1. 1 My name is Mikio Kubo from Tokyo University of Marine Science of Technology. The title of this talk is Trend in Supply Chain Optimization and Humanitarian Logistics.

2. 2 This is the agenda of my talk. I will start with the definition of the supply chain and logistics, and introduce three decision levels of the SC,and then show you the classification of inventory. Next I ’ m going to talk about several models in the SC; they are logistics network design, inventory, production planning, and vehicle routing. And finally, I will talk about the SC management and humanitarian logistics, if time allows.

3. 3 There are many definitions of the Supply Chain. But my definition is simple. IT (that means the information technology) ＋ Logistics ＝ Supply Chain To complete the definition, we need the definition of logistics. As shown in this figure, logistics optimizes the flow of products between the point of origin (here supply point) and the point of consumption (here demand point) in order to meet customers' requirements.

4. 4 In general, the supply chain is composed of three systems: One is the real system that includes real logistics objects such as trucks, ships, plants, products, machines, etc. Using the metaphor of human beings, the real system is compared to muscle of the body. Another is the transactional IT; for example, POS that means Point-Of- Sales, you can see such a system in convenience stores or supermarkets, ERP that means Enterprise Resource Planning that is an extension of legacy MRP that means Material Requirement Planning, DRP that means Distribution Requirement Planning, etc. The transactional IT is compared to a nerve net of human beings that just flows the pulse and executes automatic actions. The other is the analytic IT composed of some models and algorithms to solve them. The analytic IT is compared to the brain of human beings. That ’ s the main theme of this talk.

5. 5 The decision support systems or analytic IT models can be categorized into three levels. The top level is the strategic level that deals with decisions having a long-time effect that spans from a few years to 10 or more years. The tactical level includes decisions which are typically updated between once every month or quarter, or once every year. Finally, the operational level refers to day-to-day or real time decisions. Using some metaphors, the strategic level is compared to seeing the forest, while the operational level is compared to seeing the tree. Such a change of the point of view is important for the decision making using analytic IT models.

6. 6 This figure represents an entire supply chain; Procurement of parts or raw materials from suppliers, production at plants, stocking in DCs or warehouses, and distribution to retailers.These are the decision levels; strategic, tactical and operational that means long-term, middle-term, and short- term, respectively.

7. 7 In the strategic level, we have an analytic IT model named the Logistics Network Design that encompasses the whole logistics network and determines new suppliers, a new flow pattern of products through the network, a selection of warehouse locations and capacities, the production levels at each plant or the production line in order to minimize total production, inventory, and transportation costs that means whole supply chain costs.

8. 8 In the tactical and operational levels, we have 3 types of models; they are inventory, production planning and transportation delivery. The safety stock allocation model determines the positions of safety stocks in the supply chain, and simultaneously determines the pull-push boundary of each product. The inventory policy optimization model decides at what point to reorder and how much to order so as to minimize inventory ordering and holding costs. Production planning optimizes the acquisition of resources such as machines or workers, the size of the production lots to be manufactured or processed in a batch and the sequencing of the production lots. Finally, transportation delivery optimization involves the routes and frequencies of vehicles such as ships or trucks. The most important transportation delivery model is the vehicle routing problem that occurs at the final part of the supply chain.

9. 9 Inventory plays an important role in the supply chain. Using the metaphor of human body again, inventory is compared to blood and it acts as glue connecting several processes in the SC. Inventory is spread throughout the SC from raw materials, work-in-process inventory, to finished products in warehouses and retailers.And such inventory varies day by day or period to period due to many reasons.

10. 10 To understand the role of inventory in the SC, we need to classify it into 5 categories by their motivations; They are in-transit inventory, seasonal inventory, cycle inventory, lot-size inventory, and safety inventory. We have to optimize the trade-offs between inventory and some other factors.

11. 11 In-transit inventory is inventory moving through the arc or link in the SC. It can be seen as a flow thorough a pipeline; so it ’ s sometimes called pipeline inventory. Such inventory exists ‘ cause transportation time is positive. In-transit inventory is proportional to the flow volume and transportation time. That means if the speed of the flow is faster, in-transit inventory becomes smaller. This inventory will be treated and be optimized in the logistics network design model.

