At the last Agile Dinner in Helsinki on Tuesday 6th of July I ran The Beergame. The turnout was surprisingly good for being prime vacation time as we had a total of 16 people.
The game actually requires a lot of preparation and setup. There is quite a bit of papers that need to be prepared and printed. Good version can be found on the Beergame site but for this I created my own slightly improved ones. The group gave some good feedback and I have some good improvements for the next time such as using more of a pull process to run the supply chain and color coding the papers.
Here we see how the different team ordered from their suppliers. It clearly shows the panic reaction in some of them as they begin to order ever bigger amounts. Each of them ordering more than the previous one. The first spikes are a really big Bullwhip effect but the second is a bit smaller.
Causes of the Bullwhip Effect
1) Lack of information
In the beergame no information except for the order amount is perpetuated up the supply chain. Hence, most information about customer demand is quickly lost upstream in the supply chain.
With these characteristics the beergame simulates supply chains with low levels of trust, where only little information is being shared between the parties.
Without actual customer demand data, all forecasting has to rely solely on the incoming orders at each supply chain stage. In reality, in such a situation traditional forecasting methods and stock keeping strategies contribute to creating the bullwhip effect.
2) Supply chain structure
The supply chain structure itself contributes to the bullwhip effect. The longer the lead time, i.e. the longer it takes for an order to travel upstream and the subsequent delivery to travel downstream, the more aggravated the bullwhip effect is likely to be.
With traditional ordering, the point in time where an order is typically placed (the order point) is usually calculated by multiplying the forecasted demand with the lead time plus the safety stock amount, so that an order is placed so far in advance as to ensure service level during the time until the delivery is expected to arrive.
Hence, the longer the lead time is, the more pronounced an order will be as an reaction to an increase in forecasted demand (especially in conjunction with updating the safety stock levels, see above), which again contributes to the bullwhip effect.
3) Local optimisation
Local optimisation, in terms of local forecasting and individual cost optimisation, and a lack of cooperation are at the heart of the bullwhip problem.
A good example for local optimisation is the batch order phenomenon. In practice, ordering entails fix cost, e.g. ordering in full truck loads is cheaper then ordering smaller amounts. Furthermore, many suppliers offer volume discounts when ordering larger amounts.
Hence, there is a certain incentive for individual players to hold back orders and only place aggregate orders. This behaviour however aggravates the problem of demand forecasting, because very little information about actual demand is transported in such batch orders.
And batch ordering, of course, contributes directly to the bullwhip effect by unnecessarily inflating the orders.
The inventory of each stage in the supply chain also shows a similar behavior. It comes after the Bullwhip effect and is reversed. This is first the factory inventory grows rapidly and then it propagates down the supply chain. This of course causes everyone to stop ordering from their supplier. Once the inventory levels drop they order again and most likely cause the next Bullwhip effect.