Jun 29 2018
Costs are the object of heated arguments when working to improve manufacturing operations, with participants assuming that the notion of cost itself is self-evident while, in fact, they understand it differently and talk past each other. This is an attempt to bring some clarity to the subject.
In a recent blog post, Bob Emiliani discusses the Money vs. Process View of Cost, which I believe reveals the need for a serious discussion of what we mean by “cost.”
The accounting literature gives many different ways of calculating costs, with different results, without ever explaining what the numbers are supposed to mean. Costs can be sunk, marginal, fixed, variable, applied to goods sold or goods manufactured,… What about just “cost,” without any qualifier?
The only common feature of all the costs in accounting is that they are amounts of money. On the other hand, the TPS literature, including what the quote in Bob Emiliani’s article, discusses costs without ever mentioning money.
On Shark Tank, the judges always ask the contestants who make physical goods about their costs per piece, as if it were a physical measurement like weight or length:
At the same time, we all know of manufacturers going down a death spiral by stopping the production of items with a cost per piece exceeding their selling price, thereby increasing the overhead allocation of the other products and causing more of them to look unprofitable.
Bob Emiliani responded to my comments as follows:
Michel, I agree; the notion of cost is not self-evident. I recall a kaizen where Nakao-san was getting questioned by an accountant about the process improvements being made and their impact on cost. Nakao-san asked the accountant, “What is a cost?” The accountant was puzzled and stumbled for a while, not expecting such a question, and then replied “labor, material, and overhead.” Mr. Nakao replied, “Is that so?” In my conversations with Nakao-san, I understood him to have a very broad view of cost spanning the easily measurable and the immeasurable and non-quantifiable.”
Sid Joynson chimed in with a quote from Taiichi Ohno:
When discussing costs, we must always remember Taiichi Ohno’s warning. “Costs do not exist to be calculated, they exist to be reduced.”
Accountants’ Costs and Calculations
The accountant’s answer is typical, in that it is all about procedure and says nothing about meaning. If asked “What is length?”, the equivalent answer would have been to say “what you read on a tape measure.”
We think of the length of a rod as an intrinsic property of the rod, that exists independently of our ability to estimate it with an instrument. If we measure it with different instruments, we get slightly different numbers, but we think of the rod as having a true length around which the measurements fluctuate.
When we consider a manufacturing workpiece, we have multiple formulas to assign it a cost but our accounting manuals fail to give us any notion of the “true” cost these formulas estimate.
Perhaps, the reason for the absence of such a discussion is the close relationship between bookkeeping and taxation. A number on a tax form or a legal filing is whatever a formula produces. Its job is to participate in a tax calculation, not to represent an understandable economic fact.
For example, the Cost-Of-Goods-Sold (COGS) has to be calculated using standard depreciation schedules for equipment. On the other hand, when you have machines with overlapping capabilities and must decide which ones to use for which products, depreciation is irrelevant to your cost calculations because it will remain the same no matter what you do.
Cost in Manufacturing
In everyday life, the cost of a carton of milk is the $2.99 you have to pay for it, and that’s the starting point. If you choose not to buy it, you keep the money and it doesn’t increase what you pay for anything else.
We know it’s not true of the accountants’ “unit costs” for manufactured goods. If you decide not to make a unit, not only do you not get to keep its cost but you increase the costs of all other units you make because each one has to carry more of the factory’s overhead.
Yet many managers still behave as if the unit costs were like the cost of a carton of milk, as reflected in comments like “when we ship one of these, we ship a dollar with it.” Unit cost thinking is wrong but simple. If you have a black box that puts out unit costs, it is easy to use them to choose whether to make a product or which machines/lines to use when making it.
All the complexity has been put in the black box where you don’t need to worry about it. This may have been a reasonable thing to attempt in the 19th century but not today, because everybody has the computing power needed to do a serious analysis of the funds flows associated with a decision or event. While Excel is often rightly criticized for being misused, this is a task it actually excels at.
In everyday speech, “cost” is used to designate (1) money you have to pay as in “this carton of milk cost me $3” and (2) any kind of loss as in “this debate cost me the election.”