12. 12 The main role of inventory in the SC is to fill the gap between supply and demand. Seasonal inventory is inventory to counter predictable seasonal demand under the restriction of the limited production resources. For example, a can maker supplies tons of cans to beverage makers such as beer companies. For a high demand during the summer season, the can maker builds up inventory during low demand periods and store it. This is seasonal inventory.

13. 13 Assume that demand is constant and supply arrives cyclically. In this case, inventory changes like saw- teeth. Here, “ saw ” is a tool for cutting woods whose teeth is like this figure. That ’ s cycle inventory. Many activities in the SC have economy of scale; that means large lots decrease the cost. This is the motivation of cycle inventory. If transportation activity has a non- negative fixed cost, we have to have cyclic inventory, which is optimized in the LND model. If ordering activity has a non-negative fixed cost, we have to have cyclic inventory, which is optimized in the EQO model.

14. 14 Lot-sizing inventory is a generalization of cycle inventory when the speed of the demand is not constant. That means the demand changes over time, period by period. As in cycle inventory, lot- sizing inventory has the trade-off for fixed costs, especially production set-up costs and is optimized in lot-sizing and multi-period logistics network design models.

15. 15 Safety inventory or safety stock is inventory to protect against uncertainties of future events such as customer demands. So it has the trade-off for the customer service level. If customers request 100% of service level that means no stock out, retailers need a huge amount of safety inventory. Service level is determined by backorder or stock-out penalty costs. So there ’ s a trade-off between safety inventory costs and backorder costs. Many analytic IT models include safety inventory. They are safety stock allocation, logistics network design, and inventory policy optimization models.

16. 16 Inventory can be classified into these 5 types of inventory, in-transit, seasonal, cycle or lot-size, and safety inventory. It ’ s not so easy to treat them separately in practice. But we should optimize them separately using some analytic IT models. I ’ ll talk about such models until my lecture time is up.

17. 17 The first analytic IT model is the logistics network design which involves issues relating to plant, warehouse, and retailer locations. These are strategic decisions because they have a long-term effect on the company. The objective is to design or reconfigure the logistics network so as to minimize annual system wide costs, including production and purchasing costs, inventory holding costs, facility costs that include storage, handling, and fixed costs, and transportation costs. There are many decisions optimized in this model. For example, Where should we replenish parts? In which plant or on which production line should we produce products? Where and by which transportation-mode should we transport products? Where should we construct (or close) plants or new distribution centers?

18. 18 There are many trade-offs in the model: for example, increasing the number of warehouses typically yields: -An improvement in service level due to the reduction in average travel time to the customers. -An increase in inventory costs due to increased safety stocks required to protect each warehouse against uncertainties in customer demands. -An increase in overhead and setup costs. -A reduction in outbound transportation costs: transportation costs from the warehouses to the customers. -An increase in inbound transportation costs: transportation costs from the suppliers and/or manufacturers to the warehouses.

19. 19 Another trade-off in the LND model is between the in- transit inventory cost and the transportation cost. In- transit inventory is proportional to the flow volume and transportation time. So if we use a fast transportation mode such as a plane that is expensive, in-transit inventory becomes smaller and decrease the inventory cost. Meanwhile, if we use a slow transportation mode such as ships, in-transit inventory becomes larger. That ’ s the trade-off and should be optimized in the LND model that selects the appropriate transportation mode for each link by minimizing the sum of the transportation and in-transit inventory costs.

20. 20 We can extend the basic LND model to the multi-period LND model. Let us consider situations where demand changes over time because of seasonal factors, or growing or shrinking markers. It also allows for time-varying purchase, production, transportation, and inventory costs. The problem is to find the right balance between the cost of holding seasonal inventory and the other costs. This model is in the tactical decision level and can be seen as an extension of the master production system (MPS) usually used in a plant for determining the production level.

21. 21 The multi-period LND model optimizes the trade-off between seasonal inventory cost and overtime penalty cost. Assume that demand has a seasonal peak, say in summer and our plant has a limited resource. So we cannot supply the peak demand by the production during the peak periods. One strategy to cope with the peak demand is the so-called constant production strategy using seasonal inventory. That means the company builds up inventory during low demand periods, say in spring, and stores it for the summer, the peak season. Another strategy is to vary the production level by hiring new workers or by doing overwork; both require additional costs. The multi-period LND model finds the best balance of these strategies.

