Demand Forecasting

One logical addition to a dynamic order management application is some type of end-use demand forecasting. The ability to factor in trends and trend changes, repeated demand patterns, and special types of patterns may produce orders that more closely match demand. If so, transaction and inventory costs as well as stockouts should be reduced. Perhaps even more important, the substantial but hidden costs of nurses dealing with supplies stockouts should be reduced as well.

There are many simplistic ways of generating demand forecasts. These typically are backward-looking. They assume that the future will look pretty much like the past. But maybe it will not.

Forecasting can certainly be informed by the past but it is really about forward-looking. What you can often learn from the past is the behavior that drives demand. Knowledge of behavior can then be translated into demand forecasting rules, not into forecasts themselves.

Cyclicality in Demand

Sometimes the rules involve cyclicality. Demand may follow a weekly cycle—high early in the week and tailing off to almost nothing over weekends. Or the cycle may have a monthly or seasonal (e.g., flu season) cycle as well. Annual cycles may also be important. Vacation periods affect demand as physicians and other medical staff take time over the traditionally slow summer period.

Erratic Demand

We may see patterns that might best be termed "erratic" — random spikes up and down superimposed on a fairly steady demand mean. These patterns might call for a point of use stock level that reflects the mean demand, not peak demand, with the peak demand over mean being supported by shared stock on the floor or in central supply. Many fast-moving items exhibit this pattern.

Episodic Demand

Episodic demand is characterized by a spike, or occasional spikes of demand, separated by periods of zero demand. Sometimes you can relate the spikes to events but typically they appear to be quite random, at least in timing. What may not be random is the demand spike volume and duration. This knowledge can help in forecasting demand.

Stable Demand

Stable demand typically has a high mean volume and a relatively low deviation about this mean. For such items, safety stock might be handled most efficiently in shared locations, either on the same floor or in central supply. More importantly, stable demand can often be blanket-ordered, with a single purchase order covering a quarter-year or even a full year of demand. Vendor shipments against such orders are scheduled into their order stream automatically in the same quantity and timing for each shipment.

Just-In-Case Demand

Most hospitals have a large number of inventory items (and often a large inventory investment) that do not move. They are kept on hand, often in many locations, just in case they are needed. The demand forecast for such items is zero but the stocking strategy needs to minimize duplication. You may need just one item stocked in a location that is accessible to all users if the item must be available quickly.

Slow-Moving Item Demand

These items clog up storage locations and are often hard to forecast. Occasional use appears as small blips on a background of zero demand. It may be possible to store these items in shared locations where aggregated demand is high enough to be forecastable.

As these examples should illustrate, forecasting demand is fairly complex and is closely linked to storage and space utilization strategies.

Automated Point-of-Use Ordering

The growing use of automated stocking and ordering systems raises the issue of compliance with the system's removal, restock, and counting requirements. Compliance has a major impact on ordering ... Managing Around Compliance ...

To return to the main topic list ...

 

Mechanical vs. Smart

Most forecasting routines are mechanical, without behavioral or pattern-based capabilities. Moving averages of demand series, or regressions using a few demand-related variables, are the norm.

Smart forecasting routines go beyond the backward-looking approach to incorporate an understanding of any identifiable patterns in demand. In addition, they can be matched with stocking and space strategies to provide a sort of optimization.

Our simulation models include a number of "smart" forecasting capabilities.