In our experience, students learn best from the beer game when the instructor gives little or no advice about how to choose order quantities. Instead, we recommend that the instructor initially provide guidance only on 1) the current state of the supply chain and 2) how to set up the game and play each period.

1) Current State of the Supply Chain:

2) Setup and Play Instructions:

In order to produce results that best highlight the key lessons of the beer game, including the use of AI, we recommend guiding the students through a few rounds of play with specific settings. Below, you will find options to guide students through the classic beer game and its associated learnings as well as rounds focused mainly on the performance of the AI player.

1) One Human Player + Three Human-Like Players (Classic Setup): The student is the human player and the three computerized players play as “human-like” players. This is the closest setting to the classic four-player beer game, with the “human-like” players standing in for the student’s human teammates. The student should play the role of Wholesaler and use the “classic” demand pattern. Note: This is the default setting. No settings changes are required prior to game play.

2) One Human Player following a Rational Strategy + Three Human-Like Players: This time we want to guide the student to make ordering decisions following a base-stock policy (similar to if a Rational computerized player were to take control), to demonstrate the difference between typical human play and a more stable inventory policy. (All other settings remain as in Round 1.)

It’s easy to play using a base-stock policy: In period 1, the player should order 8 units. In every period after that, the player should order the same quantity as the demand he or she received.

This will implement a base-stock policy with a base-stock level of 32. (In period 1, the player has a starting inventory level (IL) of 12, plus 16 units on order, and receives a demand for 4 units. The inventory position (IP) is therefore 12 + 16 − 4 = 24. An order quantity of 8 brings the IP to 32. In subsequent periods, setting the order quantity to the demand keeps the IP at 32.) Note that 32 is not necessarily the optimal base-stock level, since the other players are not following a predictable ordering pattern, but it is reasonable under these settings.

3)There are two options for how you can run the final round:

a) One Human Player + Three Rational Players: The student plays as the Wholesaler, and the three computerized players are set for Rational play. All other settings are as in round 1.

b) Four Rational Players (Observation Round) – All four players are computerized and set for Rational play. The student simply watches the computerized players play.

By default, the game lasts for 20 periods. Some instructors prefer to run the game for 30 or more periods (you can have the students easily change the number of rounds within the ‘Change Settings’ option of the Default Settings Screen) so that the cost and bullwhip patterns become more evident. In either case, you should use the same number of periods in every round so the results are comparable with each other.

After each round you should encourage the students to review the resulting order quantities, costs and bullwhip effect index (BEI) values over time. (Inventory levels and fill rates can also be included to round out the overall metric review if desired.) You will guide them through a comparison of these performance metrics among the rounds.

Note: The students should write down and/or export the results of each game they play, for easy comparison across rounds. Alternately, they can open a new browser tab for each round they play.

After students have completed all three rounds of Classic Game Play, it is good to let them lead the conversation about questions such as, What did they think of their results? Was it what they expected to happen? Why or why not? If the supply chain performed poorly, was it the “fault” of the student or one of the computerized players?

After encouraging the students to interpret their own results, it’s a good time to zero in on the performance fluctuations of each player in the different rounds and introduce the concept of the bullwhip effect. You can find more documentation on this in What is the Bullwhip Effect? . Also, there are further discussion points for each round in the Sample Output section below.

One of our favorite features of the Opex Analytics Beer Game is the post-game comparison, which shows you how each of the computerized players would have performed, compared to how you performed, if they played in your role with the same settings. This gives an opportunity to discuss the AI player’s performance. Did the human player beat the AI, or vice-versa?

Below we provide further insights to be discovered within each round.

1) One Human Player + Three Human-Like Players

Observation: Order volatility increases as you move upstream in the supply chain (Demand -> Retailer -> Wholesaler -> Distributor -> Manufacturer). This is the bullwhip effect! There is also a lag of a few periods between orders of one player and the (increasingly volatile) orders of the next, due to the order and shipment lead times. The “human-like” players follow the formula proposed by Sterman (1989), which is meant to emulate the way human players play the beer game. The order quantity increases when the inventory level (IL) or inventory position (IL + on-order items) fall below a target value. In other words, the player exhibits “panicky” behavior, over-ordering when inventories get low, even if the correct amount of inventory is already in the pipeline. Conversely, when inventories are high, the player under-orders, getting complacent even when there is not enough inventory in the pipeline.