22. 22 The LND model can be formulated as a multi-commodity network flow problem. Supply chain or logistics network is modeled as a network composed of nodes and arcs. The raw materials, parts, intermediates, final products or items are modeled as commodities or flow through the network. We have to model the bill-of-material or recipe structure that represents the relationship between the items. That means which product is composed of which parts or raw materials. We also incorporate the safety inventory cost into the model that is a concave function of the flow volume through the arc due to “ statistical economy of scale ” that we ’ ll discuss later. Usually, the concave cost minimization problem is difficult; of course, NP-hard. But by using approximation of concave functions, the LND problem can be formulated as a mixed integer programming problem and can be solved by standard MIP solvers such as Gurobi or CPLEX.

23. 23 The second model is the safety stock allocation model in the tactical level that optimizes the trade-off between the safety inventory and the customer service level.

24. 24 This model is based on a basic principle of inventory called the statistical economy of scale; that means gathering inventory together reduces the total inventory volume and cost. This principle is used in many modern supply chain strategies such as risk pooling, delayed differentiation, and design for logistics. The safety stock allocation model answers how to use these strategies and also answers the following questions; where to keep safety stock? which facility should produce to stock? and which facility should produce to order?

25. 25 The safety stock allocation model is based on the classical formula to compute the safety inventory volume. Assume that the demand follows the normal distribution with average demand μ and the standard deviation σ. Service level is the probability of no stocking out and is determined by a decision maker. Lead-time is the amount of time that elapses from the instant that an order is placed until it arrives. So the maximum inventory volume is computed by the formula μ× Ｌ plus safety stock ratio×σ× square root of L.. Here, safety stock ratio can be computed from service level. If the service level is 95%, then the safety stock ratio becomes 1.65. Actually this formula is a special case of the classical newsboy problem that I will talk later.

26. 26 This figure represents the relationship between the lead-time and the average, safety and maximum inventory. Horizontal axis is the lead time and the vertical axis is the inventory volume; average inventory (drawn in black line) is a linear function, while maximum inventory (drawn in pink) and safety inventory (drawn in yellow) are both concave that represents the statistical economy of scale.

27. 27 One characteristic of the safety stock allocation model is to treat lead-time as a variable instead of a given constant. First we introduce the guaranteed lead-time that is a committed time that the facility guarantees to deliver to its customers. In this example, this facility i guarantees to deliver to its customers within 2 days represented by a yellow arrow in the upper right corner. So the guaranteed lead time denoted by L_i is 2. This facility has another lead time called the entering LT that is the GLT of his predecessor or upstream (supply) facility. In this example, the entering LT of facility i represented by a yellow arrow in the lower left corner is 1 ‘ cause its predecessor has GLT 1 day.

28. 28 The facility has a given constant production time that is the amount of time that elapses from the instant that the item or product arrives until it is ready to ship. In this example it is represented by a green rectangle in the lower right corner and it is 3 day. The net replenishment time is defined by entering LT + production time – guaranteed LT. In this example the net replenishment time represented by the red arrow in the upper left corner is 1+3-2=2. The inventory of the facility is set to the maximum demand volume during the net replenishment time to satisfy the service level.

29. 29 Next I ’ ll show you a mathematical programming approach for solving the safety stock allocation problem in general network. The main variable is x that represents the net replenishment time. The objective function is nonlinear. Here D is the maximum demand function. It is a general nonlinear function. D minus μx gives us the safety inventory volume and by multiplying it by holding cost h_i, we get the safety inventory cost. The objective function is the sum of these nonlinear functions over all facilities. The first constraint defines the net replenishment time is equal to entering LT LI_i (it is a variable)+ processing time T_i (it is a constant) – guaranteed lead time L_i (it is also a variable). The second constraint means that the entering lead time of facility j is greater than or equal to the guaranteed lead time of facility i if there exits an arc between i and j, i.e., facility i is a supplier of facility j. The third constraint restricts the lower and upper bounds of the guaranteed LT, and finally, the forth constraint defines the non-negativity of the net replenishment time. This is a nonlinear programming problem. Generally it ’ s quite hard to solve it (theoretically, it ’ s NP-hard problem) but by using some piecewise linear approximations of non-linear functions, we can solve the problem by standard MIP solvers.

