According to Plossl, one of the initial differentiators of MRP versus previous approaches was in the understanding of demand.
It may seem obvious that if you have orders for 10 cars, you will likely need 50 wheels and 20 headlights (assuming you are providing spare wheels). But it's less obvious that you might need 250 lugnuts and 3000 type A45 left-threaded fasteners (OK, I made that last one up).
Maintaining correct inventories of parts when the manufactured products have complex bills of material has been a nontrivial challenge. Earlier (pre-MRP) approaches to ensuring sufficient stock used concepts such as of Order Point, in which the "bin" of parts was monitored as a stand alone stock and re-ordered when it reached a certain, pre-designated low point.
The trouble with this is that the tail wags the dog. It constrains the productive capacity of the operation to whatever someone has defined as "usual," plus or minus fudge factors. There is no link to real demand as generated by the sales function, and so shortages and overcapacity resulted.
To quote Plossl (page 21-22):
"Orthodox inventory analysis and classification techniques are designed to determine the most desirable treatment of a given inventory item or group or items. They employ various attributes of the items such as cost, lead time, and past usage, but none of them takes into account the most important one, the nature of demand. Yet it is the nature (or source) of demand which provides the real key to inventory control technique applicability and selection. The fundamental principle in the selection of order point or material requirements planning is whether demand for the item is dependent or independent of that for other items."
"Independent demand must be forecast, but dependent demand can be calculated since it is directly related to, or derives from, the demand for another inventory item or product...Dependent demand need not and should not be forecast; it can be precisely determined from the demand for those items that are its sole cause."
"OP/EOQ systems are oblivious to the dependence of one inventory item on another. Order point views every inventory item as though it had a life of its own; thus, forecasting and OP are inseperable. But all forecasting . . . attempts to use past experience to determine the shape of the future and succeeds only when the future continues to be like the past."
"In a manufacturing environment, however, the objective is to make the future better than the past; future demand for a given part will almost invariably be different from past demand. Marketing and sales people and competitors are working diligently to make it different. Forecasting, therefore, should be the method of last resort, never used when it is possible to relate one item's demand to that for another item." (Emphases supplied.)
As a side note, it seems to me that dependent versus independent is a relative distinction. We can say that the car is the unit of independent demand, but someone might say, "no, the true unit of independent demand is the passenger mile and the demand for cars is dependent on that." An important point, but tending to the academic. I am looking for an operating model here that I can use as a basis for management.
Let's translate this into IT terms. There are various components subject to demand:
- Space
- Power
- Cooling
- CPU
- RAM
- Network
- Storage
- Skilled staff
- Software licenses
I am not including time or money, because those are more fundamental (everything above requires some amount of each). I also am not including functional applications, as this is a discussion about infrastructure. (Once I solve for this special case, I might move on to including apps.)
Now, just like multiple bins of parts, all of these can be forecast independently. But are they independent? No. The independent driver of demand is the IT service. (And per above, IT service demand is in turn driven by business need, but for the sake of the operating model, let's leave it as the independent concept.)
Demand management is discussed in ITIL version 3 in the Service Strategy volume, and capacity management in the Service Design volume. Service Strategy states on page 130 that "every additional unit of demand generated by business activity is allocated to a unit of service capacity." It's not clear to me what a "unit of service capacity" is (Service Strategy discusses "service units" but those appear to be organizational capabilities). A service capacity unit however sounds like something different - and I propose it might be the fundamentally countable unit of independent demand.
Back to forecasting: how many of you have an integrated IT capacity forecast? That is, your demand for all of the components bulleted above is integrated into an overall IT service demand model. If you have such a thing, awesome - because I can't find any IT management products supporting such a beast. (I've actually been prowling around on shop floor control software sites.)
ITIL v3 gets this at least at a high level, stating in the Service Design volume that Capacity Management as a whole requires sub-processes of Business Capacity Management, Service Capacity Management, and Component Capacity Management. But the ITIL discussion is biased to the operational, focusing on availability in terms of SLAs. What about ensuring that enough space, power, and cooling is available for incrementally added services?
This is a nontrivial bill of materials (and services) problem, which I will further examine in the next post.
-Charlie