Baselines and Optimal Baselines Assignment
Order ID 53563633773 Type Essay Writer Level Masters Style APA Sources/References 4 Perfect Number of Pages to Order 5-10 Pages
Baselines and Optimal Baselines Assignment
8 BASELINES AND OPTIMAL BASELINES
Up to this point, we have compared the results of our optimization run to other optimization runs. However, the goal of most network design projects is to improve an existing supply chain. For example, when you are presenting your results, you may be asked how much money is saved with the new design or how much the service improves. To answer these questions, you need a baseline. A baseline is a model of the existing network.
Equally important, you will also be asked a deeper question: How do you know that your model is valid? The baseline model can solve that problem as well.
It turns out that there are two main baseline models you will want to consider. Each provides its own set of insights. We will call the two models the actual baseline and the optimized baseline. Both are important and both are critical.
The actual baseline is a representation of the supply chain exactly as it was run in the past. In practice, most modelers use data from the previous year to build the actual baseline model. In the actual baseline, you have exactly the same mathematical formulation that we’ve previously covered except that you are eliminating any choices for the model.
That is, the actual baseline uses just the existing facilities and the current assignments of facilities to customers (in more complex models you will extend this for flows between facilities and where product is made and how much is made at each location). In terms of the mathematical formulation, we are taking the decision variables and assigning them a value. During an optimization run, we have let the optimization engine pick the best value. Now we are telling the model which facilities are open and which ones are closed. We are telling the model which customers receive product from each facility. And so on. In effect, you are turning the mathematical model into a big calculator.
When you are building this baseline, it is important that you set up the model exactly the same way you are going to set up the optimization runs. That is, if you are going to run the optimization with outbound transportation costs, you want the baseline to include the outbound transportation costs.
We have found it helpful to set up the baseline in two steps. In the first step, you set up the model just as we set up the optimization models in previous sections. That is, you set up the structure of the model first. It is this structure that you will use later when running optimization scenarios.
The second step is to lock in all the decisions. In this step, you input data such that all the existing facilities are used, and you specify the flow from each facility to each customer. In previous chapters, our maps and solutions all looked relatively clean. In the actual baseline, we do not expect this to happen. Here you are entering data as it happened. For example, in the actual baseline, you may have shipments from your New York facility to your customer in Los Angeles even though you have a facility in Los Angeles. Things like that happen all the time in a real-world supply chain. We do not want to pretend they don’t.
After you have this model built, it provides you a nice way to validate the structure of the model and the costs. The idea is that your baseline model reflects the business you are modeling. You compare the results from the model with the costs that the company actually incurred. You typically want the costs of your model to come to within 1% to 10% of the actual costs. There is always a good discussion on how close you need to get to the actual costs, and a lot of it depends on the culture of the company you work for and the importance of the project.
The main argument for getting close to 1% is the fact that the firm is making important decisions and they want to feel comfortable that the model that is guiding the decisions is accurate. Because you are working with historical data, it should be possible to match the model costs with historical costs. If you cannot, there may be a problem with the model. In this case, you do not want to fall into the trap of fooling yourself that costs within 1% guarantee that your model is fine. There are ways to load a baseline to get the costs you want that could make future optimization runs misleading.
The main arguments for being fine with something in the range of 10% can be two-fold. First, the availability of quality data may be an issue, thereby requiring the use of assumptions to fill gaps in data. For example, you may be missing some data on outbound shipments to customers but have information on total freight spend. In this case, you can input outbound data that is available and allow the model to determine the remaining customer assignments through optimization. These optimized assignments may be slightly different from what actually happened, thereby yielding a baseline cost that could be 10% away from actual costs. This is not necessarily a bad thing as we will be generating a similar result with the optimized baseline (discussed below) which will be used extensively during the analysis.
Secondly, the firm realizes that a supply chain is complex and always changing, so it is difficult to model costs exactly. And the firm is not going to make a decision unless the savings are relatively large compared to the baseline. So the baseline serves as an anchor, and if we find 20% savings from the baseline, we can assume that we’ve found a signiicant savings. In this case, you still want to make sure that the baseline model is functionally accurate and not off by 10% because of mistakes.