30. 30 In summary. The safety stock allocation problem can be solved by several types of algorithms. One is the mathematical programming approach that relies on MIP solvers. It can handle middle-size instances with general networks. Another is based on dynamic programming. We can extend the DP algorithm we discussed to a more general tree network case. But, unfortunately, we cannot extend the DP approach to general networks,. Other approaches are metaheuristics such as local search, iterated LS, or tabu search, etc. They give approximate solutions instead of the exact solutions but it ’ s usually faster and more robust.

31. 31 Here is a real example reported in the book written by David Simchi-Levi of MIT. Final demand (part 1 in this figure) is sold in Dallas that has a normal distribution with average 100 and standard deviation 10 and customer ’ s guaranteed LT is 30 days. Many parts are required to produce the final product and they are produced in many facilities. By optimizing the allocation of safety inventory, they could save more than 43 thousand dollars. It ’ s about 40 % cost down from the baseline model. The safety stock allocation model can be used in so-called “ what if ” analysis, too. For example, what if we change the guaranteed lead-time to the customer from 30 days to 15 days that means an improved customer service. The model answers the question. The cost increases to 51 thousand dollars.

32. 32 Next we will turn our attention to another inventory model called inventory policy optimization. Inventory policy optimization is in the operational or tactical decision level. We first introduce 2 classical models; newsboy model and economic ordering quantity model. The newsboy model introduced by Scarf in 1960 is a framework where we optimize the trade-off between lost sales and safety inventory costs. The EOQ model introduced by Harris in 1915 (not 50, it ’ s a very old model) is a framework where we optimize the trade-off between fixed ordering and cycle inventory costs.

33. 33 Newsboy model can be extended to multi period model in which the ending inventory at a period becomes the starting inventory of the next period. In this case we have to monitor the inventory position that is defined as the sum of in-hand or local inventory + inventory on order (that means the amount of orders that has not been arrived yet) – backorder (that means demand not satisfied now but carried over to future). We determine the ordering quantity so that the inventory position becomes a pre-determined value called the base stock level; that is the base stock policy

34. 34 If the fixed ordering cost is positive, the base stock policy is no more optimal. In such cases, we need modified base stock policies, namely (Q,R) policy and (s,S) policy. In (Q,R) policy, we monitor the inventory position and if it reaches a re-order point R, we order a fixed quantity Q. Here R and Q are system parameters. We may use EOQ model to compute the ordering quantity Q. In (s,S) policy, we again monitor the inventory position and if it is below a re-ordering point small s, we order the amount so that the inventory position becomes an order-up-to level capital S.

35. 35 This figure shows the change of inventory levels of (Q,R) and (s,S) policies. Here is a re-order point R (or small s) and here is the order-up-to level Q+R or capital S. When the inventory position becomes R, both policies order the amount Q and the inventory position becomes R+Q or S. After the lead time, the order arrives and the in-hand inventory increases and coincides with the inventory position. If the bulk demand occurs and the system orders below the re- order point, the (Q,R) policy orders Q while the (s,S) policy orders more so that the inventory position becomes capital S. Under some reasonable assumptions, (s,S) policy is proved to be optimal. But I believe (Q,R) policy is more practical in Japan ‘ cause the ordering lot-size is fixed in many companies.

36. 36 In the base stock or (Q,R) or (s,S) policies, we treat the time as continuous variable. But in many situations the timing of decisions is restricted to a discrete time. In such cases, discrete time model is more appropriate. We now turn our attention to the discrete time inventory model, namely the periodic ordering policy. In this policy, we check the inventory periodically, say once a day. If the inventory position is below the base stock level, we order the amount so that it recovers the base stock level. We assume the demand occurs during the day, and the ordering quantity is determined at the end of the day, and the order arrives at the beginning of the day after the lead time.

37. 37 We have talked about the inventory policy optimization. For the continuous time models, we get the optimal base stock via dynamic programming and we can get similar algorithms for (Q,R) and (s,S) policies. For the discrete time model, called the periodic order policy model, we used a simulation based optimization algorithm. Such an approach is called the infinitesimal perturbation analysis. I believe this approach will be one of the candidates for solving real inventory problems.

38. 38 Next we consider a model in production planning. The first model is the lot-size optimization model. This model supports the decision maker in the tactical level and optimizes the trade-off between the set-up cost and lot-size inventory. This model can be seen as an extension of the EOQ model in which the customer demand is not necessarily constant.

