Strategic Supply Chain Management Chapter 8 Strategic Supply Chain Management

Transcription

1 Strategic Supply Chain Management Chapter 8 Strategic Supply Chain Management Contents Topic Sales Forecasting Cost Factors & Data Aggregation Strategic Supply Chain Model from Practice Page 1

2 Strategic Supply Chain Management Scope of the Strategic SCM Strategic level of the SCM Comprises the strategic planning of locations for companies as well as basic logistics nodes in the field of procurement, production or distribution. Concerns the identification of long-term sourcing, procurement, production and distribution strategies. In many cases planning software considers flows of quantities as well as logistics costs to support the strategic level decisions. Sometimes location decisions are on a tactical level, too. Example: Renting storage space or production capacities. Page 2

3 Topic of the Strategic SCM Characteristics of strategic planning high data aggregation high imprecision of the forecasted data or even no data available long-term decisions involve high investments decisions involve the higher management level Strategic decisions heavily influence the efficiency and cost-effectiveness of a supply chain customer satisfaction. Page 3

4 Topic of the Strategic SCM Involved decisions Procurement Production Choice of suppliers and determination of the demand on raw materials. Which products should be produced where and in what amount. Location planning Number, location and capacity of new facilities. Distribution Market area Choice of transportation routes for the distribution of products between facilities and customers. Allocation between facilities and customers. Page 4

5 Topic of the Strategic SCM Make decisions in such a way that the total costs for satisfying customer demands are minimized with regard to different service levels. Balance between service level (close to customers) and Procurement and production costs Inventory costs Setup costs (storage, labor, administration, ) Transportation costs Page 5

6 Topic of the Strategic SCM Trade-off using the example of warehouses A higher number of distribution centers results in an increasing service level due to shorter transportation times to the customers increasing inventory costs due to higher safety stocks in all distribution centers higher administration and organization efforts and higher fixed costs reduced costs for outgoing transports from the warehouses to the customers increasing costs for incoming deliveries from the suppliers / factories to the distribution centers Page 6

7 Strategic Supply Chain Management Strategic Supply Chain Management Contents Topic Sales Forecasting - Forecasting Methods - Time Series Analysis - Quality of Forecasts Cost Factors & Data Aggregation Strategic Supply Chain Model from Practice Page 7

8 Strategic Supply Chain Management Sales Forecasting A sales forecasting estimates the future trend of the demand Your demand will decrease about 50% Almost every planning decision bases on sales forecasting Sales forecasting is a structured process There are several different methods to forecast future demand, depending on the application and aims of the forecast Page 8

9 Sales Forecasting Long-term forecasts are often wrong 640K ought to be enough for anybody Bill Gates, 1981 When the phone was first demonstrated to President Rutherford Hayes, he is reported to have said: That s an amazing invention, but who would ever want to use one of them? President Rutherford B. Hayes, 1876 A phonograph is just a mere toy, which had no commercial value. Thomas A. Edison, 1880 I think there is a world market for maybe five computers. Thomas J. Watson, 1948 There is no reason for any individual to have a computer in their home. Ken Olsen, 1977 Page 9

10 Sales Forecasting Forecasting methods Judgmental methods The opinions of several persons are combined to one forecast. Qualitative criteria. Demand for flights in 20 years Causal methods A function is estimated, which represents the influence of known underlying factors on the demand. Demand for service personnel for the next year Time series These methods use historical demand as the basis of estimating. Demand for detergent for the next month Page 10

11 Forecasting Methods Examples Question 1. Which energy sources will be used in 20 years? 2. How will the UMTS sales develop in the next five years? 3. How much service personnel is needed to maintain the telephone network? 4. How will the GSM sales develop within the next 12 months? 5. How many salesmen will be needed within the next week? Forecasting method (typical) Judgmental method Judgmental method Causal method Time series Time series

12 Forecasting Methods Judgmental Methods Application Are used if no foretime data is available or the data is not appropriate for forecasts. Example: Demand forecasting for a new technology Are used to fit causal and time series methods. Example: Observance of a unique promotion by time series Methods Manager s opinion Sales estimate Customer market survey Expert opinion Delphi method Scenario techniques Page 12

