8 reasons why Demand-Driven MRP (DDMRP) is not Lean (part 2)

Demand-Driven Material Requirements Planning, or DDMRP in short, promises to be the first real innovation to MRP since the invention of MRP. Furthermore, in one of its white papers, the Demand Driven Institute states that “Lean Finds a Friend in DDMRP” (2011). DDMRP is positioned as integrating Lean, the Theory of Constraints (TOC) as well as Six Sigma whilst fundamentally innovating the traditional MRP (and DRP) planning approach. What do I think of it? Some parts are OK, but most of it still has nothing to do with Lean. In a series of two posts of which this is the second (the first can be found here), I’ll try to explain my views and hope this may help you and your company in making the right decision when looking at your values, your thinking, your philosophy and your strategies.

DDMRP in a nutshell

As this is the second post in a series of two comparing Lean (JIT) and DDMRP, I will just provide a brief reminder of the key elements of DDMRP. The starting point of DDMRP is the observation that in most companies, many parts are either short or in excess. I guess many of us can relate to that. It is my experience as well. From that starting point, the conclusion is that there must be an optimal range for each part in which it is neither short (or service at risk) or in excess. This introduces the concept of the buffer in DDMRP. DDMRP is built around these buffers as its slogan also reveals: “position and protect”.

DDMRP consists of five major steps: (1) position the inventory buffers, (2) size the buffers, (3) dynamically adjust the buffers, (4) create supply orders in the DDMRP way and (5) prioritizing open supply orders based upon visual management of buffer levels. In the DDMRP terminology, step 4 is referred to as planning, and step 5 as execution.

In two posts I’ll review these five DDMRP steps in somewhat more detail and compare it to Lean JIT philosophy and approach. The first post discussed steps 1 to 3. This second post will further go into step 4 and 5.

4. Creating supply orders

Now here it really becomes interesting as this concerns the actual release of work into the supply chain (the authorization to produce or ship). DDMRP hereby reviews, daily, what DDMRP calls the “available stock”. Available stock is defined as the actual on-hand plus the open supply minus the demand for today (and possibly a known and qualified order spike that is within the replenishment lead time of the part and beyond the safety factor used for determining the red base zone). So, this is very close to the concept of economic inventory in traditional inventory management theory. Now when this “available stock” is below the green zone, the difference between the “available stock” and the top of the green zone (so the total buffer) is proposed as the supply order to be released. So, very similar to a so-called order-up-to or periodic review system (PRS, or a so-called P-Model).

To be honest, here’s why I strongly object DDMRPs claim to be similar to a pull system and that it shows characteristics of the Lean approach in this respect. It does not.

First, the pull system does not look forward in releasing work (not even today, let alone the whole replenishment period or “order spike period”); it acts purely and solely on consumption (something can still happen today…). It just releases whatever was taken today (possibly accumulated for the container quantity or agreed lot size, or the agreed review interval). Don’t forget that it isn’t only demand that consumes parts (never seen a forklift driver run into a pile of parts before?).

Second, when we have integrated safety into the buffer sizing, why look forward to a (potential) spike in the order book. Wasn’t this why DDMRP integrated variability into the initial or dynamically recalculated buffer sizing? As said, kanban does not look forward. Even so-called qualified sales orders can still (and frequently do) change, unfortunately.

Thirdly, DDMRP is in its basic appearance a periodic review system; the basic appearance of a kanban system is a fixed quantity system (a so-called Q-Model). Periodic systems (like the order-up-to system) do have a negative effect in case of spikes: they convey the spike further upstream. This is because it is a fixed timing, variable quantity system. The kanban system, as it is a fixed quantity, variable timing system, dampens it somewhat (depending upon the fixed quantity used), but instead increases the order frequency. Furthermore, the kanban system also lets itself adjust easily to a fixed period or even fixed sequence environments.

To show the effect of the typical fixed quantity kanban system and DDMRP, I have used the data of a DDMRP simulation by Chad Smith presented at the SMART Conference in Sydney, Australia in June 2013 and compared it to the kanban system, using the same data. The part is the same we already introduced before when determining the buffer and loop size.

DDMRP and Kanban simulation data:

DDMRP and Kanban Simulation Data

To better illustrate what happens in both approaches, I included the graphs of the actual on-hand stock and the generated supply orders below. 

DDMRP and Kanban Simulation: On Hand Inventory

DDMRP and Kanban Simulation: Orders

What you see happening is that an order spike is conveyed upstream which leads to a spike supply order and in the end an overall higher on-hand level, whereas kanban would only put in an additional call-off when confronted with such a spike. And I will not repeat what happens when the spike, in the end, is changed or even canceled by the customer. This clearly shows that DDMRP essentially still is an anticipatory and therefore push, system.

To finalize the brief experiment, a few figures from the simulation. The DDMRP on-hand target level was 70 parts, and the range 52 to 89. In the simulation, the average on-hand inventory was at 72 parts (so at target), and the range 32 to 105. In the fixed quantity kanban approach, the expected average inventory was 52 parts with a range of 35 to 70. The simulated average was at 50 parts (also at target). The range was from 15 to 72 parts. Overall, both DDMRP and kanban would provide 100% service, but kanban would do so with a 30% lower inventory level, and — not unimportantly — a more stable inventory and ordering pattern.