The management accounting notion of cost is an extension of the first sense to more complex situations than the purchase of an item in a grocery store. It is expressed in the language of money. Ohno and Nakao use it in the second sense, in the language of things. Both uses are legitimate, just not in the same conversation.
The Case Against Calculating Costs
The case against calculating costs is that extending the concept from a carton of milk in a grocery store to a manufactured product in a factory is anything but straightforward. The results take so much energy to produce that the organization has none left to work on improvement. One consequence is that it makes you focus you on the easily quantifiable at the expense of what is dismissed as intangible.
In Section 4 the 4th edition of Juran’s Quality Control Handbook (1988), Frank Gryna described Quality Cost as the sum of costs for failure, appraisal, repair, and prevention. The effects of less tangible issues like the company’s reputation were excluded for being “controversial” even though, as revealed for example by the Firestone tread separation case in 2000, they dwarf the easily quantifiable terms. 30 years later, the Quality Control Handbook is in its 7th edition. Frank Gryna’s Section 4 is no longer there, but the “Cost of Poor Quality” (COPQ) discussed in many sections is a close relative to Gryna’s and still excludes terms like customer ill-will.
The Case For Calculating Costs
The case for calculating costs is that, unless you do, you may spend your resources on “improvements” that make no difference.
For example, if you have been implementing 5S on the shop floor, you might want to apply it to the overcrowded hard disk of your server. This would mean reviewing thousands of files to decide whether they belong on the server or should be archived elsewhere. Then you might want to reorganize the folder structure and your databases and set up procedures to maintain the new organization… Alternatively, it may be faster and cheaper to just buy a bigger drive, and you need to know whether it’s the case.
Cost As A Schedule, Not A Number
The concept of unit cost, while intuitive, is a mismatch to manufacturing operations. You don’t make and sell one unit but a flow of units over time and this is what you should be looking at. If you increase production from 20,000 to 50,000 units/month for the next 4 years, you may need new machines and more people, materials, outsourced services,…, and you expect an increase in sales. In financial terms, it doesn’t translate to one number but to a schedule of numbers over time, and it is this schedule that you use to evaluate the economics of decisions.
According to Wikipedia, thinking in terms of cash flow schedules is not new. In fact, it goes back to the advent of money lending for interest in ancient Egypt and Babylon. Today, it is taught to anyone who takes a class in financial analysis, including the use of electronic spreadsheets to run all the calculations that made this technique cumbersome 50 years ago.
Economist Claude Riveline argues that goods don’t have costs, only decisions and events do. In his view, the cost of a decision or an event is not a single number but a schedule of outflows, representing what it causes you to spend over time that you wouldn’t have if you had made another decision or the event hadn’t happened.
The calculation requires an explicit alternative. By default, the alternative is nothing — you don’t expand production or the earthquake doesn’t occur. But the alternative can also be a different means of achieving the same result, for example outsourcing the production increase instead of expanding production in-house.
Generating schedules for both outflows and inflows is straightforward. Where it becomes complex is when you apply a discount rate to the numbers for the future in order to make them comparable to the present and calculate various measures of performance. This is why it’s called discounted cash flow analysis (DCF) and it yields different results depending on the choice of the discount rate.
One commonly used method is to find the discount rate that equalizes the net present values of the schedules of in- and outflows. It’s called the Internal Rate of Return and is commonly used in evaluating projects. It is calculated by the IRR function in Excel.
In evaluating costs, Riveline, on the other hand, does not try to reduce the schedule to a single number. To him, the cost of the decision or event is not a single number but the schedule itself. It is a sequence of numbers that is free of arbitrary allocations over time. You can visualize it in a table, bar chart or line chart, or summarize in a variety of ways. And, unlike unit costs, the schedule is easy for manufacturing professionals to generate and understand.
Allocating Shared Resources
While free of allocations over time, the numbers in Riveline’s schedules are still subject to allocations for shared resources. If you can break down your plant into focused factories, each with all the resources needed to make one product, you have eliminated the problem.
In most factories, however, complete dedication of machines to products cannot be done. Some of the support services usually stay centralized, including HR, Shipping and Receiving, IT, Technical Data Management, Safety, and others. In production, some processes still operate as common services, like Heat Treatment, Electroplating, or Painting.