13 Forecasting Methods Causal Methods Application Can be used, if the demand for a product or service can be forecasted on the basis of a known parameter. Example: Demand for cabs subject to the population of a city In general, a regression analysis estimates the parameter of a function so as to give a "best fit" of the data. Example: Demand for cabs = population/1000 Methods Linear regression Non-linear regression Page 13

14 Forecasting Methods Time series Application Construction of a forecast on the basis of observed demand over time. The operator estimates or supposes the behavior of the demand (constant level, trend,...). Different forecasting methods for different demand behaviors. Methods (Simple) Moving average Simple exponential smoothing Linear regression Method of Holt double exponential smoothing Method of Winters triple exponential smoothing Page 14

15 Sales Forecasting Time Series Notation t d t y t e t MSE t time period, t = 1, 2, historical data (demand values), t = 1,, T forecast of the values, t = 2, 3, forecasting error: e t = d t - y t, t = 2,, T Mean Square Error: also Mean Squared Average Page 15

16 Time Series Typical Time Series Patterns and Forecasting Models Constant Level (Simple) moving average Simple exponential smoothing Linear Trend Linear regression Holts Method double exponential smoothing Seasonal Effects Winters Method triple exponential smoothing Page 16

17 Constant Level (Simple) Moving Averages (MA) The forecast value y t+1 for period t+1 corresponds to the average of the previous r observations (demand values) Formula If t < r, set r = t. Remarks One of the most simple forecasting methods The forecast is easy to update from period to period The last r demands all have the same weight 1/r Special case: r = 1 Trivial forecast : Page 17

18 Moving Average Example Demand for portable televisions per week. Choose r = 4 t d t y t e t MSE 12 = Overestimation decrease Demand Underestimation increase Moving Average Page 18

19 Moving Average It is difficult to choose the right r. Small r (Extreme case r = 1) Forecasts respond to changes in demand very fast Large parts of the information about the demand in the past gets lost Large r (Extreme case r = hole demand history) Forecasts respond to changes in the demand very slowly Even very old data have the same weight as very recent ones Experimental determination of a good value for r r MSE Forecast for period t +τ: y t+τ = y t+1, τ = 1, 2, Page 19

20 Constant Level Exponential Smoothing (ES), also called Exponential Moving Average Consider all historical demands, but weight recent observations more than older ones. Forecasting formula α = smoothing factor, 0 < α < 1 y 2 = d 1 Interpretation Recursive initiation of y t into the formula above yields to Demands are multiplied with exponentially decreasing weights Page 20

21 Exponential Smoothing In addition y t+1 is a convex combination of d 1,, d t 0,2 α = 0.2 α 0,15 α(1-α) α(1-α) 2 0,1 Equivalent formulation 0,05 Interpretation The new forecast is a weighted sum of the preceding forecast and the most recent observation Page 21

22 Exponential Smoothing Example Demand for portable televisions per week. Choose α = 0.3 t d t y t e t MSE 12 = Demand Exp. Smoothing Moving Avg. Page 22

23 Exponential Smoothing Typical values for α are between 0.01 and 0.3 Small α Forecast responds to changes in demand slowly Also older data relatively strong weighted (though never as much as more recent ones) Big α Forecast responds to changes in demand fast Information of historical data gets lost very fast Experimental determination of a good α - value α MSE Page 23

24 Time Series Typical Time Series Patterns and Forecasting Models Constant Level Moving average Single exponential smoothing Linear Trend Linear regression Holts method double exponential smoothing Seasonal Effects Winters method triple exponential smoothing Page 24

25 Linear Trend Assumption The demand forecasting y t,t+τ in period t for period t +τ,τ > 1 is y t+τ = a t + b t τ The y-intercept a t is an estimator for the constant level and the slope b t for the trend. Estimation of the values via forecasting method. Forecasting methods for the constant level systematically underestimate the demand. Example MA and ES for r = 4 and α = 0.2. structural forecast error Demand Moving Avg. Exp. Smoothing Page 25