39. 39 To solve the real lot-sizing problems is very difficult. One approach is to use the MIP solver using strong formulation. Another approach is to construct heuristics or meta-heuristics. The other approach is the MIP-based meta- heuristics. The example of such approaches are: The relax and fix, capacity scaling, and MIP based neighborhood local search.

40. 40 Next we consider the second model in production planning. That is the scheduling optimization model. This model supports the decision maker in the operational level and optimizes the allocation of activities over time under the constraints of finite resources. Here activities mean jobs or tasks or operations to be executed (represented by rectangles in this figure). In this example, we have 3 machines as finite resources.

41. 41 The key concepts of the scheduling optimization are the activities and resources. There are many resources in production lines; machines, workers, raw materials, or money can be seen as scare resources. The scheduling must satisfy some precedence constraints between activities. In this example the red activities have to be executed in this order. The red activity on machine 2 cannot start unless the red activity on machine 3 finishes, and also the red activity on machine 1 cannot start before the finish time of the red activity on machine 2.

42. 42 The history of the scheduling theory is long; so many researchers have been proposed a number of algorithms. The first category of algorithms is myopic heuristics such as active scheduling scheme, non-day scheduling scheme, or dispatching rules. Such heuristic algorithms are heavily used in practice. Theoretically constraint programming and metaheuristics and their hybrid are well-studied recently.

43. 43 Finally, I ’ ll show the model for transportation and delivery, called the vehicle routing problem. The vehicle routing occurs at the final part of the SC; the distribution from the DCs or warehouse or depots to retailers or customers. A set of customers has to be served by a fleet of vehicles with limited capacities. The vehicles are initially located at a given depot. The objective is to find a set of routes for the customers and returns to the depot without violating the capacity constraint. In many distribution systems, each customer specifies, in addition to the load volume to be delivered to it, a period of time, called a time window that specifies the earliest and latest time to start the service.

44. 44 As in the scheduling problem, many algorithms have been proposed for the vehicle routing problem. The classical approaches are: the saving method by Clarke and Wright, the sweep method by Gillet and Miller, Insertion method, and local search. Recently many mataheuristic algorithms are used for solving difficult real problems.

45. 45 This figure shows the family tree of the vehicle routing algorithms. In 1970 or before, many ad hoc methods have been proposed such as saving, insertion, sweep, and local search.Local search is brushed up to metaheuristics such as tabu search, simulated annealing, and then more general adaptive memory programming that also includes genetic algorithm as a special case. Sweep method that can be seen as a cluster-first route-second approach is refined to generalized assignment heuristics, location based heuristics that is proved to be asymptotically optimal under some assumptions. Saving and insertion methods that can be sees as construction algorithms have a descendant named GRASP that is a construction type metaheuristics. And we have another branch of exact methods such as set partitioning approach, cutting plane, etc.Recently these algorithms are unified to the hierarchical building block method proposed by me.

46. 46 Recently, we Japanese had a large disruption caused by an earthquake, and then Tsunami, and finally an explosion of nuclear plants in Fukushima. Many supply chains are stopped; so we recognized supply chains should be not only efficient but also robust with respect to such disruptions. Supply Chain Risk Management (SCRM) is a new area of SCM to copy with the supply chain disruptions. This figure represents the change of the performance of a supply chain before and after a disruption event. The action before the disruption is called “ proactive ”, while the action after the disruption is called “ response ”. Our aim is to use the supply chain optimization models so that the supply chain becomes robust w.r.t the disruption and recovers quickly after the disruption.

47. 47 The SCRM has been becoming important recently. There are many reasons. The first one is the increase of disasters. These 10 years, we – human beings- encountered many disasters. Examples of natural disasters are: Non-natural that means man-made disasters are: SARS (Severe Acute Respiratory Syndrome), BSE (Bovine Spongiform Encephalopathy), CBRNE (Chemical Biological, Radiological, Nuclear, Explosive) The other reasons are the trend in SCM such as lean SC and globalization of SC. The lean system makes the inventory low, so increases vulnerability of the SC, while globalization makes the lead time longer and the trend of outsourcing makes the supply un-stable.

48. 48 Two communities of business management and consulting are closely related to SCRM. One is the risk management that is defied as the identification, assessment, and prioritization of risks. Another is the Business Continuity Planning whose objective is to write a thick manual. But, unfortunately, a series of disasters in Japan proved that both did not work well, or useless. More important related area is the humanitarian logistics.

