Pineapple BrewBeer Game Tutorial

Read this first, then beat the chain

There is nothing to win against the other players in the Beer Game. The only opponent is the cost your chain runs up when nobody understands what the delays are doing. First-time teams typically end up around ten times above what a well-run chain would pay. Much of that gap comes from one habit, and section 6 gives you a simple rule for it.

About 8 minutes4 interactive experimentsQuiz at the end

About the spoiler boxes on this page

Some boxes reveal things that change how you play, for example the demand pattern of the classic game. Reading them is not cheating, and it helps you understand the mechanics; it only means the game will feel different for you than for someone who has not read them. If you want the textbook experience first, leave those boxes closed until after your first game. If you open them, tell your host, who can choose a different demand pattern for the group.

1. The game in 60 seconds

Four stations move beer in one direction and orders in the other. Customers buy from the retailer, the retailer orders from the wholesaler, the wholesaler from the distributor, the distributor from the brewery, and the brewery brews. This app can add a malt house and a farmer, which makes the chain longer and everything upstream worse.

  1. Brewerybrews, unlimited capacity
  2. Distributorsees the wholesaler's orders
  3. Wholesalersees the retailer's orders
  4. Retailersees real customer demand
  5. Customersbuy, or wait
  • One decision per round: how many cases you order from your supplier.
  • You see your own customer only. Only the retailer knows what the market buys. Everyone else sees orders, which are somebody else's decision.
  • Everything takes time. Depending on the rule set, an order arrives on the third day or four weeks later. Your correction always lands late.
  • A backlog is a debt, in both directions. What you cannot ship now, you ship as soon as stock arrives, whether your customer still wants it or not. What your supplier owes you, he delivers later, whether your warehouse is full or not. Nobody can cancel or refuse.
  • Both mistakes cost money. Stock costs per case and round; a missing case costs twice as much until you deliver it.
  • The score is the chain, and nobody talks. In team scoring the total of all stations counts, and the no-talking rule exists to produce the failure described below.
Experiment 1

Anatomy of one round

The retailer starts the day with 8 cases in stock, 4 arriving today, and two shipments of 4 on the way. Step through what happens, in the order it happens.

Keep in mind: when you decide, the delivery for this round has already arrived. What you order now changes the situation several rounds from now.
Why the number of rounds is usually hidden

If everyone knows the game ends in round 35, the clever move in round 32 is to stop ordering and run the stock down. The original setup told players nothing about the length, and this app hides it by default. The host can switch that off; the last rounds then measure end-game tactics rather than supply-chain skill.

2. Why this game exists

The game came out of Jay Forrester's System Dynamics Group at MIT in the 1960s. Its point is that the wild swings companies blame on the market are often produced inside the chain, by delays and feedback in the ordering process itself. John Sterman turned it into a measured experiment: across many runs, average costs of first-time players were roughly ten times higher than the achievable optimum, with the same pattern every time, oscillation, amplification toward the brewery, and a lag between market and factory.

The effect is real outside the classroom. Procter & Gamble looked at Pampers: babies consume nappies at a steady rate, retail sales varied only mildly, distributor orders varied more, and P&G's own orders to suppliers such as 3M swung the most. P&G's logistics managers named the pattern the bullwhip effect, as Lee, Padmanabhan and Whang report.

3. What goes wrong: the bullwhip effect

The swings in orders grow at every step upstream. A modest change in customer demand becomes orders several times that size at the brewery, followed by weeks of zero orders while everyone sits on the stock they over-ordered. The bill is paid four times: financed stock, lost sales in the backlog, overtime and idle time in production, and capacity built for a peak that was never real demand.

Spoiler: the demand pattern of the classic game, used in the experiments below

Opening this changes your first game. The experiments on this page use this pattern, and their charts show it.

In the classic game, and in both of this app's rule sets, customer demand sits at 4 cases per round for four rounds, rises once to 8 in round 5, and stays there. Nothing else happens to it. The whole drama upstream is the chain's reaction to a single step.

Players in the original were told only that demand had been stable at 4. Once you know the step, you can stock up before it arrives; that is not cheating but using information the others lack, and it shows what information is worth in a supply chain. It also means your host should choose a random, seasonal or custom pattern for a group that has read this.

Rule set for the experiments:
Experiment 2

The amplification explorer

All stations follow the same rule: order the demand you expect, plus a share of the gap between the stock you want and the stock you have, counting some fraction of the goods already on order. Move the sliders. The presets change only the pipeline share.

Orders per station
Cost of the whole chain-
Largest brewery order-
Peak order / peak demand-

The number that matters most: the share of the pipeline you count. Sterman measured an average of 0.34: people mentally count about a third of the beer they have already ordered. Set the slider to 100% and the same chain, with the same delays and the same demand, calms down. It still swings, because the delays are still there, but the peak shrinks to a fraction.
How the bullwhip is measured, and why the numbers can look absurd

The textbook measure is a variance ratio: the variance of a station's orders divided by the variance of customer demand. Above 1 means the station amplified the swings. In the classic game demand changes exactly once, so its variance is tiny, and ratios of 100 or more appear even for moderate swings. The results screen therefore shows a plainer measure next to it: the station's largest order divided by the largest customer demand.