What you see here as well is that in DDMRP, inventory can still go through the ceiling depending upon the actual “firmness” of the orders ahead (covering the replenishment lead time). This is due to its still anticipatory nature. This violates the fundamental purpose of the kanban system, i.e., to prevent overproduction. Furthermore, as you can see from this simulation, the variability in the on-hand quantity is greater in the DDMRP case than the fixed quantity kanban case. And of course, I am sure I don’t have to explain that WIP equals lead time and what the effects of variable lead times are on a supply chain…

5. Visually prioritizing supply orders

In DDMRP, open supply orders are prioritized based upon the penetration of their corresponding buffers, a concept borrowed from the Theory of Constraints (TOC). In other words, the deeper you are in the red zone, the higher the priority. This is shown using the green, yellow and red colors and the percentage level at which the buffer currently is. The lower this percentage, the higher the priority.

For me, there are two important aspects to this. The first is technical. When one prioritizes based upon buffer penetration, knowing that each part buffer most probably will have a different buffer percentage related to it, it is very well possible that (1) priorities change every day depending upon what happens on the consumption side of the buffer, and as a result (2) that actual lead times start varying.

In the JIT system, there is one basic and simple priority rule and that is First In, First Out (FIFO). You keep the sequence in which work was released or you mess up the whole system. This ensures that lead times are stable, WIP is stable, promises can be made reliably and the whole system does not spiral out of control. Never forget, kanban is to keep lead times under statistical control. And in that respect, DDMRP is what Deming would call “tampering”. For those that are interested, I suggest you play Deming’s funnel experiment.

The second remark I’d like to make is how DDMRP piggybacks Lean when it comes to pull, but also Lean’s focus on visual control. But let’s not make the mistake to confuse the two. DDMRP is still focused on planners and their systems, whereas Lean’s focus is on engaging the whole organization. Visual control in Lean means creating a workplace where not only the managers and foremen but also all workers can detect trouble. Through visual control, all workers (not only planners) are taking positive steps to improve a lot of waste they have found (Sugimori, 1977). It is at the very heart of Toyota’s Respect-for-People principle as well as a key requirement for Jidoka and Kaizen. Let’s never forget, the physical kanban is a means of visual control in this sense and an example of Toyota’s “management-by-sight” philosophy. And visual control is not about making things green, yellow and red in a system…

DDMRP: a truly Lean approach?

So, is DDMRP a truly Lean approach? I hope this and the first post on DDMRP together have provided the arguments why I think DDMRP isn’t.

To summarize my learnings from these two posts contrasting the DDMRP and JIT approaches, I find that:

1. DDMRP combines demand and supply variability in the same buffer. The JIT approach would not do so to make clear where the origin of problems may lay. JIT points us into the direction where we can find the root cause. This is absent in DDMRP.

2. The JIT approach to buffer (or loop) adjustment is fundamentally different from the DDMRP approach. Whereas JIT (as part of Lean) forces us to see our problems, DDMRP (like many other traditional management approaches) manages around them by adjusting the buffer size. Like this, the challenge and kaizen mindset from Lean is absent in DDMRP.

3. The mixed-level loading or heijunka element of JIT is completely missing from the DDMRP approach which makes DDMRP more prone to variability than JIT.

4. DDMRP is not a pull system as it looks (albeit “mildly”) forward. It doesn’t act upon actual consumption but considers today’s demand plus qualified order spikes in the near future in determining new releases. This is fundamentally different and makes it into an anticipatory (push) system with the same issues (i.e., inventory is not necessarily in control and overproduction is not avoided by design).

5. Once defined, JIT makes use of its buffers and will not look at potential (or even “qualified”) spikes in the near future. To me, it seems like oversteer in DDMRP that it uses both buffer and future spikes in its release strategy. Although DDMRP claims to reduce the bullwhip effect, I think JIT does a better job at this.

6. DDMRP is a so-called fixed period system (a P-Model). Kanban in its most used variant is a fixed quantity system (a Q-Model). This has an important implication for the way variability is transmitted through the chain. P-Models are less suited for more variable environments than Q-Models. When demand levels change, Q-Models still always order the same amount but will adjust the frequency. P-Models will vary order quantities and order less frequently.

7. The priority rules in DDMRP will create a continually changing priority list and as a consequence, it will reshuffle orders and create variable lead times. JIT’s FIFO priority rule is not only simple and easy to understand for all, but also plays a vital role in stabilizing lead times and keeping inventory under control.

8. DDMRP is still focused on planners and their systems, whereas Lean’s focus is on engaging the whole organization. Visual control in Lean means creating a workplace where not only the managers and foremen but also all workers can detect troubles and through problem-solving can improve their workplace (kaizen). This is at the very heart of Toyota’s Respect-for-People principle and absent in DDMRP.

So, does that mean DDMRP does not work? Of course not, it will work in the sense that it will function as described. And compared to your current state of affairs it can well deliver a substantial improvement. As Nietzsche stated: “anything goes”… But let’s not mix it up with Lean and JIT. DDMRP, in some aspects, is fundamentally different from Lean.