Supply Chain Game

Intel and Georgia Tech team up to develop interactive teaching tool that mimics SCM dynamics, including psychological and behavioral aspects.

Supply Chain Game

Intel and Georgia Tech team up to develop interactive teaching tool that mimics SCM dynamics, including psychological and behavioral aspects.

By Pinar Keskinocak, Shuangjun Xia, Mani Janakiram and Tosanwunmi Maku

Georgia Tech’s H. Milton Stewart School of Industrial and Systems Engineering (ISyE) and Intel Corporation collaborated in developing an Interactive Supply Chain Game, which is a computerized simulation that mimics the dynamics of a supply chain. The game is structurally simple yet rich in terms of the learning and insights that could be derived from it. Human players can play with each other or against computers that employ pre-programmed supply chain strategies. At the end of the game, players are expected to gain some insights regarding supply chain management, especially in the following areas:

  1. management of inventory and the importance of balancing underage and overage costs;
  2. the importance of maintaining a good relationship with one’s supplier to ensure timely delivery of products;
  3. the importance of maintaining a good relationship with one’s customers to increase future orders and improve the predictability of orders; and
  4. the impact of different kinds of inventory allocation rules on customers’ forecasting and ordering behavior.

The game is intended for anyone who has an interest in supply chain management.

Motivation for the Game

Changing market conditions over the past decade, e.g., increased product variety and shorter order lead times, have forced companies to rethink their value streams, from research and development to supply chain management, to better fulfill customer demand.

The high-tech industry, particularly the semiconductor industry, experiences cyclical changes arising from product innovation and technology advancement, as well as changing customer needs. Hence, profitably balancing supply and demand is a continuous challenge, which is exacerbated due to short product life cycles, competition, price and cost pressures. Operating in this environment, Intel’s supply chain is complex due to a combination of long processing cycle times, high demand variability and technology/manufacturing complexity. Intel’s “just say yes to customer” effort, along with improvement in internal supply chain efficiencies and effectiveness, has been well received by customers and by the industry, resulting in customer excellence scores and a recognition as one of the top 25 supply chain companies by Gartner. With a focus on customer responsiveness, asset utilization and inventory management, Intel has established collaborative demand/supply operations, implemented highly effective supply chain inventory optimization and maintained strong engagement with its customers and suppliers.

Intel’s aspiration for continuously improving its supply chain metrics and planning for various scenarios, including better collaboration with suppliers and customers, motivated the company to partner with Georgia Tech’s ISyE in developing the Interactive Supply Chain Game.

Several analytical models provide insights for understanding and managing (in part) the individual factors and decisions, such as inventory management, forecasting and supplier-buyer collaboration, which play a role in matching demand and supply. However, these models often do not incorporate the psychological/behavioral aspects.

The Interactive Supply Chain Game enables users to evaluate various contract policies and study interactive behaviors in a highly dynamic supplier/buyer environment. It also supports various what-if analyses to investigate analytical and behavioral strategies of customers and suppliers with the goal of developing win-win relationships.

The intent of the game is to enhance the supply chain strategic competency of operational managers. To this end, it will be used at Intel for training purposes. Beyond internal application, the game has been used for pedagogical and experimental purposes at Georgia Tech, with several scenarios designed to assess key supply chain questions relating to procurement, inventory signaling and allocation strategies. The game will also be made available for use at other universities.

Game Overview

In the game, a single supplier (manufacturer, e.g., a computer chip maker) sells a single product to multiple customers (e.g., OEMs or distributors). (A more advanced version now under development features multiple suppliers.) Customers receive the product from the supplier, process it, convert it into a finished good and sell it to the consumers (e.g., electronics retailers such as BestBuy, schools, businesses or personal computer buyers).

One key feature of the game is that the supplier has a production lead time, so there is a delay between the time when production starts and the time when the product is ready to be shipped to the customers. The time it takes to ship the product from the supplier to the customers and from the customers to consumers is very short compared to the supplier’s production lead time, and assumed to be zero in the game. The customers may also incur a processing time delay so that goods received from the supplier become available for shipping to consumers after a few periods. Figure 1 depicts the structure of the game.

Figure 1: Structure of the supply chain game.

Figure 1: Structure of the supply chain game.

The game is played in discrete periods, i.e., every player needs to perform some tasks in each period, and the game cannot proceed to the next period until everyone has performed the required tasks. The players either assume the role of the supplier or one of the customers. The tasks for the supplier are to specify the shipment quantity to each customer (subject to available inventory and other rules discussed later) and the production quantity in each period. The tasks for the customers are to specify the quantity to order from the supplier and the forecasts for future orders.

