Ask Gemba: Volume and Variability in Demand Segmentation

Joe asked an interesting question:

I am doing some research on demand segmentation and found some rather conflicting information from two credible resources. According to The Toyota Way Fieldbook when leveling production you produce High Volume products to stock while you make the Low Volume products to order. Then according to Ken Koenemann , Practice Leader, Lean Value Chain Practice for TBM Consulting Group, one would make Low Order Variability/High Volume and low variability/low volume products to demand while high variability/low volume and high variability/ high volume items are make to stock. Am I missing something because of the fact that the Toyota way does not take into account frequency between orders and only looks at volume? I think this would be a great post topic as smoothing is the base of the Toyota Production System but yet it there is some rather conflicting information out there.

Well Joe, it’s possible you’re missing something but I can’t say for sure without seeing the full context of the TBM explanation. You’ve caught me in the field without my Fieldbook, so I can’t look it up to provide the context. I ask our readers to help out with this question also.

Both sources know what they are talking about so if there is a contradiction it’s likely that they are not talking about exactly the same thing. Before we get into it, let’s clarify that you mean “make to order” where you’ve used “make to demand” (you use both). So the two options are make to order and make to stock. Also, for the sake of this discussion “leveling” and “smoothing” will both be synonymous with “heijunka”, which we will define as “the averaging of both volume and mix over a specified time window”.

Leveling your demand can be for either make to order production or make to stock production, or a combination of the two. It is simply a question of where you hold inventory to set the trigger point. Many think lean means “no finished goods” but this is not so, especially when having enough of the right demand at or near finished goods can help level the demand and reduce total cost down the supply chain by reducing the bullwhip effect of demand variability that amplifies as changes travel down the supply chain.

The Toyota Way Fieldbook is specifically discussing production leveling or heijunka. In that example it is correct to say that you perform an ABC analysis and focus the leveling of demand on the items that are repetitive and higher volume. Since all of this low-low high-high talk gets abstract rather quickly, here is a 2 by 2 chart to visualize it.

The question of whether you can make to demand or to stock starts with the relationship of the order fulfillment lead time as compared to the required lead time. Regardless of whether it is low-low, low-high, high-low or high-high, if you have capacity and can fill an order in less time than the required customer lead time, you should make it to demand. But this is where you need to do the ABC analysis or similar segmentation to see just how much of each type you have. Of course this assumes that both your processes and customer orders are both stable and not variable. Both of these are likely true or true enough to be valid working assumptions at Toyota, which could explain the Fieldbook explanation above.

Leveling requires a lot of work to stabilize processes, shorten changeover times and reduce lot sizes. However heijunka is not strictly speaking necessary for successful demand segmentation. If you had enough capacity and scalability you could chase demand. If the cost of inventory was sufficiently low, or if you faced severe seasonality issues you may not want to level load at all. So one possible source of this contradiction is that the Toyota Way Fieldbook is talking about heijunka and leveling, not the broader topic of demand segmentation.

The context in the TBM example seems to be more of a discussion of make to order versus make to stock, or true demand segmentation rather than heijunka. Again, these are actually two related but different discussions. Hopefully this clarifies the apparent confusion.

Our friends over at the Lean Six Sigma Supply Chain blog have written some good, brief, visual examples of demand segmentation. The point by the LSSSC folks is also that variability of that demand should be factored when deciding whether to make to stock to order. Repetitive demand that is highly variable may not be appropriate for a make to stock model, while a low volume, highly predictable and reliable demand item may be good for make to stock. It depends on many factors of the business model such as the margins of the product, the comparative cost of lost sale versus total cost of ownership of the inventory, the relative speed of the fulfillment lead time versus the required lead time by the customer, and also your current capacity utilization.

The curves below are typical, but this can get really skewed if your product margins are very high, making it very attractive to optimize around limiting the cost of lost sales rather than the sum of various operational costs. This should also be taken into account when doing design segmentation to design the fulfillment model.

Heijunka, in other words, is not for the light of heart. Demand segmentation requires you to take a broader look at your business model and understand the levers that drive it so that all of the parameters are properly considered.

