The sizing and adjustment of decoupling buffers in DDMRP is covered in Chapters 7 and 8 of the book DDMRP version 3. In this article, Steve Allanson proposes an alternative approach, which is applicable to both DDMRP buffers and more generally to Safety Stocks.

It is worth stating that my analysis is based on many years in FMCG and is most specifically relevant to this sector.

It took a full reading of the relevant chapters in the DDMRP book for me to understand the buffering concept used therein. The concept of buffers is certainly a significant advance over the way traditional MRP treats safety stock and the principles behind positioning are excellent. One of the key realisations for me was that the buffer “stocks” and the “Net Flow Calculation” includes inbound stock in transit.

The calculations described in the DDMRP book are relatively straightforward and incorporate some methods for simplifying even further. The method correctly uses lead-times, variability of demand and supply and also allows for the use of cycle times and minimum order quantities. These are combined with an Average Daily Usage figure to derive the various elements of the buffer (Red, Yellow, and Green zones) and it is this last element that, in my opinion, leads to two flaws in the methodology.

- This methodology essentially assumes that the historic ADU will continue into the future. The book does mention the possibility of using a forecast to create a forward looking ADU but it is still only an average. The book touches on the problems this can create and proposes several overlays to try to address the issue.
- The result is a number of cases/tonnes/boxes etc. but this is fixed until it is reviewed and adjusted. This is not truly dynamic and certainly is liable to be a problem in very fast moving environments – e.g. FMCG.

DDMRP proposes that Buffers should be calculated as part of the DDS&OP process and in one sense I agree with this – but I suggest that this should be a more tactical level of adjustment and that this is possible if the buffer or safety stock is made more dynamic.

I think this is a key area where the ambiguous relationship of DDMRP with forecasts

*(I prefer, and will use going forward in this article, the term Demand Plan (DP). In my use of the term a forecast is the projection forward by an algorithm of historic sales. A Demand Plan is the result of overlaying on this detailed sales and market knowledge of what will be done in the future e.g. promotions, new items, new customers etc.)*

**So how can we improve buffer and safety stock calculations.**

DDMRP proposes that Buffers should be calculated as part of the DDS&OP process and in one sense I agree with this – but I suggest that this should be a more tactical level of adjustment and that this is possible if the buffer or safety stock is made more dynamic

It is worth stating that I believe that Demand Planning should be at the heart of all good Supply Chain Management, whether you use forecasts, actual orders or a mix of both to generate orders. I differ from the apparent message of DDMRP in that in my experience solid, accurate SKU level demand plans are possible with the right process and, indeed, that good demand planning reduces the uncertainty of the demand signal rather than increasing it.

In figs 1 and 2 we can see the difference between the safety stock requirements using an ADU approach (1) vs that used based on DP accuracy (2). If the DP process really is giving inaccuracies equal to or greater than simply using total variance then there is a significant issue with the process which needs to be addressed.

Fig 1

Fig 2.

The second point of difference is that I believe safety stocks or buffer stocks expressed as fixed volumes are inherently flawed in a dynamic environment. They will need constant adjustment and will never keep up with the dynamic marketplace. Many legacy MRP and MPS systems and even some APS systems still use fixed volume stocking targets. These should, in my opinion, be avoided.

**Calculating a dynamic, DP accuracy based Safety Stock**

Firstly we need to determine the correct pairs of DP and actual sales to use – they must of course be at the SKU level. This depends upon the total time required to respond to a change to the forecast. In the case where we have decoupled stocking points this will mean the lead-time is dependant upon what the DDMRP book calls the Decoupled Lead Time (DLT).

Next we need to know how long it takes for the Supply Chain to respond to a deviation from planned sales. (The Response Time RT) In the terminology of DDMRP this would be the Decoupled Lead-time. The data we need to consider is a series of cumulative actual orders for the length of the Response Time versus the Demand Plans for these periods.

Once we know which pairs of data we are comparing we can calculate the deviation for each pair. I prefer to see this expressed as Actuals/DP. So sales above DP would give a figure above 1 and sales below DP a figure below 1.

