Mitigating the Bullwhip effect in supply networks
The Bullwhip effect (BWE) is among the most central topics in Operations management. In a recent paper, we offer a new network perspective on the BWE (Osadchiy et al. 2020). In light of these findings, I thought, it might be helpful to distill what we know about Bullwhip and what is the new knowledge that comes from viewing the BWE as a supply network phenomenon.
Traditionally, the BWE has been studied in serial supply chains. That is also the setting of the famous beer game, where the typical demand pattern across the supply chain looks like this: for a relatively small change in demand seen by the retailer, the demand variability progressively increases at the more upstream wholesaler, distributor, and factory (see the figure below). This leads to capacity shortages, stockouts, excess inventory, and increased operational costs. No wonder BWE has received a lot of attention among academics and practitioners alike.
In their seminal paper, Lee et al. (1997) define the BWE as “the phenomenon where orders to suppliers tend to have larger variance than sales to the buyer (i.e demand distortion) and the distortion propagates upstream in an amplified form (i.e variance amplification).” From here, one can make two observations:
BWE comes in two flavors: demand distortion and variance amplification
In serial supply chains, demand distortion implies variance amplification and vice versa
The first point delineates between demand distortion that is characteristic of internal firm processes (for example, batch sizes, lead times) and variance amplification that occurs when orders from a customer firm reach the supplier firm.
The implication of the second point that demand distortion translates to variance amplification, however, would not be true if we recognise that the majority of supply chains are not serial, and in fact, are supply networks or webs. For a simple example, think about a supplier with two customers, both of which show demand distortion, but whose orders counter-balance each other, in the sense that one orders when the other does not. In that case, the supplier will see a pretty steady demand and its variability may be even less than demand variability seen by either of the customers. But, this, of course, is just a thought experiment. What is actually going on?
For that, we teamed up with Bill Schmidt from Cornell and Jing Wu from the Chinese University of Hong Kong and turned to data. First, we had to build a supply network and determine the positions of firms in the network: which firms are more downstream and which are more upstream? In a network that looks like the one on the left, it is not easy.
We used the distance to final consumers as a measure of upstreamness, we called it Layer. Firms that sell consumer goods would be the most downstream layer, their immediate suppliers would be one layer above, suppliers to suppliers – two layers above, and so on. It turned out that the supply network is quite deep, reaching up to 9 layers of depth. With that, the network started to look like the one on the right (only first four layers are highlighted).
The next step was to examine the demand distortion within each firm and measure demand variability across layers of the network to check if one implies the other. The results: demand distortion is overwhelmingly common, on average, variability of firms’ orders is 19% greater than variability of demand. With that degree of demand distortion, the demand variability would roughly double from Layer 0 (most downstream) to Layer 4 – mid depth of the network. Yet, demand variability was pretty steady across layers, at around 21%. That looked puzzling to us.
To resolve the puzzle, we looked at how suppliers form relationships with customers and found that adding a customer on average leads to a demand variability reduction. Terminating a relationship with a customer and swapping one customer for another also lead to demand variability reduction. Furthermore, there is evidence that suppliers select their customers in a way that reduces the overall demand variability. For example, suppliers especially like customers that place orders during a slow season when there is excess capacity.
What can managers take away from this study?
First, suppliers can use the network of customers to shield from the adverse effects of demand distortion exhibited by downstream customers. A right mix of customers can help smooth out spikes and dips in their order patterns. That, in turn, helps maintain steady utilisation of resources, and give employees predictable schedules.
Second, a customer that distorts demand is not necessarily a bad one. Quite the opposite, that customer can be very valuable as long as they place orders not at the same time with the other ones. Finally, the benefits from careful customer base selection are complementary to factors that have been shown to reduce demand distortion, such as information sharing, smaller batch sizes and lead time reduction.
To sum up, adopting the network perspective does have some implications for how we teach and think about Bullwhip: it is important to delineate between the demand distortion and variance amplification components of BWE. Both are important and can be managed in a complementary way. Managing demand distortion involves intra-firm processes (Hau Lee et al. paper and many other great papers that followed). Managing variance amplification involves optimising your customer network and that is the focus of our work in this paper.
References
Lee, H.L., Padmanabhan, V. and Whang, S., 1997. Information distortion in a supply chain: The bullwhip effect. Management science, 43(4), pp.546-558.
Osadchiy, N., Schmidt, W. and Wu, J., 2020. The bullwhip effect in supply networks. Management Science, forthcoming. Draft available at SSRN 3241132, https://dx.doi.org/10.2139/ssrn.3241132