Besides comparison of the overall cost of the model to the overall actual costs, you also want to compare the costs at a detailed level. For example, if the actual transportation costs are $10 million, you want to come close to the $10 million. And if the transportation costs from one warehouse are $3 million, you want to come close to that number as well. You can continue this process for other variables and at a finer level of detail.
This detailed analysis not only tests the cost validity, but also helps test the structure of the model. That is you, are testing the underlying details to make sure the model is behaving correctly. You can also make small tweaks to the baseline to make sure that the costs move in the direction they should. This also tests the structure, and we’ll explore this more in the optimized baseline section.
The reader may ask: which project has better results, one that ensures a 1% difference or 10% difference between the model and reality? The answer is that we have seen both types of projects be successful. You will have to use your best judgment for each project. Also remember, an important part of these studies is selling the results internally. A big part of selling internally is convincing people that the model is correct. In some firms, you will need the baseline results to be very close to actual results, and in other firms not as close.
After you have validated the results, you now have a reference point and a model that has passed at least one check of its validity. But, keep in mind, that even though you have used the actual costs to validate your model, this is only one test. You need to keep a vigilant eye on the model validity throughout the entire project.
This baseline model needs to be ready to be converted to the optimization model for two reasons:
- 1. You need a fair point of comparison. If you change the structure of the model before running the optimization, you can no longer be sure your comparison is fair.
- 2. You need a valid starting point. The actual baseline has been validated. If you change the structure, you may no longer have a valid model.
It is good practice to minimize the changes to the structure of the model in going from one scenario to another. If you change the structure, you really need to revalidate the model.
After the actual baseline is done, you will want to run the optimized baseline. In the actual baseline, we modeled everything that actually happened. In the optimized baseline, you want to replicate what should have happened based on the rules you had in place and if you executed according to plan. In our example from earlier, we want the Los Angeles customer to receive product from the Los Angeles facility, not the New York facility. You can also think of the optimized baseline as “cleaning up” the actual baseline.
An optimized baseline is not as well defined as an actual baseline. You have some flexibility in how you set up the optimized baseline model. And you may want to run several versions of the optimized baseline. The spirit of the optimized baseline is to clean up the baseline and come up with a model of what should have happened. Typically, an optimized baseline has the following structure:
- ■Uses all the existing facilities
- ■Uses existing assignments of facilities to customers (or later from one facility to another). In some versions of the optimized baseline model, you may relax this constraint and let customers receive product from different warehouses.
So, what does this model tell us?
First, it helps us validate our model. We want the results of this model to make sense. When we clean up the model and let products flow as we planned, are costs reasonable, are capacities still respected, and are other business rules respected? This reminds us that we need to continually evaluate our results. Real-world supply chains are complicated. Models are just a representation of the real world. We need to keep a skeptical eye on our models to make sure that they are giving results that reflect enough of the real world to allow us to make decisions.
Second, the model helps us identify areas for improvement. If there is a significant and validated cost difference between the actual baseline and the optimal baseline, it may signal a chance to improve. In this case, you want to understand why the actual supply chain deviated so far from the rules you put in place. You may have an opportunity to change the way different parts of the business operate so that you can better follow the rules that are in place. In other words, if you can execute better, there may be significant savings opportunities. Although no changes to the physical infrastructure of the supply chain are needed, you can still find significant savings. Occasionally, there is more benefit in running the existing supply chain better than re-designing it.
Third, this model provides a good benchmark to compare our optimization runs. When we run an optimization model, it will not make illogical decisions (it won’t ship from New York to Los Angeles if there is a capable facility in Los Angeles). So we do not want to give ourselves a false sense of the results by comparing an optimization run to an actual baseline. That is, the actual baseline may have a high cost based on poor execution of the plan and not a poor plan. By comparing to the optimized baseline, we are comparing optimized scenarios versus an optimized version of the baseline so we have a fair comparison.
Other Versions of the Baseline
In some cases, the baseline and the optimized baseline as we described previously may not fit your situation completely. The following two cases provide immediate challenges:
- 1. The supply chain you are modeling experienced significant changes in the past year.
- 2. The supply chain you are modeling does not have a logical baseline or historical set of data.
Case #1 can happen if the firm has opened and closed significant facilities in the past year or has changed the product mix in a dramatic fashion. This means that the actual costs from the past year are not going to be valid and may not provide a clear basis for validating our baseline model.