Observation: All four players exacerbated the bullwhip effect, as evidenced by the fact that all four bullwhip effect index (BEI) scores are greater than 1. The computerized players showed some panic in their ordering over time; the human player showed even more. You can read more about the BEI measure in the Game Results section

Note: The BEI is a “moving variance” and thus will change throughout the time horizon of the game. When discussing with students, simply concentrate on the value in the last period.

Now focus on the comparison graphs displayed at the end of the game. While the student is playing the game, the software plays the game several times behind the scenes, replacing the human player with each of our computerized players, using the same game settings.

The software displays two graphs. The first compares the total supply chain cost (all 4 players), when the human is replaced by each of the computerized players. The second another displays the BEI for the role the human is playing (e.g., Wholesaler) when the human is replaced by each of the computerized players.

Observation: Under the Classic demand pattern with Human-Like teammates, our AI agent is hard to beat. It achieves a cost of $281, less than half the cost obtained by a Rational player playing in the same role, $649.50. (The Rational player uses a base-stock policy with reasonable, though not necessarily optimal, base-stock levels.) Humans tend to perform pretty badly in this setup — here, the human had a cost of $1,027. Even the Random player beat our human, which is not uncommon, and can be a source of some good-natured humor in class. In this example, the Human-Like player performed the worst, though not significantly worse than the human.

Observation: The Rational player has the lowest BEI, since it uses a base-stock policy. (Remember that a base-stock policy, by definition, produces BEI values close to 1.) The BEI for the AI player is quite a bit higher, even though its costs are lower, reinforcing the point that higher BEI does not always mean higher cost. The Random and Human-Like players had rather large BEI values, as did our human player.

The Cumulative Cost rises continuously as time goes on (over $1,000 total cost), again showcasing the effect of the continuous fluctuation in complacency or panic of the players causing exorbitant carrying or stock out costs.

2) One Human Player following a Rational Strategy + Three Human-Like Players

Observation: In this case the student (playing as the Wholesaler) follows a base-stock policy. This shows in our graphic as the Wholesaler orders are exactly the same as the Retailer’s, two periods later. Notice that the Wholesaler neither increases nor decreases the bullwhip effect–s/he simply passes along the orders s/he receives.

Observation: Notice that in this case the student (as Wholesaler) performs with a bullwhip effect index (BEI) of close to 1. (It doesn’t equal exactly 1 because of the difference in order quantities in the first period, and the lag between the Retailer’s and the Wholesaler’s orders). This confirms that the Wholesaler neither increases nor decreases the bullwhip effect, as we saw in the order quantity graph. On the other hand, the computerized, Human-Like players contribute to the bullwhip effect, similar to round 1.

Other sample output

The total cost is lower than in round 1 (roughly $650 vs. $1,000), also showing the improvement from the Wholesaler following a base-stock policy. Fill rates are better, too.

3) There are two options for how you can run the final “Classic Game Play” round.

3.a) One Human Player + Three Rational Players

Observation: Because it follows a base-stock policy, the Retailer reproduces the Demand pattern in its own orders. (The Demand is obscured by the other curves, but you can see it by clicking on some of the other curves in the legend to turn them off.) The Wholesaler (the human player) breaks the pattern, but then the Distributor and Manufacturer reproduce the Wholesaler’s ordering patterns.

Observation: As in Round 1, the computerized Rational players have BEIs close to 1, and the human player has a much higher BEI.

Other sample output

The total cost is again lower than in previous rounds ($450 here vs. $650 in Round 2 and $1,000 in Round 1). This shows the additional improvement from more Rational players being added to the team.

3.b) Four Rational Players

Observation: Now each player passes along the orders that it receives, except in the first period when they order up to the base-stock level. There is no bullwhip effect.

Observation: All four players have BEIs close to 1. The Retailer, Wholesaler, and Distributor have BEIs below 1; this is an artifact of the fact that the first-period orders differ from the demands.

Other sample output

This round also represents the lowest cumulative total cost, at $180. Because the base-stock levels are sufficiently high and the demand is almost stable, the fill rate is 1 for every player.

Opex Analytics has a team of educators that are available for hire to plan and deliver customized trainings in many areas. With a deep background in academia, Opex spends lots of time and resources on researching, developing and delivering analytics educational materials. We would be happy to visit your classroom or company to lead your group through the beer game and even expand the session with lessons from other supply chain analytics concepts as well.

Samples of other Opex Academic Materials

Supply Chain Network Design Book - http://networkdesignbook.com/

Fundamentals of Supply Chain Theory - https://coral.ise.lehigh.edu/sctheory/

Slick Oil Distribution Game - http://www.scdigest.com/experts/DrWatson_17-10-10.php?cid=13119