12 January 2010

What does “demand-driven” really mean?

Recently I stumbled across a whitepaper entitled Demand-Driven Inventory Management Strategies: Challenges & Opportunities for Distribution-Intensive Companies (Fraser and Brandel 2007) prepared by Julie Fraser and William Brandel, principals at Industry Directions, Inc. What I found amazing about this article is that it doesn't really get to the point of "demand-driven inventory management strategies." Instead the focus is on better systems, better data, better use of the data, and better forecasts at the SKU level (rather than at the "product family" or "product category" levels).

Now, maybe I'm an idealist, but when I think of "demand-driven" inventory, I think of an integrated supply chain that functions in such a way that when an end-user takes a unit of product off the shelf at the retailer's store (or whatever model is being used), that action triggers the production of one unit at the manufacturer's plant with a minimum of mid-stream manipulation.

See the accompanying illustration, and let me describe for you my concept of a demand-driven supply chain. We will start at the bottom of the illustration – where the action begins – at the retail store. Ideally, each retail store should stock just enough product to cover one day's sales plus a "buffer" to allow for expected variability in demand.

At the end of each business day, the retailers should transmit to their associated distribution center (DC) the quantities sold of each SKU in the supply chain depicted. It doesn't matter whether the DC is owned by the retail chain or the distributor, the process would and should work the same.

Next, depending upon the agreed replenishment cycle (although daily is ideal), the DCs would prepare replenishment orders to be shipped to the retail outlets. The goal would be to replenish exactly the quantity that was reported as sold (plus or minus any adjustments for seasonality, special promotions, etc.) at each outlet. Meanwhile, the DCs will have reported to their supplying warehouse how many units of each SKU that they have sold (again, plus or minus any adjustments).

Each DC should stock only enough of each SKU to cover the replenishment cycle from the domestic warehouse plus a "buffer" to cover any variability in demand. On its scheduled replenishment cycle – and, again, daily is ideal – the domestic warehouse should ship out replenishment orders to the DCs. In the meantime, if the replenishment cycle is longer than one day, the domestic warehouse will have transmitted daily sales numbers back to the off-shore warehouse, so that the consumer purchase made at the retail outlet is transmitted all the way back to the manufacturer within one business day.

Following the pattern we have discussed already, the domestic warehouse should carry just enough of each SKU to cover variability in demand and supply (lead-time). The size of the "buffer" should include a calculated allowance for disruptions in the supply chain where it is most vulnerable (e.g., overseas transportation, or other). Naturally, since this represents aggregate demand, estimates of demand will be more accurate at this level than they will be at either the DCs or the retail outlets. As a result, the domestic warehouse inventories will be larger, but not nearly as large as if each lower level in the supply chain tried to estimate (read: forecast) demand for periods into the future. This approach helps keep inventories to a minimum and makes the whole supply chain more responsive to changes in demand.

Likewise, the foreign port warehouse should carry just enough stock of each SKU to cover variability in demand from the warehouse(s) on the opposite shore plus any variability in lead time from the manufacturing plant which, as we shall see, should be near zero.

Since estimates of demand variability will be most accurate at the level that demand is most highly aggregated, the manufacturer is the most reasonable place to keep the largest "buffer" of inventory for the SKUs in the supply chain. The manufacturer should carry enough stock of each SKU to cover production lead time variability. (Typically, this buffer length should be only three times the actual production time for each SKU. That is to say, if a day's aggregate supply of SKU #1001 can be produced in a single day of production, then the buffer for SKU #1001 should initially be set for three days. Then production should be scheduled in batches as small as is practical for the SKU.) Generally, production should be scheduled at the manufacturing plant based on producing the actual demand reported via the supply chain plus enough to fill any "holes" created in the buffer created by unusual demand in a prior period.

Now, that's what I call a "demand-driven" supply chain. It is do-able and it makes life better for everyone. Here's why:

  • Lower inventories everywhere
  • Reduced write-offs due to obsolescence
  • Lower inventory carrying costs all across the supply chain
  • Manufacturing is more responsive to changes in market demand – not separated from real feedback by weeks or months
  • Fewer lost sales due to stock-out (And the value of lost sales are almost always under-estimated in supply chain calculations simply because they are done by "averages," but the items most likely to suffer stock-outs are the most popular selling items, not average performers.)
  • Reduction or elimination of expediting costs across the whole supply chain
  • Dramatic reduction in overstocks (and about 73% of companies report overstocks simultaneously with expediting for items that are running short), which leads to less price-cutting to liquidate unneeded inventories
©2010 Richard D. Cushing

Works Cited

Fraser, Julie, and William Brandel. Demand-Driven Inventory Management Strategies: Challenges & Opportunities for Distribution-Intensive Companies. White paper, Boston, MA: Industry Directions, 2007.


Liza Mellin said...

The demand-driven inventory management pretty much sounds like how the Kanban method works.

Downstream = manufacturers -> retailers
Upstream = retailers -> manufacturers

In Kanban, the reporting of the number of SKUs sold happens upstream, while the delivery happens downstream. I have a question though. How large should the buffer for that variation of demand be?

RDCushing said...


You are correct. There are certainly similarities between kanban and the method suggested in my post. However, kanban suggests a signal when inventory reaches a specified level, whereas near real-time feedback through the supply chain means the entire supply chain is kept up-to-date with demand data at all times.

As to buffer sizing, there is no one formula to apply. It depends on many factors and the size of the buffer for one item in your operations may be entirely different than the size of the buffer for another item.

Some of the factors that should be considered in buffer sizing are:

1) Seasonality
2) Demand variance from the mean
3) Replenishment time
4) "Sprint" capacity (that is, when necessary, how fast can the supply chain make up a shortfall in short bursts of high-levels of effort, and how costly is it do so)
5) Throughput value of the item
6) Product affinities (if sales are lost on this item, is it likely to affect the sale of other items, as well)
7) Difference between the mathematical mean and mode of product demand

I trust this helps. Good to hear from you.

Liza Mellin said...

Thank you so much RDCushing! More power to you and good health.