One way to accommodate this is to run two or three different baseline models. The baseline models would capture the cost of what would have happened if the changes hadn’t occurred, the costs of what would have happened if the changes had happened at the start of the year, and the combined cost that attempts to match up the costs with what actually happened. Then, going forward in the analysis, you want to use the model with the changes as your starting point.
As an example, let us say that a firm shut down a plant in July last year, and we are looking to use data for the last calendar year. With a planned plant shutdown, you would expect to see production going down at this plant in the months leading up to July, and consequently volume picking up at other plants through July and beyond. In this situation, we can create two baselines—one representing January to July, with the old plant showing production, and another baseline for August to December, representing the network without the shut-down plant. After we validate both models using their respective data, we can then use the second baseline model (August to December) as the basis going forward as this is the new reality. For purposes of consistency, we can take this baseline and annualize its volume and use this version (call it revised baseline) to apply demand growth forecasts for scenario analysis.
Case #2 can happen if you are modeling an entirely new supply chain or are modeling so far out in the future that last year’s demand is not considered a valid starting point. This case is a little trickier and requires some creativity.
The first thing to keep in mind is that you need to build some baseline. You need a way to validate the structure of the model and validate that the costs are reasonable. In some cases, the firm will have a preliminary budget in mind and you can use that as your reference point. Or you can use a version of the optimized baseline as the baseline. That is, you determine the cost of the new business using your existing structure. There is no correct answer here, but the more similar you can make the baseline to your existing network, the easier it will be to validate your model.
Baseline Case Study—Illinois Quality Parts, Inc.
Let’s go through a case study to learn and understand how to develop baseline and optimized baseline models before starting to evaluate alternatives. We will take the case of Illinois Quality Parts, Inc. (IQP), a large distributor of industrial parts and components based in the Chicago area. The company distributes a wide variety of industrial parts ranging from small components such as bearings and belts to heavy parts used by original equipment manufacturers (OEM) that make of automotive, heavy equipment, and industrial products.
The company distributes these products through their network of three warehouses located in Riverside, California; Addison, Illinois; and Bridgewater, New Jersey. The parts are procured from specialized manufacturers of the parts located all over the U.S. The parts are delivered to the three warehouses with freight paid by the vendors.
The customer base includes original equipment manufacturers, small distributors, and retailers located across the U.S. The company consolidates customer orders and ships them in full truckloads to its customers. Each warehouse has a customer territory assigned to it, based on which, the customer orders are routed to the appropriate warehouse and shipped out. The map in Figure 8.1 shows the customer territory by warehouse (or called a DC-distribution center by IQP).
Looking deeper into their supply chain costs, the team calculated that they were spending $78 million in outbound transportation costs annually. See Figure 8.2 for a breakdown of outbound freight costs by warehouse. (Note that the totals may not add up due to rounding.)
Figure 8.2 Annual Outbound Freight Costs by Warehouse
The management team had not performed a network analysis in over five years and wanted to understand whether there were opportunities to reduce costs—this included evaluation of customer assignments to warehouses, and changing the number and location of warehouses in the network.
Analyzing Shipment Data and Creating Customers and Demand
Before we set out to answer questions around alternatives, first we want to build a baseline model that accurately represents how the supply chain operates today. To build such a model, we will want to collect and use actual historical data as this will show how the company’s supply chain actually worked.
We will start our analysis by collecting historical shipment data and building the customers and demand. Because this analysis will primarily focus on transportation costs, we will use weight (pounds) as the unit of measure to model customer demand. From a transportation perspective, the vendors pay for inbound freight, and this will likely not change with any network changes; so the management team wanted to focus on only the outbound costs.
A sample portion of the shipment data is shown in Figure 8.3 . This data serves two purposes: (1) We can use this to derive the demand by customer for the model, and (2) this tells us which warehouse shipped to which demand point over the past year.
Figure 8.3 Sample Extract from Historical Outbound Shipment Data
We then take the shipment data to identify the customer locations (from the Destination fields in the shipment data) and the total demand (sum of shipment weight) for the baseline model. To create the list of the customers, we group all the same destinations together. That is, the shipment file may have 100 line items for shipments to the same location and these will be grouped into a single customer. Then, we add up the total demand for each of these customers. Figure 8.4 shows the customers plotted on a map and the states shaded by their relative demand.
