Demand Driven – do not try to adjust everything to (re)size your buffers

The DDMRP model is based on the use of buffers that adapt to changes in demand or in the production environment. This adaptation comes mainly from the regular updating of the parameters that govern the size of the buffers.

However, behind each buffer there are several parameters to be adjusted at the same time; such as lead time or variability factors for example.

When you want to set up a DDOM or update it in the DDS&OP process, is it necessary to set all these parameters at the same time?

Is the performance the same or better if some parameters are defined as « fixed »?

This raises a number of industrial and scientific questions; addressed in Guillaume Martin’s PhD work defended in November 2020 « Dynamic Control of Demand Driven Sales and Operations Planning » carried out for AGIRE joint research laboratory.

What are the problems encountered?

Taken independently, each buffer is sized by a set of at least 5 parameters;

  • the Average Daily Usage (ADU),
  • the Decoupled Lead Time (DLT),
  • the Lead Time Factor (LTF),
  • the Variability Factor (VF)
  • and the spike detection horizon (for its integration in the net flow equation).

The definition of these parameters is outlined by the Demand Driven Institute (DDI), but not necessarily the rules for changing them over time.

Taken as a whole, one must also take into account the effects of changes in buffer parameters on the entire system; with much more security, my reference will certainly be delivered on time, but at what « cost » to the other references?

Finally, from the manager’s point of view, it is sometimes humanly unfeasible to update several hundred buffers easily and quickly.

Work conducted to address these issues

In order to help managers in their decision, a design of experiments was conducted to test different combinations of possible settings. These methods were extracted from the DDI training courses. Techniques used in other management methods and feedback from projects realized by AGILEA.

In order to be able to test all possible combinations, a simulation tool was developed to assess the impacts of changes in buffer settings on the entire organization.

Conclusions of the study

The results of the study are summarized in a decision tree to guide the decision maker (Figure 1).

The two most impacting parameters are (logically) the DLT and the ADU. The synthetic tree shows only the decisions affecting these parameters.

Figure 1: synthetic decision tree and recommendations

Two main categories stand out for choosing the policy to apply and for updating the parameters; does the stored product have a seasonality profile or not?

In a second step, the user can choose his update method according to the performance criterion that is the most relevant for him:

  • maximizing the customer service rate,
  • minimizing inventory and work-in-progress,
  • minimization of lead time,
  • or minimizing the load on the bottleneck.

In many cases, the most effective setting policies happen to be counterintuitive:

  • Keeping the parameters constant instead of varying them all the time allows you to be reactive but not too reactive. Because too much reactivity can make the system unstable (and have the opposite effect of the one you want),
  • It is sometimes more interesting to control the size of the buffers by varying the DLT rather than the ADU. In particular thanks to a policy called « single DLT », which proposes to give a single DLT value to all the products of a buffer stage.

In conclusion, before wanting to size or resize your buffers, it is necessary to identify which parameters are really important to review. This depends essentially on the type of application and the performance objective.

The work carried out has thus made it possible to identify 8 typologies of relevant combinations for the parameterization of buffers according to the various situations studied. For more details, you can access the thesis manuscript.

Of course, this work opens up a large number of perspectives. One of them would consist in the industrialization of a complete tool that automatically recommends the type of parameters to be modified according to the nature of the organization being evaluated (demand, seasonality, etc.).

Authors: Romain Miclo & Guillaume Martin