16. Demand Segmentation Overview Typically, when examining Customer and Market demand, the approach is to calculate the figures based on averages. In reality this gives only half of the picture, because variation in demand can have as big (and in fact usually much greater) an effect on a process. Demand Segmentation[26] examines both the average and the variation in demand in one graph, as in Figure 7.16.1. [26] For more detail see Manufacturing for Survival: the how to guide for practitioners and managers by Blair R. Williams.
Figure 7.16.1. Volume of Demand versus Variability in Demand.[27] 
[27] Source: SBTI's Lean Sigma for Healthcare training material.
Products in the top left corner of the graph exhibit smooth high volume demand. Products in the bottom right-hand corner exhibit variable, low volume demand. Most production planning, forecasting, and process leveling approaches mistakenly consider these in the same way, which misses opportunities to create more predictable, responsive processes and reduce inventory all at the same time. As I tell Belts in my classes, the statistics tools are great, but Demand Segmentation is probably the most powerful tool in the Lean Sigma arsenal. Project examples include Industrial Product rationalization, raw material rationalization, warehouse stocking Healthcare On-floor medications inventories Service/Transactional Product rationalization
Logistics As a simple analysis tool this can be applied by the Belt without the rest of the Team; however, the data might have to come from multiple sources and often requires Team involvement to collect it. The analysis itself is done entirely in a spreadsheet, such as Excel, and can be done in a matter of a few minutes after the data is in the correct format. Initially, data is historical, but after the organization understands the application of the tool, forecast data can also be used. Roadmap Step 1. | Identify which entity types are to be examined. Demand Segmentation works best with at least five entity types to make the graph sensible. If fewer entity types are examined, it is probably best to look at a simple Demand Profile for each, which gives more detail.
Data is captured for each entity type separately. Identifying the entity types is not as easy as it first appears. The trick is to ask what problem is to be resolved. For example, in Production Planning there might be 25 different Customer products and you would like to schedule production based on their demand. At first glance, you might jump to the conclusion that there are 25 different entity types. Perhaps the 25 products are only six differently labeled manufactured products for different Customers. If you look at Production Planning, you would choose to use the six entity types rather than the 25, because those are the products you manufacture. However, for the same problem, if you look at Finished Goods rationalization in the warehouse and products are stored there pre-labeled, then the 25 entity types would probably make sense. The valuable tool "Why do we care?" helps us here.
| Step 2. | Identify the demand buckets, for example, days in a month or weeks in a year, depending on the drumbeat of the demand. It is necessary to have at least 25 buckets of captured demand to get good measures of the average and particularly variation. The buckets themselves have to be meaningful and often it is useful to take the typical replenishment cycle as the driver. For example, in hospital care units, drug inventories are kept on the floor in electronic vaults with the medications being dispensed directly from them. It is obviously important not to run out of medications, and volume and variation in demand are key to calculating how much to put in there. Therefore, if the vaults were replenished once daily then a day would be a reasonable time bucket for the segmentation.
| | | Step 3. | Data is collected from the downstream Customer for different entity types across the time buckets, as in Table 7.16.1. The most common mistake is to look to the Process Planning group for when we decided to make the entity, not when it was actually demanded by the Customer. Customer demand rates are typically much smoother than we care to admit, and in fact, their usage rates are even smoother. Internally, we tend to batch entity processing into large lots, which we make on an infrequent basis; so it shows much higher variation in demand than is actually there. For Demand Segmentation, we need to look at demand patterns, not our own planned process patterns.
Table 7.16.1. Capturing and Analyzing the DataProduct Line X | Week 1 | Week 2 | Week 3 | Week 4 | Week 5 | Mean | S | CV |
|---|
Product A | 1700.00 | 9.00 | 230.00 | 1.00 | 10.00 | 390.00 | 660.69 | 1.69 | Product B | 420.00 | 333.00 | 380.00 | 550.00 | 390.00 | 414.60 | 73.24 | 0.18 | Product C | 10.00 | 1.00 | 3.00 | 0.00 | 2.00 | 3.20 | 3.54 | 1.11 | Product D | 7.00 | 5.00 | 3.00 | 1.00 | 4.00 | 4.00 | 2.00 | 0.50 | Demand | 2137.00 | 348.00 | 616.00 | 552.00 | 406.00 | 811.80 | 739.48 | 0.91 |
Examples include
- Healthcare floor medication inventories Consider actual pulls from the vault
- Raw Material segmentation Production usage of the raw materials
- Production segmentation Customer demand of manufacturing codes (not product codes)
- Product portfolio Customer demand of final product codes
| | | Step 4. | After you have the demand data for each entity type across multiple time buckets, you can calculate the mean of the demand and the coefficient of variation (COV), defined as
The COV is used rather than the standard deviation because it is the size of variability relative to the total level of demand that is important, rather than the size of the variation itself. For example, if there is demand for an entity type from the Customer with a mean of 1,000,000 units per month, then a standard deviation of 50 units is neither here nor there and the demand is considered to be smooth. However, if the mean of the demand is only 100 units and the standard deviation is the same 50 units, then the demand cannot be considered as smooth. The general rule here is that a COV less than 1 is considered to be very smooth. A COV greater than 2 is highly variable.
