Northern Illinois – my neck of the woods – is supposed to get a foot and a half of snow tomorrow – maybe. With their $5 billion budget and network of satellites, tracking stations and massive computers, staffed by legions of thoroughly degreed technical wizards, the National Weather Service is still not sure – an 80% chance of it happening, they say. Two days ago there was no snow in the forecast at all.
Don't get me wrong. I am not criticizing the National Weather Service. Far from it. I don''t know an isobar from Snickers bar and I am amazed at what those folks know and do. And I believe them when they look at all of the data at their disposal and acknowledge the reality that there are so many variables at play that things could change even in the next 24 hours, causing the weather tomorrow to be completely different.
What I am questioning is the folly of supply chain managers who still think the key to customer delivery and factory optimization lies in forecasting. While the weather people don't know for certain if a monster snow storm will hit me tomorrow, lots of supply chain people (maybe most of them) delude themselves into thinking that either sales people or some newfangled variation on regression analysis can accurately predict the quantity of each widget customers will order in any given week a couple of months out into the future.
I don't know much about arithmetic, but I do know that if the demand for something is pretty stable, like the graph to the left, predicting what the future will look like doesn't take a genius, and it certainly does not require software to determine. I can forecast the demand for this item on the back of a bar napkin.
And I also know that, when the demand for an item looks like the graph to the right, then no amount of bashing the sales folks, and no amount of money poured into software is going to forecast the next point on the graph with any certainty. That is the problem with the impending snowstorm – it as an anomalous spike on a graph. The demand for this part is inherently unforecastable … as in 'it can't be done with much accuracy so don't waste your time and the company's money trying'.
Both of these items average 50 per week, but the variability in the rates of demand make them two completely different animals. Somewhere along the way, the completely illogical idea came to be accepted as 'fact' that the item with the stable demand rendered itself quite nicely to kanban, or demand pull, while the item with all of the ugly spikes did not and therefore still had to be managed with goofy MRP thinking – forecasting, primarily.
While lean thinking has progressed by leaps and bounds over the last twenty years, it seems to me that, when it comes to inventory management and factory scheduling it may have even regressed. It is that scoundrel Doc Hall's fault, of course. The young whippersnappers involved in lean may not be aware of the fact that lean leadership initially came from the materials management function. The first understanding of the Toyota Production System was all arounf JIT. Kanban and pull was the heart of it all, and materials managers all around the US and Europe were trying to get their arms around it kanban versus forecasting and master scheduling. When Doc and a handful of co-conspirators, unhappy with APICS reluctance to lead the effort to embrace lean, led the revolt that became AME, the focus of lean began to change. Doc and the AME gang understood lean to be broader than just JIT and the focus evolved to more of a factory operations one, and less on inventory and scheduling.
Too bad because a lot of momentum was lost. The fact is that the only way to achieve excellent rates of delivery with the unstable demand items is with pull. And, yes, it takes a mountain of inventory in most cases to do it. The size of the mountain is a clear mathematical function of the degree of variability in demand, lead times and lot sizes. So the only way to deliver on time every time with mimimun inventory is to reduce the variability, lead times and lot sizes. You will never get there with forecasting.
We need to get materials management back in the lean game. The myth of forecasting need to be killed with a stake driven through its crooked heart. Somebody else is going to have to it besides me, however. There is an 80% probability that I will be snowed in and unable to attend.
Elizabeth Taranto says
I’ll admit I skimmed thru this post rather quickly this morning but for the first time, felt compelled to write–how do you address the lack of forecasting when lead times for a vast range of products (electronic components) typically range from 16-32-48 weeks & beyond? That doesn’t include the single-customer parts that manufacturers stop making without warning. Schedule pull-ins to ‘keep the factory busy’ or deal with a pushout on another build add to this problem, especially in a high-mix, low-volume manufacturing environment. Distributors will pipeline product if there’s some idea of usage & quantity ahead of time. I’d be interested in any publications that address this. Thanks BT
Bill Waddell says
Hi Elizabeth,
The post was aimed at specific, discrete part or product forecasting. There is value in aggregate forecasting … essentially saying we project selling 100 of them over the course of the next year. That sort of planning is much more likely to be reasonably accurate, and is a valuable tool for long range capacity planning. It is when we presume to take the 100 and attempt to put them into discrete daily or weekly buckets that we get into trouble.
For parts such as those you describe, often the best approach is to issue a blanket order to the supplier ir distributor for about 2/3 of the projected annual requirement: A PO that promises to buy at least 70 of them in the next twelve months, also specifying a requirement that, in exchange for the commitment the supplier must maintain an inventory (or at least the raw materials necessary to provide much shorter lead times).
If all goes well, the PO requirement will be met in about 8 months, at which time a new blanket PO is issued for 2/3 of the next 12 months projected demand.
This sort of arrangement (1) protects you from downside forecast error of 30% or so, provides assurance to the supplier justifying their investment in inventories necessary to provide acceptable lead times, and (3) assures overall supply chain inventory minimization with optimum customer delivery.
Of course, the long term solution is to root out and eliminate the drivers of the extremely long lead times you describe.
I don’t know of a source in which these concepts are explained very well. Feel free to shoot me an email if I didn’t answer the question very well.
Sam says
While I agree with some of the evils of forecasting in manufacturing, I would say the National Weather Service was spot on this time around.
Weather forecasting is getting to be very reliable and they deserve a lot of credit.
Brian Dreckshage says
Bill:
A number of us have been working with APICS in order to update their lean offerings. A couple years ago, a number of us revamped their Lean Essentials Workshop and in another month Bill Kerber and I are releasing a book aimed at the typical APICS member. It is called Lean Supply Chain Management Essentials: A Framework for Materials Managers.
I was a member of AME for a while and I believe that we discussed some of their shortcomings. AME was founded on a great idea but they failed to deliver on the education side. APICS has a great “delivery process and network” but has lacked in their lean offerings.
As a long time APICS member, I feel that I owe a lot of my career success to APICS, directly and indirectly and I hope that the Lean community has not given up on them. Bill Kerber felt the same way and I was able to drag him back into the APICS fold after his successful journey into lean. (Unfortunately, we lost Bill to cancer late last year.)
You do not have to publish this…it is more for your information. Keep up the great work!
Bill Waddell says
Glad to hear that someone is trying to blow some relevant life back into APICS, Brian. It would be a great thing for manufacturing to have them back in the game.
N.H.Kalyanakrishnan says
I like this para
“So the only way to deliver on time every time with mimimun inventory is to reduce the variability, lead times and lot sizes. You will never get there with forecasting.” Forecasting is required only one time when the product is launched first time to estimate the initial inventory required to service this product. Subsequently, the inventory should be dynamic based on the actual consumption and the replenishment time. Production should be replenishment to consumption and should not be based on forecast. This is what is suggested by TOC (Theory of Constraints). However people still find it difficult to come out of the forecasting paradigm !!