6 Comments

  1. Fernado Flora

    August 2, 2009 - 1:23 pm

    “Toyota Way is practice not theory” I like more and more this sentence!
    I have more than 15 years in automotive industry. In the last 7 years in production and a lean manufacturing specialist in a big multinational. I gave lean training for all levels of management and private costumers. I sow one thousand different problems and unfortunately, TPS don’t give us the answers that we can imagine
    Now as operational manager I have a new challenge… Working for one of the most important brand in automotive industry (is German). What is the challenge?
    per day I should sew about 350 covers. I have 1300 different part numbers, 5 complete different models and many different colours (threads). In a normal production day maybe I can have about 100 different part numbers. It means, 1,2 covers for the same part number.
    The costumer, change every day the weekly planning, some times twice, always some new emergency and so on…
    So, we start analyzing cycle times vs. tack times, doing a combined cycle time (according historic), doing the best lay-out its possible (at least the best flow for the highest model) and trying and trying different combinations of lay-outs, improving our CONWIP for one point start, and this and this and more this….
    I have no more than 75/85 % of efficiency, no stock (for this I need at least more 25% of labour (I work in one shift) and a big pressure… So, high variety, low volumes… for each problem, different solution. Its true that the Lean thinking help… a lot. But in the end, there are not easy solutions or graphics that can give us the way for paradise. Only the way of thinking can help us. What can we do more??? And what can we do Different today? This is the right questions but attention; don’t bring us always the right answers. Just try and try and try.
    Fernado Flora

  2. Bruce Baker

    August 3, 2009 - 7:18 am

    Does it make sense sense to look at variability ‘scaled’ to demand? Something like “stddev of demand / avg of demand” in order to differentiate between ABCs?
    I did ABC segmentation for a previous employer and we learned (in general) that leveling lowered variability, and that looking at TRUE demand was important. We found that if we used our historic shipment data that we were looking at a model of our own dysfunctional systems instead of TRUE demand. The shipment data showed our customers’ created demand because of our artificially long stated lead times (so that we would never be late) and our inability to hit even long lead times as well as our customers’ internal problems as far as variability of demand. We also learned that many B and C demand FGs shared precursor materials with the As at some upstream point therefore we could often combine the upstream demand from As, Bs, and Cs in supermarkets in upstream pull loops and reduce the lead times for make to order stuff (B & C FG skus). We also learned that you could find lower variability in RMs due to FGs using common RMs but that they used WIPs that were common as well so we could create shorter lead times for B & C FGs by stocking them in a partially processed state. Also it is good to try to include you vendors as much as possible because you might find similar opportunities there as well. We started including number of SKUs in each data box of a current state vsm and we would look for SKU explosions. This was really fun. We came up with big challenges that nobody would have thought of had we not done current and future state vsms and we took a lot of cash ut of inventories.

  3. Larry Loucka

    August 3, 2009 - 9:03 am

    Fernado Flora, To add to Jon Miller’s comments … if you have the customer historical demand by day you can calculate a measure of variability or linearity called the Coefficient of Variation. Take the demand by day for each product and for each calculate the daily average and the daily standard deviation. If you use Excel be sure you have a zero in every blank cell. Excel treats blanks and zeros differently. Now calculate Cv = Standard Deviation / Average. The greater the Cv the more variability in demand.

  4. Mike

    August 4, 2009 - 7:36 am

    I think you are all guilty of making this overly complicated and confusing. Why do you need to have mathematical models that will predict the future for you and/or tell you what to do. Communication is the key–internal and external communication coupled with properly flexible processes and people will allow you to know what to make, when to make it, and how much “safety stock” to keep. Try utilizing some common sense and a couple of whiteboards instead of looking for a Rosetta Stone algorithm.

  5. Joe Molesky

    August 4, 2009 - 11:11 am

    Thank you for all of the feedback. I appreciate all of the thought that has been out into this. Our issue today is that we make nearly everything in a make to order model. All of our product is configured to order. We do not sell any two jobs that are the same. However, all jobs are made out of a list of thousands of possible components so there is some similarity. The problem comes in due to the high degree of seasonality in our business. Capacity is not an issue 9 months out of the year. However, during the three peak months we are hard pressed to get our work done. I am simply looking for a starting point.
    Mike, I do not think anyone was attempting to find a Rosetta Stone Algorithm, instead I am looking for a common understanding to two conflicting pieces of information. John truly helped with that.
    Have a Great Week,
    Joe

  6. Ken Koenemann

    March 21, 2012 - 8:15 am

    Jon,
    Your comments are true, however I believe their is a misunderstanding on the replenishment or planning techniques I have recommended.
    Low variability/High volume products should be placed on either a rate-based or pull replenishment system.
    Low variabliity/Low volume products should be placed on a pull system. The biggest issue you encounter with these items are the minimum lot sizes one produces (if not capable of one piece flow) are larger than demand in the period.
    High variability/High Volume should really be on a Reorder Point model. Due to the variability one would not want to carry a lot of inventory to cover it. Therefore looking at forecasts and using ROP is an appropriate method. If you have unlimited capacity, Make to Order is another method.
    High variability/Low volume should be make to order.
    The big challenge for most organizations is that the Toyota system which I worked in was smoothed at the point of distribution. As a supplier to Toyota you typicall saw less than 10% variation – truck to truck, day to day, week to week. Unfortunately most buinesses do not understand the concept of Heijunka from a network perspective.
    These are all great discussions.