The mean and the COV are calculated for each entity type as shown in Table 7.16.1. A Total demand calculation is also shown in the table, which gives an indication of the total variability in demand seen by the process. However, at this stage the important numbers to examine are for the individual entity types.
| Step 5. | From the data table, create a graph similar to the one shown in Figure 7.16.1, taking the mean of demand as the x-axis and the COV of demand as the y-axis. Each entity type has its own point on the graph.
|
Interpreting the Output Interpreting the Demand Segmentation graph depends heavily on application but can be addressed by breaking the graph into zones, as shown in Figure 7.16.2. Figure 7.16.2. Demand zones. 
Application: Replenishment Demand Segmentation can be applied to materials management and specifically replenishment as replenishment to a Customer or replenishment of materials internally or from a Supplierfor example, the delivery of product to a Customer, a materials delivery to a line, or a medications delivery to a Care Unit. In this case, the zones would be treated as follows: Zone 1Deliver direct to line. High volume, low variation materials usage is so smooth that it allows us to add service value to the Customer by managing their inventory for them. Materials would be delivered on a rate-based system directly to the point of use (POU). There is probably no need to keep anything more than a small buffer of Finished Goods at the end of the supplying process for these types of entities. Payment could be based on a Blanket Order and then Call Offs made as entities are used. Zone 2Pull System from Customer. For the middle majority, simple pull triggers from the Customer would allow replenishment when needed (see "Pull Systems and Kanban" in this chapter). If the Global Process Lead Time for entities in this Zone were too long (for example, it is impossible to process the entity from scratch and get it to the Customer POU within acceptable timeframes), then a solution would be to replenish from stock, by keeping a small inventory of Finished Goods at the end of the supplying process. See also "TimeGlobal Process Lead Time" in this chapter. Zone 3Make (or Deliver) to order. High variation, low volume gives such unpredictable demand that it forces the Supplier to deliver only when there is an order. If the Global Process Lead Time for the supplying process is short enough, then there isn't too much of a problem. However, if the Global Process Lead Time is beyond acceptable bounds for delivery Lead Times, then either: The supplying process (or the POU) has to stock inventory, which is probably highly unpalatable Work should be done quickly to reduce the Global Process Lead Time The validity of having the entity types in the portfolio should be questioned (see the next application section) Or the worst case is that the promised delivery time needs to be extended with the Customer
Application: Product Portfolios Demand Segmentation can be applied to the Product Portfolio to identify opportunities for rationalization and to validate the value in the portfolio. In this case the Zones would be treated as follows: Zone 1Dedicated Business Unit. Products in this Zone have high-volume, low variation demand and could be managed independently, either by dedicating a small internal group to manage them or to spin them off as a whole new Business Unit. If the entity types in question have reasonable margin, then these businesses often are known as the "Cash Cows." Zone 2Majority of portfolio. Products in this Zone are probably best left as they are. There is always opportunity to rationalize in this majority, but they typically aren't a primary focus. Zone 3High-value niche products. Unless they are high-value products, the validity of keeping low volume, highly variable products in the portfolio is questionable. This is the first portfolio area to look to for rationalization or obsolescence opportunities. Other opportunities here might be to combine products in these Zones from multiple operating sites and make them just at one site, thus creating an elevated volume at the single site (and usually a reduced variability too, due to the Central Limit Theorem) and freeing up the other sites from the burden.
Application: Production or Operations Planning Demand Segmentation when applied to Production or Operations Planning allows processing of the entity types in the different Zones to be planned more effectively, as follows: Zone 1Repetitive flow rate-based scheduling. Entity types that have a large, smooth demand do not need to be scheduled individually, they can be rate-based, so that during each time-period a consistent amount is processed with the knowledge that the Customer/Market uses that amount. Slight variation in demand is catered for by using small buffers of inventory, preferably at the POU. For example, in the light bulb manufacturing industry, there is a consistent high level of demand to which the processes are paced. No one in Operations Planning for those companies takes all the orders from DIY and hardware stores on a daily basis and determines what should be made during that shift. Zone 2Hybrid control Pull System. For the middle majority of entity types in this Zone, Operations can be successfully governed using internal Pull Systems, to minimize work in process (WIP) inventory. Zone 3Discrete job order. For entity types in this Zone, the orders are few and far between and highly variable, so it makes little sense to process them without an order or Customer request.
|