The forecast horizon (the number of periods into the future the customers provide forecasts to the supplier) is equal to the production lead time of the supplier. Forecasts are intended to give the supplier a rough estimate of future orders and can potentially be considered in the production decision in each period. Forecasts can also be considered as “soft orders” that can be changed as the game moves on; actual orders do not have to be the same as forecasts. However, both the supplier and the customer have access to the forecasts and the orders exchanged between the two, so that the accuracy of the forecasts is common knowledge to all players.

In the scenarios played so far, the end-consumer role is played by the computer with demand streams modeled using stationary demand distributions. However, the game is flexible to accommodate other types of distributions or structures for modeling end consumer demand.

The goal for both the supplier and the customers is to maximize their own profits. Players can be offered some rewards (increasing their profits) to encourage participation and to motivate the players to seriously play the game. Profits for both the supplier and the customers are calculated as follows:

Profit = Revenue - Procurement Cost - Inventory Cost - Backlog Cost

Revenue comes from shipping products to one’s customers; procurement cost is proportional to the units received from one’s supplier. If there is positive inventory left at the end of a period, an inventory holding cost is charged for each unit of inventory. If demand exceeds one’s finished goods inventory, the unmet demand is backlogged and each unit of backlog is charged a backlog cost. The supplier’s revenue is equal to the sum of the procurement costs of all customers in each period. Unit selling price, procurement cost, inventory cost and backlog cost are private information to each player (with the exception of the wholesale price charged by the supplier to the customers, which is known to both parties).

The customers also know the probability distribution of the consumer demand in each period. Each customer has an independent stream of demand, but under some game settings, backlogged demand could move to other customers or leave the system if held for too long, causing the customers to compete for consumer demand. The game could also be configured to allow for different types of information sharing, e.g., where the supplier can observe each customer’s actual demand or even inventory levels in real time. Table I summarizes the information available to the players, the tasks the players need to perform and the goal of the game.

Table I: Summary of the players’ roles in the game.

Table I: Summary of the players’ roles in the game.

Flow of the Game

In each period, the following events take place:

  1. Each customer submits an order to the supplier asking for a certain number of units to be delivered in that period. Customers also submit forecasts for the next few periods.
  2. The supplier receives the orders and the forecasts from the customers, and specifies the number of units to be shipped to each customer, as well as the production quantity for the current period. If the supplier does not have enough inventory to fulfill every customer’s order, some orders will be backlogged and fulfilled in future periods. The backlogged orders are known as the “on-order” quantity to the customers.
  3. Customers receive the shipments from the supplier immediately, and their inventories are replenished.
  4. Demand for each customer occurs and is filled immediately using each customer’s inventory. If the demand exceeds available inventory, the customer backlogs the demand and fulfills it in future periods.
  5. Profit is calculated for the current period, and the period ends. The events in each period are summarized in Figure 2.
Figure 2: Sequence of events in each period of the game.

Figure 2: Sequence of events in each period of the game.

Observations and Insights from the Game

During the years 2009 and 2010, more than 100 human subjects participated in the interactive supply chain game. Most of the subjects were undergraduate and graduate students from ISyE at Georgia Tech. Several employees from Intel also participated in the study. In all experiments, the supplier was automated and two human subjects were asked to play the role of the customer. The focus in these initial experiments was on understanding the impact of: (i) the supplier’s service level, (ii) the inventory allocation rule, and (iii) the information given to the subjects regarding the allocation rule, on the customers’ forecasting and ordering behavior.

To answer the first question we designed four sets of experiments (labeled V1.0 below), with high (85 percent ~ 95 percent), medium (70 percent ~ 80 percent), and low (55 percent ~ 65 percent) service levels (fill rate). We also have a setting with medium service level where the customers do not know the exact demand distribution. The supplier’s production decision is determined by a base-stock policy (i.e., the supplier’s production quantity is equal to the sum of orders in each period), and the supplier uses the uniform allocation rule when there is not enough inventory to satisfy all the demand. The results are summarized in Table II.

Table II: Summary of results from the first set of experiments.

Table II: Summary of results from the first set of experiments.

In all four settings, subjects demonstrated a clear learning effect in terms of higher profit per period, lower fluctuation of inventory levels and higher consistency of order-up-to levels in the second compared to the first half of the game. Therefore, all analyses were based on the second half of the game. As expected, customers’ profits go down as the supplier’s service level goes down. The highest amount of forecast inflation is observed in the medium service level setting. While this seems a little counter-intuitive as first, it can be potentially explained as follows:

• When the service level is high or low, the supplier either consistently delivers orders without any delay or consistently carries a backlog. In either case, the supplier’s inability to fulfill orders becomes more apparent to the customers. However, when the supplier’s service level is medium, his ability to fulfill orders is not consistent. This may have puzzled the customers, leading them to inflate forecasts more in the hope of securing supply in the future. When customers do not know the demand distribution, both forecast inflation and mean absolute forecast effort is higher, which is intuitive.

