Guns and algorithms have a lot in common. Both are tools that, in the hands of the right person, can be powerful forces for good. Of course, in the wrong hands, they are often instruments of evil. Following my post Better Off Smart Than Sexy, Gary Cokins implored, "Can you find in your heart some graciousness that analytics and algorithms are nice to have compared to traditional ways, including intuitives and gut-feel guess-timates?"
The answer is, of course I can – but whoever said supply chain management was a choice between gut-feel guess-timates and "frequent re-freshes of a demand forecast"? Neither one will get the job done. I am foursquare in favor of algorithms – I have been on a decades long quest to settle on the optimum one, in fact – just not a forecasting algorithm.
Here's the deal:
Manufacturing's big problem is the gap between cumulative lead times (production lead time + purchasing lead time), and customer lead times. : the customer wants the stuff in two weeks, but it takes you six weeks to get the parts and materials and make it. So you have to order materials today and start planning production for what you need to ship/sell six weeks out even though you only know what you need two weeks out. What to buy and what to make now? That is the question.
Gary – you and I can agree that seat of the pants, gut feel is a dumb way to go about it. You suggest forecasting (and when I say you, please don't take it personally; I mean the ERP, push system thinkers). the problem is this:
Here we have two items, and two radically different demand patterns over the last 26 weeks. The one on the left is what the supply chain folks refer to as having stable demand. A well trained chimpanzee can forecast it on the back of a bar napkin. If only all of our products had such stable demand. The one on the right has – and bear with me while I throw a highly technical APICS supply chain term at you – lumpy demand.
The gut feelers do dumb things like set inventory at 4 weeks average demand. Both of the above items have an average weekly demand of 50, so both items would have an inventory of 200. That would be more then 3 times greater than the item on the left would ever require, but not enough to meet half of the orders that come in for the item on the right. It is this sort of silly planning that causes many companies to be drowning in inventory, yet still having lousy on time delivery.
The problem with forecasting, however, is that, while the item on the left can be adequately forecast by a primate without need of a computer – no computer, no algorithm, no wizardry no matter how clever the PhD's may be working on it, can accurately forecast the item on the right. It is inherently, mathematically un-forecastable. Those "retailers analysts" stuck in their cubicles do indeed have a tiger by the tail if they think forecasting all of those thousands of SKU's is going to maximize profits and balance stock shortages with excess inventory. They are going to fail – it is written in the mathematical stars that they will fail and there is no algorithm anyone can give them to prevent it.
Here is an algorithm I am particularly fond of:
It is a kanban formula, and all those junior buyers need to do is plug it in, then replenish every week – sold 12, buy 12;sold 183, buy 183; … we are back to tasks doable by chimps on bar stools. You have my personal guarantee that if you use it, those buyers will have a >98% availability rate with close to the minimum inventory possible to achieve that >98% on time rate. It leads to setting an inventory of 71 for the item on the left, and 428 for the one on the right.
To summarize the formula for those still suffering through their first cup of morning coffee. it basically says inventory required is a function of lead times and the rate of variability in demand. The only way to have smaller inventories and still maintain a high on time delivery (or availability) rate is to reduce lead times, reduce variability, or both. Surprise! Surprise! Lean is chock full of tools to reduce lead times, and Six Sigma is really all about reducing variability. Apply the tools – reduce lead times and variability – lower the inventory levels while still shipping on time – starting to hone in on JIT and one piece flow.
The kanban-based pull provides clear insight into how to improve. The pull approach, when driven by the right arithmetic, results in no stockouts,a right sized inventory, and facilitates continuous improvement. With forecast push, both stockouts and excesses are inevitable, and forecast error is just forecast error and all you can do is spend more time and more money in the pursuit of the unattainable accurate forecast.
As I am sure Gary can point out, however, he specifically mentioned retailers analysts – most of whom have 8-16 week Chinese lead times and zero customer lead times, and high rates of variability due to changing trends. Their lead times are so long that, but the time they have enough history to calculate a kanban quantity, the item is discontinued. The only way to avoid the stock out / stock excess problem plaguing the retailers is for them to shorten their lead times – but since their accounting systems revolve around purchase price, rather than a holistic view, they will continue to chase cheap labor to get low prices for items that are inherently un-forecastable; yet they will continue to try to forecast them; and they will continue to miss sales while simultaneously discount merchandise to unload it. In short, they are, well, …. screwed (not to out too fine a point on it).
Pull systems with sound statistics driving their sizes are the only way to assure on time delivery with minimal inventories. No amount of diligence or mathematical analysis can change that fact.
(Thank you Gary for your comment and for being a good sport about it. I ook forward to your views.
As long winded as this post is, it is nowhere near comprehensive enough to explain the formula or the kanban/pull approach it is used in. Shoot me an email if anyone has any need for clarification of anything concerning the approach.)