plan out so I've yet to see a company I don't know have you seen one that really tries to model the whole thing down to distribution usually very disconnected once you create the finished product but so usually what we'll see and we'll see this in SC2X you kind of divide this in half so you look at coordinating all the total ingredients it doesn't say how many vendors how many sources it comes from but usually it's like a usually a ton of raw materials locations goes down to very few production plans so you're back out to a lot of points of sale so it kind of looks like those two diamonds together and so having 20,000 SKUs in 80 different production plants my guess is that many production plants you're delivering locally I mean there's no reason this looks like food so you're probably doing a lot of this within a region so you can probably decompose this problem to be honest because if I say a production plant in I don't know Shanghai and something in Omaha in the United States they're not really going to interact right they're going to be separate so one thing when you have a problem like this can I segment can I isolate and solve some of these problems independently or do they connect at some point the second thing is you want to look at the SKUs 20,000 SKUs a SKU stop keeping you we'll talk more about this in SC1X could be just the different packaging it might not be that different so you want to look at families things that are like common and we'll talk about this in SC2X because this is one of the critical things when you do a network design you know how do I group different products together because you can't model each individual SKU like if I'm doing shoes a shoe manufacturer every size is a different SKU of the same thing you know I'm not going to model them that differently they're going to I can group them together so that's one thing having 80 production the other thing it doesn't say what the problem is they're trying to solve so are we trying to solve roughly some rough estimation of what the costs are going to be are we solving how things should flow from which plants should make which products going to which distribution centers and flowing it down or are we or those set and we're just determining when to do it is it a truly a schedule plan so they're totally different time frames and they use different techniques and they would also probably the aggregation ideas I think very important here right if you're planning a network then you're working on an aggregate level so you can probably consolidate SKUs maybe separate different production plans because and not separate different production plans if you're working on that specific scheduling problem you can probably go and look at each production site individually and so the second question here can they can these be solved with software like sass or ample maybe but I wouldn't there's these problems are so big and there's specialty software for this and we'll talk about them in SC4X but companies like SAP German company they have something called APO Advanced Planning Optimization APO I forget that he Advanced Production Optimization I think anyway there's different tools in these that are out there specifically doing say production planning or flow optimization companies like Oracle JDA which used to own the i2 tools which used to do a lot of this there's a lot of specialized software that is used for this sass and ample are generic tools sass is more like an analyst tool ample except for forecasting sass is used in production for forecasting it's one of the top forecasting tools out there ample is a really production grade simulation but no one that I know uses simulation to solve a scheduling problem because a simulation doesn't give you an answer it tells you what will happen if you do a certain thing so I haven't seen have you seen simulation used for this yeah because just the question is I want to know what the optimal schedule is not what will happen if I do something so different tool yeah so I think perhaps instead of simulation it's about heuristics right yes how can I get to a if you can't solve an optimum maybe not simulation but heuristics is the right word to look for and look up different methods and techniques to solve the problem maybe not optimally but close to optimal yeah and then and I think I talked about this and I see 0x people think you know optimization and then heuristics bad even for a large-scale problem like this the dirty little secret is if you use large-scale commentary optimization at the end of the day you're not solving it optimally you're solving it as close to a gap they call it the you know it's the integer programming gap between what you've solved and what you know the unconstrained linear solution will be so actually don't solve it for these real problems exactly exactly you get real close so heuristics aren't a bad thing yeah alright yeah what else anything else on this one no I think we're good do we have questions about I don't see any on this why we do the last one because it's it's pretty quick too isn't it in terms of introduction maybe take some time but perhaps in discussion it's okay and then we'll do questions yeah so this is by Molina Ibanez and it's talking about forecasting ordering for a Spanish grocery store chain so they are operating for the whole state of Catalonia in Spain and they have an automated warehouse to manage all the meats, fruits and vegetables that are dispatched to different locations so they have 40,000 storage locations in that facility and have around 1700 highly perishable SKUs that go through that warehouse perishable means Chef Lef is very very short so meats, fruits, vegetables they have like one to two weeks perhaps if they are pulled down I don't know I'm not an expert but something that you're not an expert on chickens or I consume them I can actually do something with them in the kitchen but I'm not sure how to store them so they have that guaranteed shelf life that they have to deliver to in store otherwise they have to sell it out at a discount so we are coming back to and ruin that discussion and a high turnover so 90% of the SKUs go through the warehouse within one day so a lot of volume that is going through that and we see that the volume varies a lot so of course one hand seasonal depends on winter or summer he brought he or she brought the example that meats are more important in the winter and in the summer it's more about the vegetables and it varies also across the week right because at the weekend you need to have a lot available so Thursdays and Fridays a lot of material is pushed out whereas on Monday it's not that important so the key here is that they have to dispatch the trucks to different locations and they have different lead times because far away trucks need to be dispatched earlier than close by trucks or trucks that go to close by but they go on the same day we're talking three hour difference saying well we have these slots of 12, 13, 14 and 15 where we push out the different trucks and his core question is how much inventory to hold in the warehouse to meet the critical cut of time so because it's again that question of having too much and having product that sits there too long but also having too little and not being able to push out in time so to me the critical question that is is where is it coming from so if they're a retailer they're sourcing this so I think the limiting thing is how quickly I can source it because if they're shipping it looks like a one day was it the third fourth? 