4. Why it happens

The structural layer

Lee, Padmanabhan and Whang identified four operational causes: demand signal processing, the rationing game, order batching, and price variations.

  • Demand signal processing. Each station reads the orders it receives as demand and adds its own safety margin. The margin of the margin grows upstream.
  • The rationing game. When supply is short and allocated in proportion to orders, everyone inflates orders, then cancels once supply recovers.
  • Order batching. Order cycles, full truck loads and minimum quantities turn a smooth stream of sales into lumps.
  • Price variations. Promotions make customers buy when it is cheap, not when they need it, so orders stop telling you anything about consumption.

Three of the four are switched off in the game: prices are fixed, order frequency is fixed, and nobody is rationed by a competitor. The bullwhip appears anyway.

The behavioural layer

What remains is the decision rule in your head. Sterman's explanation is misperception of feedback: people manage the stock they can see and discount the stock that is on its way. His shower analogy: the water is cold, you turn the tap, nothing happens, you turn it further, and a moment later you are scalded. The delay was not in your mental model. In the game it sounds like this: my backlog is growing, my last order still has not arrived, order more.

The trust layer

Croson, Donohue, Katok and Sterman ran the game with a demand that was constant and announced to everyone in advance. The chains still swung. Their explanation is coordination risk: even when you trust the market, you do not trust the other players to behave, so you keep a buffer against them, and your buffer becomes their surprise.

Why it survives experience: your inventory is visible and costs money every round. The pipeline is invisible and costs nothing yet. The cheapest habit in the game is to write the pipeline number down before you decide.

5. How supply chains fix this in the real world

  1. Share real demand instead of orders: point-of-sale data, shared forecasts, vendor-managed inventory, joint planning.
  2. Shorten the lead time. The bullwhip grows with lead time; try the third slider in Experiment 2.
  3. Order more often, in smaller quantities, which needs cheaper setups and mixed loads.
  4. Keep prices stable, so that orders reflect consumption rather than promotions.
  5. Allocate by past sales, not by current orders, so that inflating orders stops paying.
  6. Fix the ordering rule itself: subtract what is already on order and adjust gradually. That is section 6, and it is the part you control alone.
Experiment 3

Would talking to each other help?

Four runs of the same chain, same demand, same delays. The only difference is what each station knows and how it decides.

A is the typical human chain measured by Sterman. B adds shared demand information. C keeps everyone in the dark but fixes the ordering rule. D does both.

The limits

  • Information helps, but does not cure. In experiments where inventory information was shared across the chain, the amplification was reduced but not eliminated, and players kept underweighting the supply line. Wu and Katok found that training alone did not help significantly either; only training combined with communication did.
  • The delays stay. Brewing, transport and production ramp-ups are physical. The best you can do is plan with them instead of against them.
  • Interests differ. Firms in a chain have separate books, so sharing data touches margins and bargaining positions. The practical instruments are contracts and joint planning, not goodwill.

Inside the game, talking is banned. What you can do is stop sending noise: your order is the demand signal your supplier forecasts from.

Why "just pass the order on" is so hard to beat

On the results screen you will see a run in which every station simply orders what it was asked for. It sometimes beats the rule of thumb, and sometimes the human team. In a chain that starts in balance, not reacting at all sends every change upstream once, without amplification, and pays only the temporary backlog while the pipeline catches up. Any rule that holds a safety buffer pays for that buffer every round, and with a demand that only wobbles around a stable level, the buffer never earns its keep. Pass-through fails when demand shifts for good or when the chain starts out of balance; it then never recovers what it lost, and in the simulations it costs up to almost twice as much as the rule of thumb.

Why care about the whole chain and not only your own cost

Inside the game, your supplier's backlog is your missing delivery a couple of rounds later. Outside the game, every supplier eventually prices his inventory and overtime into what he charges, and the chain passes it on to the end customer: expensive chains lose to cheap chains, not one link to another. And every station optimising only itself is one of the causes on Lee's list. That is why team scoring is the default here.

In real life I could just cancel the order

You could, at a price, and the beer would not disappear; it would sit at your supplier. That is the second act of the bullwhip: when demand softens, everyone cancels at once and the top of the chain is left with the stock. In 2001 Cisco wrote off roughly 2.2 billion dollars of components after such a cancellation wave. The no-cancel rule in the game keeps the cost where you can see it.

6. A rule of thumb that works

One line, four numbers, about 20 seconds per round. It keeps a chain stable in every demand pattern we simulated and beats typical human play by a wide margin. It is not the cheapest rule in every situation; the second box below says where it loses and what to change.

