DDMRP: What's the fuss all about?

First of all I do not think #DDMRP has anything to do with whether or not we need a #forecast: We do, but I'll get to why later in the article. Arguing about it is just noise.

The real breakthrough of DDMRP is that for the first time we have incorporated #stochasticity in #supplychain planning. We have needed this for some time. In fact, since the beginning of computer based planning in the mid-1990s.

We are all inherently aware of stochastic behavior, but we are conditioned as engineers to believe in a deterministic world. Everything we are taught is based on deterministic behaviour. As a consequence most of our solutions for supply chain planning do not incorporate stochasticity.

I never tire of asking people if they understand the basics of Little's formula. In just a few seconds I can understand whether they understand stochastic behaviour and whether I am going to have a meaningful discussion.

Few do.

And hence we get to the situation where the supply side makes statements like "the forecast is always wrong", and the demand side says "you never deliver to plan" to the supply side. Well, duh! This is because the world is stochastic, not deterministic. We refer to this as uncertainty, but it's not, it's variability. And variability is something that cannot be designed out of the system or process entirely. The silly thing is that we measure variability, especially demand variability, but we don't use it. Well, not beyond inventory optimization anyway. With the introduction of DDMRP we are now extending it beyond inventory optimization to influence haw we plan demand/supply matching.

I can't tell you how frustrated I get when I hear people state that the purpose of S&OP is the generation of a one-number plan for the organization. What responsible CEO/CFO would stand up before financial analysts and give a one-number prediction of future financial and operational performance? They don't, because they know shift happens. And yet in supply chain we measure the supply side on plan conformance and the demand side on forecast accuracy. Don't get me wrong, these are useful metrics, when combined with other metrics, particularly variability.

Some might say that they do measure variability of demand, and use that to determine forecastability of a particular segment or product. But what is it that they give to supply side of the house? A single number by time period. Similarly on the supply side, which has a single number for lead time, or yield, or throughput, or asset utilization, etc. And a single supply commit, with no measure of likelihood.

Enter DDMRP. For the first time we have an approach that starts from a stochastic perspective. The essence of DDMRP is that there is variability to both demand and supply so we need to place buffers to absorb this variability (size of buffers) at the points of greatest variability (placement of decoupling points). By doing so we allow the rest of the supply chain to operate in a more deterministic manner. Signal to noise. Plan to the signal, absorb the noise.

I'm sure the DDMRP experts will object to my characterization of DDMRP, and there are no doubt some aspects I have missed, but I am in fact saying this is a huge step forward. And, yes, we need a forecast beyond supply chain lead time to decide on capacity levels - staff and equipment, long term supplier contracts, etc.

Many of the more recent concepts, such as the Toyota Production System, are aimed at reducing the variability, and this is a worthwhile activity, but this applies only to the supply side. And none of these approaches will remove variability entirely, nor can they. They reduce it to an acceptable level. But at their core they are deterministic approaches that ignore stochasticity. Hence my interest in and excitement about DDMRP.

However, I am also going to say that DDMRP does not go far enough. It doesn't suggest how we should operate differently in order to better deal with the variability. And as we begin to move to the supply chain of 1 because of end consumer behaviors, personalized medicine, and other market factors, the variability is only going to increase. We think of this effect on the demand side, but the variability flows down into the supply side too because of smaller and smaller batches sizes, reduced inventories, etc.

This is where I find the work of Ian Glenday around Repetitive Flexible Supply interesting. Glenday's observation, across multiple industries, is that only 6% of items consume 50% of capacity, and that 30% of items only consume 1% of capacity.

Glenday's essential point is that the "greens" the signal and the "reds" are the noise. This means that we should set aside capacity for the "greens" and use Lean principles to plan these items.

The classical single echelon inventory optimization equation is below, which is made up of the average and standard deviations of demand and supply.

If we produce the same quantity of "green" items every cycle, we can reduce the amount of inventory required tremendously because this reduces the standard deviation of lead time to zero, eliminating the second term in the square root, meaning we only have to absorb the demand variability. Even if we cannot eliminate lead time variability entirely, reducing it will have a big impact on the amount of safety stock required. In addition, because we are producing the "green" items every cycle, we reduce the lead time too. The combination will reduce the required inventory for the greens a lot. Because they are the high volume products, this will also reduce the overall inventory by a lot.

On the other hand, the "red" items are only produced when there is available capacity, meaning their supply variabiltiy and lead times will go up, which in turn will mean that more inventory will be required for these items. However, because they are the low volume items, the impact on the overall inventory will be dwarfed by the gains from the "green" items.

Of course there are other more sophisticated ways of calculating safety stock, but they all come down to the elements of demand volume, demand variability, lead time, and lead time variability, meaning that the overall observation offered by Glenday is still valid.

I want to come back to the acceptance that stochastic behaviour is the norm, and the incorporation of these ideas into how we construct and operate our supply chains. This is a huge leap forward for which I am very thankful to the likes of Carol Ptak and the Demand Driven Institute, who have been promoting DDMRP for years.