Modeling Historical Data as Predefined Flows
We can see from the sample data (in Figure 8.3 ) that customers in Tennessee (TN) were served from both Addison and Bridgewater warehouses, even though Tennessee falls under the Addison warehouse territory according to their customer assignment rules.
Why would this happen? This is very typical of any given supply chain as there are several reasons that require departure from the published or expected policy. The most likely reason for out-of-territory shipments is inventory availability, wherein there may not be sufficient inventory of the ordered items to ship from the assigned warehouse, thereby requiring shipment from another warehouse. In these cases, it is cheaper and faster to ship to the customer from another warehouse than to get the product replenished at the Addison warehouse from Bridgewater warehouse and then ship to the customer.
We will revisit this topic after we run the baseline model to understand the magnitude of these out-of-territory shipments.
So, our next step is to replicate these shipments in our model. For example, we want to show customer demand in Tennessee being served from both Addison and Bridgewater warehouses. We can model this by predefining the flow of product in each lane based on the shipment data. Each commercial network modeling software has its own way of inputting this data into the model, but the general concept is to specify the total volume (in pounds in this example) moving from each warehouse to each customer.
We see an example of the predefined flow for the Nashville, Tennessee, customer in Figure 8.5 . Based on the shipment data, the total demand for this customer is 4,259,050 pounds; we are predefining that 2.981 million pounds of this demand would be served from the Addison warehouse, and the remaining demand (1.27 million pounds) would be served from the Bridgewater warehouse. We then fix the variable to honor these constraints. This concept of predefined flows is applied across all the ship-to locations in the baseline model.
Figure 8.5 Example of Predefined Flow for Baseline Model
Modeling Transportation in the Baseline
Now that we know the flows, let’s add the transportation costs. The shipment data file included shipment weights and actual freight costs by truckload shipment. We can use this to derive transportation cost per mile rates for the model using regression analysis (as described in Chapter 6 , “Adding Outbound Transportation to the Model”).
We do not yet know the right level of granularity that would provide an acceptable level of accuracy to the baseline model—for example, will it be acceptable to calculate and apply a $/mile rate across all lanes, or do we need to estimate the rates at some lower level?
As mentioned before, we typically aim for getting the baseline model costs to within 1% to 10% of actual costs. The higher end of the range is typically acceptable if the quality of data was poor, requiring the use of assumptions to substitute real data. If our modeling analysis yields a baseline model that falls within this tolerance and gives a reasonable level of confidence that it is accurate, we have met our goals for this task.
To go through this process, we will start modeling transportation with an overall $/mile rate across all lanes. If this does not yield a sufficiently accurate result, we will recalculate at a lower level and reapply in the model.
The Logistics team at IQP estimated that their average truckload rate was $2/mile across all lanes, with a minimum charge of $450/load. Based on a quick regression analysis using all the lanes as the sample, we come up with something close to this as well. We will start with using this $2/mile rate in the baseline model.
Baseline Model Results
The maps showing results of the baseline model are displayed in Figure 8.6 . The maps from the baseline model depict the reality of how customer demand was met. There were out-of-territory shipments from all three warehouses, including products moving from the Bridgewater warehouse all the way to California.
Figure 8.6 Maps Showing Aggregate Baseline Model Results and by DC
Now let’s look at what the baseline transportation costs look like relative to the actual freight spend, based on the $2/mile TL rate (see Figure 8.7 ).
Figure 8.7 Results Comparison—Actual Versus Baseline Transportation Costs
When we compare the total transportation costs between actual ($77.55 million) versus the baseline model ($83.51), it shows that the model results were about 8% different from the actual costs. This is actually not a very bad result and can be acceptable in a lot of cases. Before we make a decision on whether this is acceptable, let us dig a little deeper at the cost comparison by warehouse. The costs out of Addison are very close to actual (1%), but the freight costs for Bridgewater outbound are off by 25% from actual costs, while Riverside outbound costs are off by 7%. The 7% delta can be ignored for now, but the 25% delta is very large. What does this mean? The comparison shows that the $2/mile is roughly 25% higher than actual for New Jersey origin. This is important to note—all things being equal, the model will assign customers to a warehouse that provides the lowest cost per unit, which is a factor of both distance and $/mile. If the rate is lower than $2/mile, that means the New Jersey warehouse can potentially serve customers farther away than previously estimated. It could also affect the choice of warehouse location picked because it may be cheaper to locate a warehouse in a zone or state that has lower outbound rates.