Order = incoming order + 30% × (inventory gap + pipeline gap)

Incoming order (E)

The order you received this round. It is your best guess of what your customer wants next, so it is the base of your order. No rounding up because you are nervous.

Inventory gap

Target inventory minus net inventory, where net inventory is stock minus backlog. The target is the buffer you want, by default the stock you started with. With a backlog of 6 and no stock, net inventory is −6 and the gap against a target of 12 is 18.

Pipeline gap

Target pipeline minus what you have ordered but not yet received. The target pipeline is E times the lead time minus one, because this round's delivery has already landed. With E of 5 and a lead time of 4, you want about 15 on the way; if 22 are on the way, the gap is −7 and you order less than E. This is the term everybody drops. Dropping it is the same as setting the pipeline share to zero in Experiment 2.

Why 30% and not 100%

Because of the chain. When your incoming order rises by 4, your target pipeline rises by 4 times the lead time minus one, so the full gap is suddenly 12 or more cases. Order all of it at once and your supplier sees a jump of 16 where the market moved by 4; he does the same, and the brewery sees a mountain. In the simulations, closing the full gap every round costs a chain of four ten to twenty times as much as closing a third of it; the second slider in Experiment 2 shows it. Taking 30 percent passes the change upstream in small steps and forgives errors in your forecast and your pipeline count. The price is that a backlog takes a few rounds longer to clear. Between 20 and 40 percent is a sensible range.

How high should the target inventory be? The cost ratio decides

The buffer covers rounds in which demand exceeds your expectation, and how much of it is worth paying for depends on the two cost rates. The rule from inventory theory: aim for a service level of backlog cost divided by the sum of both costs. With the game's usual 1 to 2 ratio that is 67 percent, a modest buffer. With equal costs a buffer is worth nothing; you aim for zero net inventory. If a backlog also cost you the customer or a penalty, say 1 to 10, you would aim for 91 percent and hold much more. With a demand that never surprises you, no buffer earns its keep at all, which is why the near-optimal reference on the results screen usually keeps a small one at most.

Where this rule loses, and what to change

When demand only wobbles around a stable level, reacting to the last order turns every wobble into an order swing. Taking E as the average of the last three incoming orders cuts the chain's cost by about a third in that case, and simply passing orders on is cheaper still. When demand shifts for good, as in the classic step, the last order is the better E and the rule beats pass-through clearly. So: random or seasonal pattern, average the last three; step-like pattern, use the last order. Everything else stays the same.

Experiment 4

Run the rule of thumb on your own numbers

Two checks before you press Order: Did I subtract what is already on order? If my order is more than twice what came in, why?

7. Playbook for the next session

  1. Say the pipeline out loud. "22 on the way, I want 15" is a complete decision. The cockpit shows it as "On order".
  2. Never chase a backlog with one big order. It arrives in one lump, several rounds later, and you cannot refuse it.
  3. Expect the lag. After you correct, nothing changes for as many rounds as your lead time. Silence is not evidence that the order was too small.
  4. Upstream stations: discount the peak. If the orders reaching you have tripled while the chain is starved, you are looking at accumulated panic, not at the market.
  5. Unwind slowly. When the backlog clears, do not order zero for five rounds. That is the second half of the oscillation, and it costs as much as the first.
If you remember one thing: the inventory in front of you is history, the pipeline is the near future. Manage the pipeline.

8. Quiz

Six questions, with the reasoning after each answer.

Sources

  • Forrester, J. W. (1961): Industrial Dynamics. MIT Press. The production-distribution model behind the game, which was developed as a teaching exercise in Forrester's System Dynamics Group at MIT during the 1960s.
  • Sterman, J. D. (1989): Modeling managerial behavior: misperceptions of feedback in a dynamic decision making experiment. Management Science 35(3), 321–339. Source of the measured ordering rule and its average parameter estimates: θ = 0.36, S′ = 17, α = 0.26 and β = 0.34, the supply-line weight.
  • Lee, H. L., Padmanabhan, V., Whang, S. (1997): Information distortion in a supply chain: the bullwhip effect. Management Science 43(4), 546–558, and the companion article in Sloan Management Review 38(3), 93–102, with the Pampers example and the origin of the name.
  • Croson, R., Donohue, K. (2006): Behavioral causes of the bullwhip effect and the observed value of inventory information. Management Science 52(3), 323–336.
  • Croson, R., Donohue, K., Katok, E., Sterman, J. (2014): Order stability in supply chains: coordination risk and the role of coordination stock. Production and Operations Management 23(2), 176–196.
  • Wu, D. Y., Katok, E. (2006): Learning, communication and the bullwhip effect. Journal of Operations Management 24(6), 839–850.

The simulations on this page run in your browser with the same engine as the game itself. Numbers are deterministic, so you can re-run them and get the same result.