26 Linear Trend Linear Regression LR Basic idea Determine a line L with y-intercept a and slope b, which minimizes the mean square error Demand Slope b over all observations. MSE(a,b) is a convex, continuous differentiable function in a and b 5 Y-intercept a Differentiate with respect to a and b. Setting the derivative to zero yields the optimal values e t Time Page 26

27 Linear Regression a t and b t are computed based on the regression line, which interpolates the r > 1 latest demands best. Result for t 2 as optimal values for the y-intercept and the slope. Remark If 1 < t < r, set r = t. If t = 1, set y 2 = d 1 Page 27

28 Linear Regression Example Demand for portable televisions per week. Choose r = 4 t d t y t e t MSE 12 = Demand Regression Page 28

29 Linear Regression Problems Complex formula All periods are equal weighted Updating the parameter is cumbersome Special case: r = 2 Again an experimental determination of a good r-value r MSE Page 29

30 Trend Holds Method Double Exponential Smoothing Similar to simple exponential smoothing, but with a smoothing factor for the y-intercept (α) and one for the slope (β) of the forecasting function. Parameter whereas t 2, 0 < α < 1 and 0 < β < 1. Interpretation a t b t Combination of the new observation and the preceding forecast (identical to simple exponential smoothing) Combination of the difference between the new constant level and the previous one and the preceding estimator for the trend Page 30

31 Method of Holt Double Exponential Smoothing Remarks Initial values: a 1 = d 1 and b 1 = (d T - d 1 ) / T-1 For β = 0 one gets the simple exponential smoothing as a special case. New estimators are simple to compute Consideration of older demands with exponentially decreasing weights Experimental determination of good values for α and β. Page 31

32 Method of Holt Double Exponential Smoothing Example Demand for portable televisions per week. Choose α = 0.3 and β = 0.4. t d t y t e t MSE 12 = Demand Holt Regression Page 32

33 Time Series Typical Time Series Patterns and Forecasting Models Constant Level Moving average Single exponential smoothing Linear Trend Linear regression Holts method double exponential smoothing Seasonal Effects Winters method triple exponential smoothing Page 33

34 Time Series Seasonal Effects If there exist seasonal effects in the run of the demand curve, there is a demand pattern, which repeats every P periods. General idea Decouple the seasonal component in the demand curve from the trend component Demand a t + b t τ Trend Component (a t + b t τ)c t s 1 s 2 Seasonal Component Page 34

35 Seasonal Effects Forecasting formula s t+τ P seasonal factor length of the seasonal period Computation of a t, b t and s t+τ whereas t > P and 0 < α, β, γ < 1. Page 35

36 Seasonal Effects For the recursion it is necessary to know the Initial values a 0, b 0 and s t, t = 1,, P Demand values for N 2 seasonal periods Then where average slope of the trend function average demand in the i-th seasonal period Y-intercept of the trend function Non-normalized seasonal factors Page 36

37 Sales Forecasting Quality of forecasts Various measures for the forecasting error Mean Squared Error Oftenused Good theoretical properties Mean Absolute Deviation Intuitively better to interpret than MSE Mean Absolute Percentage Deviation Allows to estimate the quality of the forecasting method: 10% very good > 10%, 20% good > 20%, 30% medium > 30% bad Page 37

38 Sales Forecasts Forecasting Control The tracking signal shows structural forecasting errors, i.e. permanent over- or underestimation. s max continuous overestimation -s max Reconsider the choice of the parameters or the forecasting method Page 38

39 Strategic Supply Chain Management Strategic Supply Chain Management Content Topic Sales Forecasting Cost Factors & Data Aggregation Strategic Supply Chain Model from Practice Page 39

40 Cost Factors & Data Aggregation Cost Factors Transportation costs are one of the most important aspects in the planning and optimization of supply chains. Depend on Freight rate Quantity Distance Freight rate depends on Product conditions: size, manageability, vulnerability Means of transport: truck, tank lorry, rail, airplane Batch size: full truckloads are cheaper than less than truckloads or package freight Freight rate can be taken from tables for each product. Page 40