Based on this previous analysis, we arrive at the conclusion that a generic $2/mile rate will not be appropriate for all lanes. The logistics department was able to provide a state-to-state matrix with $/mile rates that we can use in the model. See Figure 8.8 for a sample extract from the matrix.
These types of rates typically have large states such as California and New York broken into North–South or East–West, given the differences in demographic and economic activity within the state.
Let’s rerun the baseline model substituting the $2/mile rate with these detailed rates. Note that we are changing only the transportation rates; the predefined flows are not changing, so the maps for this revised baseline will look exactly the same as the previous iteration.
The results (as shown in Figure 8.9 ) show that the baseline model costs are much closer to actual with the state-to-state level rates. The total cost is within 3% of actual and within 5% of actual by warehouse. This tells us that we are reflecting the differences in outbound rates by origin and potentially by destination, thereby yielding model costs that are very close to actual.
Figure 8.9 Results Comparison—Actual Versus Baseline with New Rates
Based on these results, the management team can feel comfortable that our baseline model is calibrated appropriately and is a good representation of how their supply chain operated historically.
Building the Optimized Baseline(s)
Now that we have a validated baseline model, our next step is to create and run an optimized baseline scenario. This scenario is intended to show what it would look like if the current network operated optimally and without any operational exceptions. In this case, we can come up with two such versions of optimized baseline:
- 1. Optimized baseline with customers served according to their assigned territories
- 2. Optimized baseline with optimized assignment of customers to warehouses
The first version will show what the cost would be if there were no out-of-territory shipments based on the existing territories. Because most of these exceptions are attributed to inventory availability, it will help quantify the financial impact of not having the right inventory in the right place at the right time.
In other words, the combination of baseline and optimized baseline models can help quantify the financial impact of these operational issues, and help the team initiate immediate corrective action.
The second version of the optimized baseline will tell us whether the customer territory assignment itself can be improved—that is, whether there are opportunities to reassign specific customer states or locations among the warehouses to reduce overall costs. The customer territories have not been evaluated by the logistics team in a few years, so it may identify quick opportunities.
Note that in both scenarios we are not making any changes to the warehouses or the network itself. We will keep the warehouses fixed and not look at any new locations in these scenarios—these will come later after the optimized baselines are completed.
The results of the two optimized baseline scenarios are shown in Figures 8.10 and 8.11 . The map for the scenario with current territory assignments shows flows as expected according to the current business rules. The map for the scenario with optimized territories shows the Riverside warehouse picking up states such as Colorado, New Mexico, and Montana. We also see Bridgewater shipping to Southern Louisiana and Southern Mississippi, which were originally part of the Addison territory.
The cost and volume comparison provides us with a lot of insight on improvement opportunities.
This analysis shows that there is a $2.6 million or 3% cost reduction opportunity if the outbound orders were shipped according to their assigned territories. This means that the company could save over $2 million if they had the right inventory at each warehouse—this is usually attributed to an insufficient amount of safety stock buffer at the warehouses. The company can use these cost savings versus the cost of higher inventory levels at these warehouses to determine what makes sense.
This scenario also shows that Riverside and Bridgewater are currently handling more volume (as represented in the Baseline) than they should based on their customer territory assignments (Optimized Baseline—With Current Territories). The converse applies to Addison, which is serving less volume than it should be according to the territory assignments.
When we look at the second optimized baseline (with optimized territories), it shows that the company could save $5.75 million or 7% over the baseline model if the territories were realigned optimally. In this case, Riverside should be serving states farther east of its original territory, as well as other minor reassignments between Addison and Bridgewater warehouses as noted earlier.
The consequence of this realignment is that the Bridgewater and Riverside warehouses will handle additional volume. Note that we have not modeled capacity in this model so far, so this is not factored into these outputs. We can easily add the capacity constraints and rerun the model. However, there is value in running this scenario without capacity constraints—it tells us what the optimal throughput would be for each warehouse. This is valuable information that the company can use to determine whether this can be executed operationally.
