This phenomenon of "upstream amplification" occurs in many chains. It happens in food chains, in the housing industry and is notorious in supply chains, where it has been labeled the bullwhip effect. Every change downstream in the chain results in a relatively greater change one step upstream in the chain. The amplification ratio is greater than 1.0. So, if mobile phone sales increase 10%, orders to the semiconductor industry for mobile phone chips go up 30%, and orders for new wafer steppers and other equipment used to produce IC easily double from one year to the next. This phenomenon was already observed and masterfully analyzed by Prof. Jay Forrester in the late 1950s and described in his Industrial Dynamics , still a must-read for all of us systems thinkers.

Delays between perceiving that a change has occurred and reacting to this change form one part of the explanation for the bullwhip effect. But, there are more. Several of the explanations focus on locally optimizing behavior of actors in the chain that result in sub-optimal performance of the chain as a whole. One important type of behavior is shortage gaming . When, in 1999, mobile phone producers noticed that the semiconductor manufacturers of this world could not keep up with their demand, they inflated their orders for ICs. If I know that you can make 100 and my competitor and I both need 60, then we normally will both get 50 if the supplier allocates his stock fairly. But, what if I cheat and ask for 90? Then the supplier will see a total demand of 150, and if he then tries to allocate fairly, he is over extended by half of his capacity. So, he will only ship 2 units for every demand of 3, and so I will receive the 60 I need, while my competitor will only get 40. Great idea, if only my competitor hadn't thought of the same idea. So, total demand is inflated and everybody is worse off.

Behavioral routines such as these are the "usual suspects" of the bullwhip effect and you will find them occasionally in retrospective accounts of a boom and bust. Unfortunately, these routines are typically forgotten as soon as the next boom takes off. Much less obvious causes of demand amplification in chains are our inventory control systems. Wasn't inventory intended to buffer against fluctuations in the environment and hence smooth what happens behind it? Yes, but then you need the right algorithms to do so. A widely used inventory control policy is the order-up-to rule: try and keep the stock level at X, where X is so many weeks of average demand. So, every period, you try to produce to stock as much as you need to bring stock levels back to X, taking into account the expected demand in the coming period. Sounds reasonable enough, doesn't it? Yes, but let's now assume that it takes you 3 weeks to produce to stock. Now, demand in one week goes up with 10%. What many order-up-to policies will assume, is that demand will remain at this higher level, or at least for the coming three weeks. This means that you will start producing not 10%, but 3 times 10%. Hence, you will order 30% more materials, and amplification happens again. This is one example of the unintended destabilizing effects of inadequate inventory control policies. MRP systems typically work this way, and MRP lies at the basis of ERP. And ERP, didn't we just spend several billions in various industries on making that the corner stone of our goods flow control systems? Yes, we did. And since so many lemmings can't be wrong, few people are questioning if the algorithms in these systems are really what we need to make our supply chains perform better.² ERP may be great to get rid of patchworks of legacy systems, but the MRP algorithms embedded in it as production control system suck. With the advent of Advanced Planning Systems (APS) on top of these ERP systems, life does not get better, on the contrary, because the underlying model remains flawed.

So, what can be done to reduce demand amplification? One can obviously shorten response delays both in cycle times and in order fulfillment. But, even better, why don't we move from push to pull? MRP pushes goods through chains, but if you only pull through the chain what you need, you will not be making more than you need, provided your production cycle times are short enough.

Before we diversify into the many pitfalls of cycle time reduction and pull systems, let's return to our traffic congestion example one more time. This was not an example of just amplification, but of oscillation as well. Oscillation, a repeated fluctuation around some long-term average, is very much present in the visual metaphor of the bullwhip, and also very much a part of the everyday driving experience of the Dutch commuter. Road speeds keep going up and down, not just once, but all the time. This need not be because there is an infinite number of truck drivers that keep moving to the right lane one after another. In many dynamic systems, it takes only a one-time disturbance to create a continuous, or at least long-lasting, oscillatory movement, like a pendulum that keeps swinging for a long time after you have pushed it once.

The important thing to note here is that this happens because many supply chains are inherently unstable . Mathematically speaking, they belong to a class of systems that, for specific parameter settings, have no stable equilibrium value. Key parameter values here are the lengths of the various delays in the system. There are no simple answers as to what the "right" lengths will be, or how they should relate to each other. What you can do is create a System Dynamics model of the structure of the system and then conduct a sensitivity analysis: for what parameter values will this system start to oscillate after a one-time disturbance and when will it remain stable? Just try and you will be amazed how tiny differences in values turn an inherently unstable system into a robust one, and how other sets of parameter values will make fluctuations go totally out of control.