TOCICO

Dynamic Buffer Management (DBM) was developed by Dr. Goldratt to ensure that product availability is protected against market demand fluctuations. Buffer size, trigger for buffer revision, frequency of buffer revision, and quantum of buffer change are the four important parameters which determine the effectiveness of DBM. DBM practices were developed when access to market demand data was limited, computing power was moderate, and application of AI/ML was still in its infancy. Now that we have access to faster, more frequent, and granular level demand data, computing power is higher, and AI/ML can be used effectively for short-term demand prediction, it is time to revisit our current practices of setting and revising the above four parameters to make DBM even more effective. Buffers are expected to be set at ‘maximum demand during average replenishment lead time (RLT), adjusted for supply reliability’. Initial buffers are set using this formula and changes are made based on whether the actual inventory stays in the ‘Red’ or ‘Green’ zone for a specific time. If that happens, buffer is resized, either up or down, by a certain percentage point, which is 33% in most implementations. Such periodic quantum changes often cause stress on the supply side, which takes time to adjust to the new reality. In a retail scenario, for example, where demand is predictably low on weekdays and high on weekends, we end up setting buffers at the higher level throughout the week. We experimented with sticking to the definition of buffer size as ‘maximum demand during average RLT, adjusted for supply reliability’. We used AI/ML based algorithms to predict short-term (RLT) demand at the most granular level where buffers are kept. These predictions were used to reset the buffer size and refreshed daily. This resulted in smaller and more frequent buffer changes, with smoother supplies. The results have been encouraging, with a substantial improvement in availability and reduction in overall inventory.