These experiments provide significant evidence that human subjects choose to inflate forecasts even when there is no benefit in doing so. Recall that the supplier produces according to a base-stock policy and allocates inventory using a uniform allocation rule. Hence, the forecasts in these experiments do not have any impact on the supplier’s production or allocation policy. However, not knowing the supplier’s decision mechanism, the average human customer player seems to think that inflating forecasts (at least a little) will be beneficial.

To address the second question, we designed a forecast-accuracy-based inventory allocation rule and introduced it to the medium and low service level settings of the game, labeled as V1.1. Here the supplier still uses a base-stock production policy, but in case of shortage, allocates (proportionally) more products to the customer who has better forecast accuracy. The results of V1.1 compared to V1.0 are summarized in Table III.

The results show that the magnitude and the frequency of forecast inflation are significantly lower when the new allocation rule is announced and implemented. Recall that the supplier still uses a base-stock production policy and there is no improvement in the supplier’s service level overall. Hence, if both customers equally improve their forecast accuracy, neither of them will receive a higher service level from the supplier. Only when the two customers’ forecast accuracy differs significantly will the better forecaster be rewarded a higher service level. Despite this, forecast inflation is much lower when the new allocation rule is implemented. This leads us to ask the third research question: Is the reduction in forecast inflation due to the new inventory allocation rule or due to subject’s perception that bad forecasts will be punished?

We designed two other experiment settings to answer the third question. In V1.0/1.1, the subjects were given the same instructions as in V1.0, but the actual allocation rule being used is one that rewards forecast accuracy as in V1.1. In V1.1/1.0, the subjects were given the same instructions as in V1.1, but the actual allocation rule used was the uniform allocation as in V1.0. By comparing the results of these four experiment settings, we aim to understand whether the reduction in forecast inflation is truly due to the new allocation rule or the “belief” of the human subjects regarding the rule.

As Table IV shows, both forecast inflation and the mean absolute forecast error are lower in V1.1/1.0 than in V1.0/1.1, meaning that telling the subjects that the supplier will reward accurate forecasts without actually implementing it is more effective in reducing forecast inflation than implementing the allocation rule without telling the customers about it. This provides support that the psychological effects on the subjects may outweigh the real effects generated by the allocation rule itself, confirming our intuition based on the results of the second set of experiments.

Conclusions

The results of the third set of experiments also prompted us to ask, “What if the supplier improves his overall service level when the overall forecast accuracy of his customers improves?” In other words, if we let the supplier’s production decision also depend on customers’ forecast accuracy, each customer may see a tangible improvement of supplier’s service level when he improves his own forecast accuracy regardless of how the other customers are forecasting. This may change the forecasting behaviors of the customer yet again – a research topic now being studied.

The supply chain game has been piloted at Intel with several business managers from various areas of the supply chain. The consensus feedback from experiments has been quite positive, as practitioners can assume specific supply chain roles and experience in “real time” the impact of their decisions on supply chain performance metrics. This attests to the flexibility of the game in modeling various supply chain scenarios.

The debrief session held following the gaming simulation also provides a knowledge-sharing forum to discuss analytical and behavioral strategies used during the game. From a business perspective, practitioners compare their strategies against other competing strategies and discuss their impacts on costs and customer service. The supply chain game thus can be applied in coaching strong supply chain strategic thinking skills and validating the implications of various supply chain strategies.

To further derive value from the supply chain gaming research, future plans include adapting the game in various Intel supply chain business settings. Areas being considered include assessing various raw material procurement strategies and customer relationship management policies, with the goal of developing win-win relationships. Additionally, Intel expects to utilize the tool in piloting business strategies prior to implementation.

Mani Janakiram is a director of supply chain strategy at Intel where he has managed several projects in supply chain, strategy roadmap, modeling, capacity planning, process control, analytics and research. He has more than 20 years of industry experience, including stints at Honeywell and Motorola, and has published more than 50 papers.

Pinar Keskinocak is the Joseph C. Mello Professor in the Stewart School of Industrial and Systems Engineering and the co-founder and co-director of the Center for Health and Humanitarian Logistics at Georgia Institute of Technology. She also serves as the associate director for research at the Health Systems Institute at Georgia Tech.

Tosanwunmi Maku is a senior operations research engineer at Intel Corporation. His experience spans management consulting and supply chain analytics, with interests in network design, strategy evaluation and systems simulation. He has Ph.D. in industrial engineering from Texas Tech University.

Shuangjun Xia is a Ph.D. student in the Stewart School of Industrial and Systems Engineering at Georgia Institute of Technology.