41 Transportation Costs Freight costs per unit generally decrease inversely proportionally (possibly with steps) with the quantity increase (piecewise) linearly with the distance The considered distance is based on Street (rail) distance Euclidean distance Can be computed with geographical information systems (GIS). Problem with the Euclidean distance Underestimates the real street distances Multiply distance with correction factor: Urban areas: 1.14 Europe, general: 1.3 Page 41

42 Cost Factors Inventory Costs Inventory costs basically consist of three cost components: Handling costs Consists of labor costs and material costs, which are proportional to the stock turn-over. Storage costs Encompasses the costs for the inventory, which are proportional to the average positive stock of inventory. Fixed costs Comprise absolute and step-wise costs, which are proportional to the size of the warehouse, but not to the quantity of goods. Handling costs are easy to determine. However it is a problem to calculate the average positive stock of inventory as well as the size of the warehouses. Page 42

43 Inventory Costs In the case of strategic planning for several years the data is extrapolated A result of a strategic planning may be: The estimated stock turn-over of a new warehouse has an amount of units per period. Question What is the average positive stock of inventory? How much capacity is needed in the warehouse, to guarantee the operational business of the inventory at any time? Should be proportional to the maximum stock/ handling quantity. One can use the inventory turn-over ratio ITR to compute the average positive stock of inventory. Inventory turn-over ratio for several product types known. Page 43

44 Inventory Costs Reference Simchi-Levi et al., 2003 Considering the relation the average positive stock of inventory per period can be determined as and therewith the inventory costs. Page 44

45 Inventory Costs Size of the Warehouse The required capacity for storage equals approx. twice the average stock of inventory. Furthermore room for offices, packing, handling goods and so on needed. Real size of the warehouses equals approx. three times the pure inventory capacity. Therewith the fixed costs for the inventory can be calculated. Page 45

46 Cost Factors & Data Aggregation Data Aggregation If one models the complete supply chain, one often has to consider many thousands of customers and products. high effort to obtain and handle detailed data aggregate data for planning and optimization Aggregation also for smaller supply chains useful, since data not always fully available forecasts for the trend of demand and costs often imprecise Keyword: Risk Pooling Effect Since the data aggregation reduces the variability, the forecasts for demand are more accurate on an aggregated level. Page 46

47 Data Aggregation Aggregation Error Planning with aggregated data and original data, respectively, leads to different costs and results. Consider the trade-off for less exact results due to the aggregation and unnecessary high complexity Two classical alternatives for aggregation customers (demand) products Page 47

48 Data Aggregation Aggregation of Customers (Demand) Clustering is based on Geographical position Aggregate geographically close customers to a customer zone. All customers within a single cell/cluster are replaced by just a single customer located at the center of this cell/cluster. Aggregation for example based on - Network Grid Aggregate all customers within the same grid cell. - Zip codes Aggregate customers according to zip code (e.g. all customers where the first two are the same). Page 48

49 Aggregation of Customers Aggregation of American ZIP-code areas (Simchi-Levi et al., 2003) digit zip codes digit zip codes Page 49

50 Aggregation of Customers Similar characteristics Aggregate customers with - similar service requests - the same supply frequency Efficiency of aggregation depends on the number of customer zones distribution of the customers within the zones Recommended at least 300 customer zones similar customer demand in each zone Page 50

51 Aggregation of Customers Example (Simchi-Levi et al.) Locating factories. Only transportation costs considered. Total costs: $ Number of customers: Total costs: $ Number of customers: 800 Aggregation error < 0.05% Page 51

52 Data Aggregation Aggregation of Products Clustering is based on Model similarity Aggregate products, which only have marginal differences (e.g. variations of the same model: color, equipment details or type of packaging). Distribution Sample Aggregate products, which are produced/ loaded in the same facility and delivered to the same customers. Further alternatives Aggregate products with the same weight, same size, same shipping method (unit load, frozen cargo), Page 52