49. 49 Humanitarian Logistics can be defied as a branch of logistics which specializes in organizing the delivery and warehousing of supplies during natural disasters to the affected area and people. It is different from the usual (commercial) logistics with respect to the following points. It is decentralized and there are many players (government, self defense force, NGOs). And No SCM unit nor trained staffs Everything is ad hoc No performance measure (fairness, speed, … ) No information & communication technology

50. 50 To cope with risks, we have to classify them. The first approach named risk mapping classifies risks into 2-dimensional space; That is usually used in risk management. The horizontal axes is the impact of risks, while the vertical axis represents the frequency of risks. This figure shows a risk mapping of an imaginary company. For example. this company categorized typhoon and earthquake into the area that has large impact but rare. This red area represents that the impact is large and high frequency. That includes strike and the fluctuation of exchange rate. This yellow area represents that the impact is small but frequency is high. That includes line stops in the plant and supply delay. This white zone is that both the impact and frequency is small.

51. 51 Risks can be classified into supply, internal and demand risks. Environmental risk is outside of the corresponding SC.

52. 52 Another classification. Disaster risk that is caused by natural and man- made disasters Political risk that is caused by contracts, laws, regulations Social risk such as child labor / abuse Intellectual property risk related to patents, trademarks, copyrights Example of other types of risks are: Financial risk, employment risk, reputation risk, …

53. 53 Strategies to copy with risks are: 1)Accept the risk and do nothing. 2) Avoid the risk factor, if possible, 3) Transfer the risk using insurance or option 4) Alignment that means share risk with other SC partners by contract, 5) Or, finally, strengthen the supply chain, by adding desirable properties to the SC.

54. 54 Such desirable properties to strengthen the SC are:robustness, resiliency, redundancy,flexibility, compatibility for proactive strategies,and agility and visibility for response strategies. Such technical terms are not well-defied and sometimes called BUZZWORDS. We propose the robustness is defied as the depth of the valley. The time resiliency is the duration between the disruption and recovery. The performance resiliency is the percentage of the performance recovery.

55. 55 An example of redundancy is the strategic inventory that is the inventory for preparing the disruption. Also it is shard by many SC partners. Remark that it must be considered separately to the safety stock that is for the demand uncertainty.

56. 56 A concrete example of flexibility is the multiple sourcing strategy that procures from two or more suppliers. Make-and-buy strategy is another kind of multiple sourcing. Remark that these two suppliers should not be located in the area. Otherwise, both suppliers may be down.

57. 57 This is another concrete example of flexibility called the process flexibility. If each plant can produce exactly one products, it is called 1-flex. If each plant can produce 2 products like this, it is called 2-flex. Graves and Tomlin at MIT showed that 2-flex is the similar performance as Full-flexibility that means each plant can produce all the products. But recent simulation study showed that it is not true when the supply disrupts.

58. 58 The other types of flexibility is transportation flexibility. That is multi-mode (that means using ship and air), multi-carrier (that means using DHL and UPS) and also multi-route (that means using ship for the south coast (and then using ground transportation) and via Panama channel).

59. 59 An example of compatibility is the risk pooling strategy that is very common in modern SC. Before the Earthquake in Japan, we Japanese had too many types of caps of the bottles. After the Earthquake, caps became a bottleneck to produce bottles. So Japanese beverage makes recognized that the compatibility is important and switched to produce the white one only.

60. 60 Using the strategies to copy with risks, we can avoid, transfer and reduce the impact and probability of the risks.

61. 61 We can also use optimization techniques to the SCRM. Basically we can apply all the SC optimization models using what if analysis. For proactive decisions, strategic and tactical models are useful. For response decisions, we use operational models such as scheduling, transportation and vehicle routing models. If you want to model both decisions, we need to use the framework of stochastic programming in which decisions before the disruption is modeled as here and now variables, while decisions after the disruption is modeled as recourse variables.

62. 62 We have developed many SC optimization systems; we are now extending them for the SCRM. For example, LND is extended using stochastic and robust optimization framework. Safety stock allocation model is extended by incorporating sourcing decision. We are also developing quick and dirty solution system for the vehicle routing problem without using ITs. We wish such a non-IT system can be used in humanitarian logistics for last-mile delivery. Anyway, much remains to be done in this are. Thanks for your attention.