90% one day so the nice thing about this so I would ask with any of these you got to ask more questions so I would ask from the supplier side how quick is their response do I have to order a week in advance can I order it the day before if I can have a really the shorter my lead time for ordering the more flexibility I have same thing on the other end when do the orders come in do the orders come in a week in advance or do I know it at 12 o'clock what gets ordered so as I have more lead time there that helps me right because I know so for the procurement side having a shorter lead time is good for me because then I don't have to do as I can now wait to make my decision but I want as much lead time on the other end so knowing those two things to me that would dictate everything because what we'll see is lead time is a critical component to determine the optimal inventory level as you have a longer lead time to get something when if I have to if it takes a week if I order something and I don't see it for a week that's different than if I order something I get it for a day because if it takes a week then I've got to keep a week's worth of demand of supply here to handle the demand while the things getting to me right and so as I shrink my order lead time the amount of inventory I need goes down as I increase the amount of lead time I have to place the order of what to put so say my customers let me know a week in advance exactly what they want that's much better so you were I think you'd want to look at the trade off between those two lead times and so if the lead time to order is longer than if the lead time that the customers give you is longer than the lead time you give your vendors your golden if it's the other way around you're kind of screwed to be honest anyway it's kind of the difference between having warehouse just for purposes of recombining different truckloads or whether it's actually serving as a buffer between your supply and your demand right so that's kind of if you can go to it yeah if you can go to a total flow through that's the best and we talked about this in 1x and when we talk about DC's the whole idea of cross stocking this is the idea if you can cross stock it do it and keep just minimum amount of the things on hand that takes longer to procure alright I'm gonna start asking questions let's see what the questions have let's take from the top probably running simulations what level of aggregation do you recommend example do I consider the individual distributions of each individual source or aggregate aggregate it depends on the problem but it kind of ties into what we discussed for the second problem it depends on what you want to answer but make sure that you have enough detail that you need but not more because it's computationally expensive there's an Einstein quote your model should be simple enough as simple as possible but not not over too simple it has to be complex enough to capture it but you want it as simple as possible figure out that bulletin spot right how can we perform inventory optimization using optimization tools are we going to cover this in SE 1x yes you're gonna cover this a lot when you say optimization tools what we'll do is we'll show you how to solve it mathematically we'll do it in like Excel spreadsheets because you'll do it if you have like SAP your ERP systems in a company will be doing this generally they'll be giving you what the order quantity is given the lead time and all these other characteristics we'll show you what the math is so yeah you'll do that and we're doing that for in SE 2x we're doing that for production planning too so we're not only doing that in inventory management and show you how that works and how you can implement that on a smaller scale if you're not using some advanced software like SAP and we're doing that in SE 2x for production planning and scheduling same thing so EFS will later courses give advice on communicating these results findings to company leaders may not understand the underlying methods not really we do that we think so online education is awesome it's the best way to learn everything you've learned so far in SE 0x in personally online education online courses, video courses are horrible for teaching change management and that's what you're talking about how do I convince someone to make a change and we talk about that that's why we run boot camps that's why we do live events like this because it's better to have interaction we just ran a course this morning for a large beverage manufacturer where we had 30 students and we could see them live and we could interact that's the best way to understand how to do these change management techniques we can talk to you about it and say here are some tips but it doesn't work as well unless you really discuss it does that make sense? it's a setting for a classroom how can these concepts be applied to urban design and planning eh well all the stuff we did for routing when I actually showed algorithms and approximations I killed two birds with one stone there because I wanted you to see network optimization and that's a way to teach that really quickly because that's a great skill to have so a lot of that urban design uses some of those approximations but a lot of the other stuff it really doesn't apply the techniques apply you can use regression to find out what flows are when I used to do I used to be a passenger transportation guy and you develop the the gravity model you determine what generates traffic and what not for urban planning and you're using regression but it's very far away from supply chain