53 Aggregation of Products Example for model similarity (Fraunhofer ITWM) Various intense aggregation of electro-closets coated, 2 doors coated, 2 doors not coated, 2 doors not coated, 1 door coated, 1 door coated, 1 door coated, 1 door Page 53

54 Aggregation of Products Example for logistics characteristics (Simchi-Levi et al.) Aggregate products with similar weight-volume ratio Weight (lbs per case) Rectangle denote clusters Volume (pallets per case) Page 54

55 Aggregation of Products Example (Simchi-Levi et al.) 5 factories, locating warehouses. Total costs: $ Number of products: 46 Total costs: $ Number of products: 4 Aggregation error < 0.03% Page 55

56 Strategic Supply Chain Management Strategic Supply Chain Management Contents Topic Sales Forecasting Cost Factors & Data Aggregation Strategic Supply Chain Model from Practice Page 56

57 Strategic Supply Chain Management Strategic Supply Chain Model from Practice Consider a model with capacities several periods several products several highly general, non-hierarchic levels Page 57

58 Strategic Supply Chain Management Make decisions on locations procurement and production storage and distribution satisfaction of customer demands in consideration of costs for procurement and production warehousing (inventory, stock turn-over) opening and closure of facilities transportation unsatisfied customer demands to minimize these costs. Page 58

59 Strategic Supply Chain Model from Practice Components of the model Facilities Very general term. May be anything i.e. customers, Warehouses, factories, production lines, cross-docks etc., and make everything i.e. produce, store, handle products, consume, etc. Page 59

60 Strategic Supply Chain model from practice Uniquely defined relationship between facilities and locations. There already exist a facility at a location or this location is a candidate for a new facility. Distinguish facilities in Selectable ones Subject of the planning and may change their status, i.e. can be opened or closed. Typically: factories, distribution centers, Not selectable ones Fixed. Typically: suppliers, customers, as well as factories, inventories, which should be maintained In the following we only talk about facilities. Page 60

61 Strategic Supply Chain Model from Practice The facilities modeled in the supply chain do not necessarily have to be part of the own organization. E.g. external supplier. Notation L = Set of all facilities S = Set of all selectable facilities S o = Set of all selectable facilities, which could be opened S c = Set of all selectable facilities, which could be closed S = S o S c. = Set of all products = Set of all time periods Page 61

62 Strategic Supply Chain model from practice Location decisions Open Close OC t l = Opening Costs for the facility l S o at the beginning of period t and for its operation for the rest of the planning period. CC t l = Costs for closure of the facility l S c at the end of period t and their operation until then. Page 62

63 Strategic Supply Chain Model from Practice Demand Facilities can have demand for products. Forecast future demands. If the forecasts are not exact enough, then consider the problem several times for different scenarios of demand trends - pessimistic (worst-case) - normal (average-case) - optimistic - Notation D t l,p = Demand in quantity units for product p at facility l in period t. Page 63

64 Strategic Supply Chain Model from Practice Satisfaction of customer demands It may be that the demand can/shall not be (completely) satisfied. Example: - Costs for the satisfaction of demand is too high (compared to profit) - Supply within the given service time is not possible, or just with very high efforts - Capacities are not sufficient Unsatisfied demand incurs penalty costs. However, they are difficult to quantify. Possibility: lost profits, service level which has to be satisfied z t l,p = Number of quantity units of demand at facility l for product p in period t, which were not delivered. PDC t l,p = Penalty costs per quantity unit of product p, which were not delivered to facility l in period t to satisfy the demand. Page 64

65 Strategic Supply Chain Model from Practice Procurement Facilities can buy products from external, i.e. from external suppliers. Example - raw material or semi-finished goods, which can not be produced at the own facilities - products, that are cheaper to buy than to produce them (Out-Sourcing) Notation b t l,p = Amount of product p, which is procured at facility l in period t. BC t l,p = Costs for procurement of one unit of product p at facility l in period t. Page 65