Among the products, only the Runners get dedicated lines; Repeaters, dedicated lines by product family; Strangers, a shared job-shop. As a consequence, the problem of determining how much of a shared resource is consumed by each product does not go away.
In the accountant’s answer to Nakao in Bob Emiliani’s story that cost is “materials, labor, and overhead,” the most problematic term in this sum is overhead, which is an allocation to each unit of product of a lump sum including everything that is neither labor nor materials. It is also the money that would have to be spent even if the factory produced nothing.
In a mid-19th-century textile mill, overhead was tiny compared to labor and materials, and the way you allocated it didn’t matter enough to influence decisions. In a 21st-century semiconductor plant, it’s a different story. What goes into overhead and its allocation are vital questions because the plant spends ten times more on overhead than on direct labor and materials combined.
A practice from the late 19th century that is still common is to allocate overhead to products based on direct labor. But overhead includes activities like Technical Data Management, for which the resources consumed by a product have nothing to do with its direct labor content but is driven instead by the flow of Engineering Changes.
A new product may undergo 50 engineering changes in its first year; a mature product with the same direct labor content, 1 or 2. As a consequence, the new product consumes 25 to 50 times more of the Technical Data Management resources than the mature one but is charged the same amount for this component of overhead. The managers in charge of the mature products may argue that it makes the cost of new products appear unfairly low. This is but one example of the way allocations can be gamed.
Harvard’s Robert Kaplan has proposed Activity-Based Costing (ABC), as an approach in which each component of overhead is allocated to products based on its own cost-driver. For Technical Data Management, it would be the number of engineering changes, and other cost drivers need to be defined and measured for Production Control, Maintenance, Quality, Human Resources, etc.
ABC requires considerably more data collection and analysis. In Real Numbers, Orry Fiume and Jean Cunningham also point out the risks of using cost drivers like the number of setups on a machine that may encourage line managers to do fewer setups and increase the size of production runs.
While it is complex to apply, it is still a simplistic assumption that every activity can be assigned a single driver as a measure of the amount required by a product. Imagine, for example, an electroplating facility for workpieces the size of bedsheets, in which you load the racks with parts the size of teaspoons in the gaps between the bedsheets. Once the racks are loaded, the direct labor to run the electroplating tanks is not affected by the presence of the teaspoons, and the extra amount of chemicals needed is imperceptibly small.
How should you split the cost of electroplating between the bedsheets and the teaspoons? The teaspoons manager thinks that piggybacking these parts on the bedsheet operation was a brilliant idea that saved the company the cost of a benchtop electroplating machine. Since it adds no expense to the factory’s operations, this manager thinks it should be free.
The bedsheet manager disagrees and doesn’t want to let the teaspoons manager freeload on the racks. From this manager’s perspective, the teaspoons department should be charged the same as if it had its own machine, and that amount should be deducted from the bedsheets’ plating costs.
There is no obvious cost driver that could be used here. In Game Theory: A Nontechnical Introduction , p. 209, Morton Davis addresses this question in the case of an airport runway shared by four classes of aircraft. A jumbo jet requires a longer runway than a regional jet but regional jets can use the long runway.
In the simple case of the bedsheets and the teaspoons, his solution boils down to splitting the difference and charging the teaspoons department half of what it would have spent to buy and operate its own, dedicated equipment.
If you want to make economically sound decisions based on unit costs, you have to deconstruct them, undo the allocations, and reaggregate the numbers into funds flows associated with the flows of materials you want to make decisions about.
Given that unit costs are misleading, as an organization, you might as well bypass their calculation altogether. Unit prices, on the other hand, are real. The average selling price multiplied by volume gives you a revenue. Your analysis should match the schedule of revenue inflows against the schedule of outflows needed to produce the goods and, for that, you don’t need unit costs.
As we have seen, the problem of allocating the costs of a shared resource among products does not go away, and it doesn’t have a one-size-fits-all solution. Allocating all overhead based on direct labor is simple but sufficiently off to lead to wrong decisions in 21st-century factories; allocating for each activity separately based on its own cost driver makes sense for some but not all activities.
The local line managers need to work with the accountants and come with a meaningful approach for each common service and centralized support department, as long as they exist.
And you need to keep in mind that, just because you don’t know how to quantify an activity or an outcome doesn’t mean it’s not vital to the business.