Dynamic Buffer Management (DBM) – the breakthrough idea and several problems to solve

The most common procedure for maintaining stock of items is relying on a forecast, translate it to average daily sales/consumption and aim at holding fix number of sale-days in stock or defining min and max number of sale-days. That number of sale-days (or sale-weeks) is determined by a policy for a whole category of items, defined broadly by the supply lead time.

This common procedure leads to significant deviations from the determined levels in both directions causing shortages and huge surpluses at the same time.

The main flaws in the rationale of the common procedure:

1. The common procedure monitors the demand fluctuations and based on it forecasts the future. But it ignores the uncertainty in the supply time.  The stock-level should consider both the demand and supply fluctuations.
2. The current forecasting method is based on predicting the average demand, but ignores assessing the level of uncertainty (forecasting error). Thus, information regarding the stock that is required to satisfy the constantly fluctuating demand is missing.
3. Frequent forecasting increases the noise in the system.
4. The min-max definition encourages batching and slows down the replenishment frequency, which increases the impact of uncertainty.

The TOC key insights for holding stock are:

1. Considering not just the on-hand stock, but also the items ‘on the way’, meaning all the open purchasing orders should be part of the mechanism to provide good availability. The Target-Level defines the buffer of stock, including both on-hand and open orders.
2. The Target-Level is kept constant until clear signal is received that it is not appropriate.
3. Fast and frequent replenishments to the target-level.
4. Buffer Management is used for establishing one priority system for moving stock from one location to another.
5. Tracking the behavior of the buffers to decide whether the Target Level is too small or too large. This is the objective of the DBM algorithm.
   1. The idea is to check the combination of two different sources of uncertainty:
      1. The market demand – its ups and downs!
      2. The replenishment time – its own ups and downs, including the impact of the frequency of replenishments.
   2. There is no point in introducing small changes.
   3. The signal for increasing the buffer is too long stay and too deep penetration into the Red Zone of the on-hand stock.
   4. The signal for decreasing the buffer is too long stay at the Green Zone.

The breakthrough idea of DBM is monitoring the effectiveness of the protection mechanism rather than re-calculating the buffer-size. Both the demand and the replenishment time behave in an erratic way, which is difficult to describe.  The main difficulty is frequent changes in the environment, which upset the key parameters of the demand and supply time.  Events like the emergence of a new competitor, a controversial article in the media, changes in the economy or regulations all could cause a quantum change in the market demand.

The replenishment time is highly influenced by the operational management of the supplier and the state of load versus capacity. Changes in both factors could lead to significant changes in the replenishment time.

Re-calculation of the buffers when such a drastic change happens is problematic because the calculations rely on past performance. Sensing the actual state of the protection mechanism leads to taking quick actions based on the most recent past.  The quick response does not try to speculate the exact size of the change – just its direction: up or down.  Goldratt recommended increasing or decreasing the buffer, once DBM signals the need, by 33%.

The impact of DBM on the performance of the organization is quite strong and faulty DBM signals might be very costly. Constant learning should be used to tune its algorithm to the specific reality, especially identifying situations that require different reaction.

When the reason for the deep and lengthy penetration into the Red is (temporarily) inability to replenish, like when the source lacks inventory or capacity then DBM should not increase the buffer.

A conceptual issue is the fixed-ratio change of buffers. It even does not matter whether it is an increase or decrease. It is always possible that a change has been invoked, but after some time reality shows there has been no real need for the change.  In other words, short time after the increase there is a signal to decrease.  However, if we use 33% for any change then we end up with about 90% of the buffer before the increase.  The problem is that it is hard to explain that inconsistency.

An idea, raised by Dmitry Egorov, was to check carefully the behavior immediately after such a buffer increase in order to validate that it is truly needed. The result of an increase in the buffer is that the buffer status is deeper into the Red relative to the new buffer size. If after very short time the buffer goes up into the Yellow, then it should signal returning to the former size.

Similar behavior should be taken after decreasing the buffer. This move would temporarily make the on-hand stock to be above the new Green line.  If the buffer status goes down into the Yellow very soon – DBM should recommend increasing the buffer back to its previous size.

A related issue is the asymmetry of the DBM algorithm between increasing and decreasing the buffer. For buffer increase the algorithm considers the depth of the penetration into the Red-Zone. For decreasing the buffer the amount of penetration into the Green is not considered at all.  Actually there is a good reason to be much more conservative about reducing buffers than for increasing them.

The use of the replenishment time as part of the DBM algorithm is of concern to me, because the TOC algorithm does not monitor that time and its relevancy for the decision is dubious. The whole point of DBM is monitoring the combination of demand and replenishment time.  The only important need for the replenishment time in the DBM algorithm is for stopping further increases until the effect of the new size can be evaluated.  However, this can be done by monitoring the arrival of the specific order generated by the buffer increase.  The algorithm for buffer increase could be based on continuous stay in the Red-Zone taking the depth into account.  For decreasing the buffer there is no reason to refer to the replenishment time.  All that is required is a time parameter for too long stay in the Green.

DBM works in a similar way to forecasts, meaning it looks back to the past to deduce the near future. However, DBM looks only to the very recent past and considers only the actual state of the on-hand stock.

Should we use forecasts as additional information?

The idea is NOT to change the buffer unless there is a clear signal that the buffer is inappropriate. The additional information based on a forecast that considers additional parameters than DBM would be a rough estimation whether the current buffer is definitely too large or too small. Considering seasonality, knowledge of a change in the economy or the emergence of new products could add valid information to the decision whether to change buffers, and also give a rough idea by how much.  When the forecast points to a minor change in the buffer. less than 20%, the buffer size should be kept as is.

The above issues are, to my mind, central for coming up with an overall more effective way to control stock buffers. I always prefer to leave the final decision to humans, but give them the most relevant information to do that.  When millions of stock buffers are maintained throughout the supply chain at various locations, and 1-2% of the buffers seem to be inappropriate at any given day, it is practically difficult for humans to consider the changes for so many buffers.  At that case there is a need to let DBM, coupled or not with forecasts, to change buffers automatically.  This means the effectiveness of DBM directly impacts the financial and strategic performance of the organization.

DBM is important enough to encourage TOC experts to collaborate in order to come up with effective DBM specifications for software companies to follow. The full detailed solution should have a wide acceptance.  TOC is clearly against any “black box” algorithms.