66 Strategic Supply Chain Model from Practice Production Manufacturing of finished goods from different inputs. Example - classical manufacturing of finished goods in factories from raw material and intermediate goods - packaging or picking products in distribution centers. E.g. drill machine from factory A with boring head from factory B packed together. Manufacturing processes are specified by lists of materials. Example Intermediate product Z1 is manufactured by raw material R1 and R2. The numbers on the arcs indicate the related material consumption factors. The production of one unit of Z1 needs 2 and 1.5 units of raw materials Z1 and Z2, respectively 2 R 1 Z R 2 Page 66

67 Strategic Supply Chain Model from Practice Simplify multi-stage lists of material to single-stage ones. P P 1 1 R 1 Z 1 2 R 2 Z R R 1 R 2 R 3 6 Notation a l,p,q = Number of units of product q, needed to manufacture one unit of product p in facility l. h t l,p = Amount of product p, produced in facility l in period t. HC t l,p = Costs for manufacturing one unit of product p in facility l in period t. Includes costs for material, machine utilization,. Page 67

68 Strategic Supply Chain Model from Practice Storage Products (raw material, intermediate products, finished goods) can be stored in facilities from one period to the next. Notation inv t l,p = Amount of product p, stored at facility l in period t. IC t l,p = Costs for storing one unit of product p at facility l in period t. Include costs for inventory, stock ground, Page 68

69 Strategic Supply Chain Model from Practice Distribution Transportation link between all facilities possible. Notation x t l,l,p = Amount of product p, transported from facility l to l in period t. TC t l,l,p = Transportation costs for one unit of product p from facility l to l in period t. The transportation costs depend on the distance, but also on the product and the means of transportation. Include often costs for goods issue (e.g. order picking, shipment) at the starting facility and for incoming goods (warehousing) at the destination facility. Sometimes costs for storage (within a period) at the starting location, too. Page 69

70 Strategic Supply Chain Model from Practice Capacities Displayed via resources. Example - machine, stockyard - storage-, order picking system - staff, shift Resources characterized by Base capacity (e.g. production capacity of a machine, maximal throughput of the picking system per period). Consumption factor states for each product the consumption of resources in resource units per quantity unit of a product. Expansible capacity of the resource (e.g. overtime, leasable storage or production capacity). Penalty costs per unit, that extend (overload) the base capacity. Page 70

71 Strategic Supply Chain Model from Practice Relations between facilities and resources one to many The same resource can be used on several facilities. Example: executive producer, which is responsible for several production lines one to one The same resource is used by all products of one facility. Example: flexible configurable machine many to one Several resources attached in the same facility. Example: facilities correspond to production lines and resources to executive producers Consider resources for - production and - incoming goods and goods issue (handling) Page 71

72 Strategic Supply Chain Model from Practice Notation R p = Set of production resources R h = Set of handling resources μ l,r,p = Consumption factor of production resource r R p per unit of product p at facility l. λ i l,r,p, λ o l,r,p = Consumption factor of production resource r R h per unit of product p at goods receipt respectively issue at facility l. v t r = Number of units, the resource r R p R h has been extended in period t. RK t r = Base capacity of resource r R p R h in period t. ERK t r = Maximally allowed extension of the capacity of the resource r R p R h in period t. RC t r = Penalty costs per extended resource unit of resource r R p R h in period t. Page 72

73 Strategic Supply Chain Model from Practice Mixed integer linear program Objective function procurement and production distribution resource extension unsatisfied demand inventory costs location decisions Page 73

74 Strategic Supply Chain Model from Practice Constraints Flow conservation procurement incoming transports production inventory last period incoming goods outgoing goods outgoing transports consumption demand to production inventory this period unsatisfied demand Page 74

75 Strategic Supply Chain Model from Practice Resources Production Handling Incoming goods and goods issue, respectively, by transports Feasibility of extension goods issue by procurement Page 75

76 Strategic Supply Chain Model from Practice Location decisions Selectable facilities can be opened and closed, respectively, only once and Define Activities at selectable facilities Procurement Page 76

77 Strategic Supply Chain Model from Practice Production Storage Outgoing distribution Incoming distribution Page 77

78 Strategic Supply Chain Model from Practice Integer and non-negativity constraints Page 78