Of course where the Sales Order visibility does not extend as far as lead-time then without taking into account Demand Plans the dynamic nature of the buffer would again be compromised

For the following analysis to be statistically valid we need at least 30 data pairs and the more we have the better will be the analysis. It is worth plotting the results but I find generally it is safe to assume they will follow a Normal or Gaussian distribution. (Where this is not the case there may be other more suitable distribution curves or in extremis it is possible to simply sum the data points to derive the numbers to be used).

We need to know the level of service (typically casefill in an FMCG environment) we need to achieve. In the UK 98.5% is a standard requirement of the retailers. From the Normal distribution we can compute the Mean and Standard Deviation of the dataset of Actual/DP as in Fig 3.

Fig 3

From the standard Normal Distribution algorithm we know that the 98.5th percentile will be at 2.175 Standard Deviations above the Mean. So now we know that if we are to be able to deliver against orders during the response time to cover 98.5% of possible sales occurrences then we need to know how far from the DP this 98.5th percentile is.

98.5th Percentile = Mean + (2.175*STD)

But what if there is any ongoing bias in the DP? If the mean is below 1 then we need to cover less upside from the Mean but if the mean is above 1 then we need to cover more upside from variability because we know we will routinely have actuals above the DP (Figs 4 and 5). We can see that our calculation automatically compensates for any ongoing bias.

Fig 4

Fig 5

We know that our demand planning process will mean that we have stock to cover the DP so in our calculation we need to remove this from our safety margin.

So the Safety margin is calculated by

(Mean-1) + (2.175*STD)

And finally we need to factor back the response time so the full Safety margin calculation is

Example: Item 0201 – Response Time 2 days

**{(Mean-1) + (2.175*STD)} x RT**

Note because RT is in days and all other factors in the equation are scaleless then the result is a number of days of cover.

Our sets of pairs result in a mean of 0.9 and a Standard deviation of 0.2 and so with a Response Time of 2 days then the result will be

**{(0.9-1) +(2.175*0.2)} x 2 = 1.67 days**

**So for this item TMS = 1.67 days.**

N.B. Stock includes Stock in transit not just on hand stock.

*Note – if you really want to stick with the DDMRP methodology and ignore DP, or if you really do find that your DP variance is higher than just looking at total sales variance (Highly unlikely in my experience!) then just substitute the STD of cumulative sales across the Response Time.*

**What are the benefits of a Dynamic Days of Cover Safety Margin?**

Firstly Demand Plan accuracy does not change rapidly and therefore the agreement of the TMS figure sits comfortably in the S&OP cycle. Let’s look at our item from above and examine the production and stock status with TMS set to 1.67 days (Fig 6).

Fig 6

We can see that the TMS in cases rises prior to the Demand and Stock follows. Production is offset from the demand. This is all achieve without the need to manually adjust TMS. In the case of DDMRP buffers – if they were similarly expressed as Days of Cover then there would be less need for adjustments to Buffers in anticipation of spikes or short, medium or long term changes to demand.

Of course where the Sales Order visibility does not extend as far as lead-time then without taking into account Demand Plans the dynamic nature of the buffer would again be compromised.

**In Conclusion**

DDMRP adds useful functionality over traditional, standard MRP in its use of decoupling buffers. However, possibly due to the ambiguous relationship of DDMRP and forecast (or Demand Plan) the methodology for sizing the buffers falls back on fixed volumes.

Fixed volumes for target stocks, Target Minimum stocks, Safety Stocks etc. have been a flaw with many MRP, MPS and APS systems for some time. I suggest that they are also a flaw with DDMRP – Buffer stocks can be expressed and used as Days of Cover – although with long lead time items this will require the use of Demand Plan for full dynamic resizing of the buffers.

Finally I suggest that with a robust Demand Planning process the variability required to be covered by buffer or TMS is significantly lower when based on deviation from Demand Plan rather than total sales variability. This has certainly been the case in my long career in FMCG. ♦♦♦♦♦