As discussed in Chapter 5 , “Adding Capacity to the Model,” it is usually hard to precisely define warehouse capacity, so it is typical to run some scenarios ignoring capacity and evaluating the model outputs versus baseline (or current) in order to understand feasibility. In this case, the model recommends that Riverside handle 21% more volume and that Bridgewater handle 10% more compared to last year. The management team needs to evaluate whether this is feasible given current capacities. Based on past experience, we have seen that a 10% increase in volume can be easily accommodated even within warehouses that are at full capacity; a 20% increase is usually a bit more challenging if capacity is tight, so this needs further analysis.
Though the primary focus of this case study was the baseline and optimized baseline models, we will wrap up this case by showing the best two-, three-, four-, and five-warehouse solutions. Figure 8.12 and Figure 8.13 show how these solutions compare to the baseline.
Lessons Learned from Baseline and Optimized Baseline Modeling
The actual baseline model is a representation of the current supply chain and how it operated in the past. This is an important first step in the network modeling process because this helps validate that the network model accurately represents the supply chain and its flows.
After the baseline model is validated, it is important to run the optimized baseline scenario(s). This represents the current network and its locations but shows the impact if everything happened according to the current business rules. There can be several variations of the optimized baseline, including those in which some of the rules are relaxed (e.g., customer assignments to specific warehouses). This model also serves several purposes. It shows the potential improvement opportunities within the current supply chain without having to make major infrastructure changes. It also provides a good basis for comparing what-if scenarios—because this scenario includes optimized outputs, it serves as a good apples-to-apples comparison with optimized outputs from what-if scenarios. The optimized baseline can also help validate the model itself well.
A well-designed and well-developed baseline and optimized baseline model serves as a strong foundation for development and running of what-if scenarios.
- 1. What is the difference between an actual baseline and an optimized baseline?
- 2. Open the file from Chapter 3 , “Locating Facilities Using a Distance-Based Approach,” called MIP for 9-City Example.xls on the book Web site. Assume that the baseline had St. Louis serving itself, Chicago, and Cincinnati while Boston served all the other locations. How would you set this up in this model and what is the baseline solution? If, in the baseline, Atlanta, New York, and Charlotte were served by both St. Louis and Boston with a 50% split, how would you set up the baseline now and what is the solution?
- 3. If your baseline model’s costs are 20% lower than the actual costs, under what conditions could this still be a valid baseline?
- 4. If your baseline model’s costs are within 0.1% of the actual costs, under what conditions might this still not be a valid solution?
- 5. In the Illinois Quality Parts case study, we first ran the baseline model with a rate of $2/mile for all lanes. Figure 8.14 shows the output of this scenario showing total cost and total miles traveled out of each warehouse.
Figure 8.14 Baseline Cost Comparison
- a. Even though we applied a rate of $2/mile, the table (total cost / total miles) shows a rate higher than $2/mile for all warehouses. Why is that? (Hint: Think about the fact that the true transportation rates include minimum charges.)
- b. The solution with the best two warehouses shows an average distance to the customers that is 144 miles higher than the optimized baseline with optimal customer assignments. However, the freight costs for the two scenarios are very close (~$74 million). How can you explain this?
- c. When we look at the results of the scenarios shown in Figure 8.10 , it looks as though some customers are not assigned to their closest warehouse. Why is that? Does that make sense?
- 6. You are starting a network design study for a company that wants to reduce costs in its distribution network. Their financial year is the same as the calendar year. Last May, the company changed their packaging strategy and the gross unit weight of their products increased by 20%—that is, for the same number of units ordered, the total weight shipped went up by 20% due to increased weight of packaging material. The products were shipped in full truckloads that typically weighed out. The demand is fairly stable all year with no seasonality. What timeframe would you use to pull data for the baseline? If you used all 12 months, what would be the advantage? What would be the disadvantage? How will you model demand for the scenarios?
Tired of getting an average grade in all your school assignments, projects, essays, and homework? Try us today for all your academic schoolwork needs. We are among the most trusted and recognized professional writing services in the market.
We provide unique, original and plagiarism-free high quality academic, homework, assignments and essay submissions for all our clients. At our company, we capitalize on producing A+ Grades for all our clients and also ensure that you have smooth academic progress in all your school term and semesters.
High-quality academic submissions, A 100% plagiarism-free submission, Meet even the most urgent deadlines, Provide our services to you at the most competitive rates in the market, Give you free revisions until you meet your desired grades and Provide you with 24/7 customer support service via calls or live chats.