any reading materials about books for large-scale problems it depends what you're talking about so you're the one who put that question in I recommend going to the vendors go to SAP go to Oracle go to JDA because I think you want to understand how to actually solve those problems not the large-scale math that's sitting underneath it if you want to go to Nemhauser and Woolsey and they do comment to our optimization and it's really hard but that's at the heart of the optimization if you want to find out what's done in practice go to the vendors Oracle, JDA, SAP a simulation really a question here is simulation used to check the performance of your inventory problems it can be that's actually a good way to use simulation to check a certain instance so if you do when we do EOQ optimal quantity to order you could actually run a simulation and you'll come up with the exact same answer it depends what you give as the underlying distribution but yes it's used to check that the first question you've obviously taken some of SC1X because the power of two technique I'm not going to go into it it's a nice approximation technique that we talked about for inventory for class C items that makes sense I don't want to go into that though because that's not really relevant here so Param has also an interesting question about the chicken problem the first problem that we covered about how to factor in the mortality so if we have that deterministic setting and we buy 50 so if they die within then there is a probability that a chicken dies so we have a second uncertainty so you don't you not only have demand uncertainty but you have also production quality let's say you know 10% die then if you know it's always 10% then what do you do? you order 10% more that's great, you can just however SC3X we actually do this problem we do the chicken problem I use it to say system dynamics because here's the dynamic is as I have more chickens they tend to cross the road the chickens cross the road they get run over so there's this dynamic so we do a little more sophisticated optimization because the number that die can be a function of the size so that makes it more complicated and we talk about the system dynamics in SC3X so stick around it's an interesting question is there any course referring to WMS? what is WMS? warehouse management system I talk about it in SC4X and I talk about distribution centers in general just kind of one lecture in SC1X in week 9 or 10 so yes a little bit not too many people do too much on WMSs to be honest there's John Bertoldi B-A-R-T-H-O-L-D-I out of Georgia Tech it's in a reference for the SC2X SC1X key concept document he is the master and he has a great book that's available online that's called warehouse science that you can look at so go to the key concept documents you have access to them on all the courses and then look at the end for SC1X for the warehousing section and that will give you the link for that I think we would need to have the geographical location of the suppliers density of the products where we need and I don't know what the question is so excellent point you're probably right oh that probably refers to this yes you're exactly right if it refers to this problem for practice 3 you're exactly right we don't actually need to know where they are we need to know how long it takes time is all that matters unless it's crossing a border and get some other things alright I'm fixed we covered that already because we can charge whatever we want exactly we covered that excellent point do we need to buy more than demand to address unplanned yes exactly we covered that as well same thing we covered it last question here for pf3 we answered that yeah we did it's all about the lead time for the Spanish grocery store it's comparing how much lead time you've got to give your suppliers and how much you can give your customers so we have that so you get to pick the last question I get to pick the last question let's do the heuristics versus optimization what's the threshold when do you switch from a heuristic to an optimization in fact where did it go it's from Giovanni the third one right when you can't solve it well yours that can be an optimization an exact optimization but I think if you can solve something optimally in a short amount of time do it the benefit of a heuristic is when you can't and so like when we did the news event we did the traveling salesman problem it's very hard to solve that optimally actually so in approximation a heuristic works well so it's really if I can solve it exactly within the time I'm allowed to solve it do it we do a project with Walmart to determine where they use their for hire fleet and their private fleet and we can solve it in two hours exactly now if they wanted to know within 10 minutes we'd have to go to heuristics but because a two hour run time is sufficient for them we could do it optimally so it's really a function let me answer the bootcamp question I was just there and we ran a bootcamp I was in Mingbo 10 days ago and we sent this note out to people and we didn't have that many show up we end up with 20 people so I've been in Asia I've been in Europe and we've been doing these bootcamps so you need to pay attention to your emails because I ran one there I didn't see you there Shara but we're trying to do these more across the world we're trying to see what the demand is because it costs it takes a lot of effort for us to go there and so we have to bounce that out and with that I'll hand it over to you to wrap it up wrap it up take a quick look at the initial poll that we had where we covered the different words word cloud because it fits very well they like us, they say we're awesome and we are supposed to teach everybody here at MIT how to do a course effectively and efficiently so I think challenging is the best way to describe SC0X and that's essentially your first sentence today we threw a lot at the students that is our toolbox and we're going to use that toolbox to go through the courses alright well guys I hope you liked it hope you liked the SC0X SC1X is more fun we cover much more real things and the math is not as the focus, the math supports the analysis it isn't the focus and we're going more into different methods and problems that you're facing every day so thank you very much Chris for taking the time that was awesome thank you very much Lurena for attending for being so active we appreciate that very much and we wish you good luck with the final exam and hopefully see you all in SC1X which starts tomorrow thank you very much bye-bye