Response Management is all about what you do when what you planned to do (forecast) does not match what is happening (actuals). This can be applied to any forecast, whether that is the traditional sales forecast of the number of units sold in a region, the projected cash position of a company, the expected completion of a new factory, or the commercial availability of a new product. For this discussion I will focus on traditional supply chain forecasting because that is the underlying data I have available to support my description.
First let’s examine why companies create a revenue or sales forecast. It is intuitively obvious that they do so to determine how much of their product they believe the market will purchase. But this is only the 10 percent of the iceberg that floats above the water. The other 90 percent of the story is that the forecast is used to drive investment plans in marketing, engineering, manufacturing capacity, inventory planning, and strategic sourcing. In other words, if the forecast is inaccurate there are a whole bunch of consequences beyond just getting the revenue forecast incorrect. Of course here I am referring to the longer term forecast used to drive the S&OP process or even the annual operating plan/budget and the principal forecast is investment.
The medium term forecast is used to drive how an existing infrastructure, particularly the supply chain infrastructure, can be used to meet market demand. This is the period in which investments are not going to make a difference. As one colleague told me, nine woman are not going to be able to produce a baby in one month, no matter how hard they try. But it is important to realize that even within this period the significance of forecasting varies by industry, as is reflected in the diagram to the left from an Aberdeen report titled “Demand Management – Bridging External Market Inputs with Internal Statistical Forecasting” published in June 2011. In this context it is important to understand how the difference between demand lead time and supply lead time varies across industries. Airplane manufacturers typically have a 3-5 year lead time from a firm order to delivery date, with a manufacturing lead time of about six months. A diaper manufacturer has between a one day to one week demand lead time, and about a three week supply lead time. On the other hand, an airplane is highly customized to a specific customer’s order, whereas a diaper is a diaper, even though there are variations. The complexity in a plane is in how bought materials are coordinated to assembly a customer specific product. The complexity in a diaper is in how it is distributed with hundreds, in some cases, thousands, of distribution locations through a multi-tier distribution network.
The combination of the difference between demand and supply lead time and degree of final product configurability determines the extent to which a company can use postponement as a strategy to mitigate the risks associated with forecasting incorrectly. CPG companies, such as diaper manufacturers, typically have little opportunity to postpone at the manufacturing stage, and therefore will use a make-to-stock supply model. However, CPG companies can postpone distribution through their multi-level distribution network. On the other had an airplane manufacturer will seldom order components before they have a firm order, which is a “build-to-order “ postponement strategy. Imagine the financial risk that an airplane manufacturer would be taking on if they built an airplane and then tried to find someone to buy it. In reality a diaper manufacturer is taking on similar risk, but the risk is distributed over millions of diapers and many thousands of consumers, so their risk per item is much smaller.
And yet a semiconductor company takes on the same level of risk when they commit $3B-$5B to build a new factory over the next 24-36 months that an airplane manufacturer would be taking on if they used a “build-to-stock” supply chain model. It is an equivalent risk when a company decides to invest in penetrating a new market and needs to invest in establishing a local legal entity, office rentals, marketing, and hiring and training local staff. These are big commitments of funds that are based upon the anticipated behavior of the market, a forecast, and once in execution they take a lot unwind.
Terra Technology, one of the leading forecasting technology companies focusing on the CPG space, where statistical forecasting is very prevalent, published a study on forecast accuracy titled “2011Forecasting Performance Benchmark Study” in which they study best practice demand forecasting in leading CPG companies. The reason that statistical forecasting is so prevalent in CPG is that demand is relatively stable – when compared with other industries – and products have fairly long life cycles so there is a lot of history to rely on. And yet in the introduction the authors note that:
- Promotional volume jumped about 75 percent in 2010 as companies looked to drive sales by offering consumers additional value. Contrary to conventional wisdom, promotional periods are actually forecast as accurately as non-promotional periods for the same items. Perhaps this is due to the extra time demand planners spend on promotions. Not surprisingly, the bias is considerably higher during promotions.
- New products remain hard to forecast with weekly item/location error rates of 65 percent, compared to 46 percent for existing products
- Demand Sensing continues to provide a consistent step change in forecast accuracy for all scenarios, including promotions and new products. For the combined 2009-2010 period, Demand Sensing reduces average weekly error by 40 percent.
- Outdated mathematics and optimistic marketing departments continue to undermine the performance of Demand Planning. This highlights the opportunity for a structured approach to forecasting based on additional demand signals and new mathematics.
- MAPE is the correct measure for supply chain performance since it is the error that Product Supply contends with. However, insight from the report raises questions regarding MAPE as the proper metric to evaluate the performance of Demand Planning because it may not properly reflect the value add by planners. Using the dataset as a resource, Terra plans to evaluate a number of different metrics in the future editions of the study.
Before analyzing the numbers, let me reiterate that CPG, when compared with industrial equipment manufacturing for example, has stable demand and long product life cycles, which, in theory, means that CPG companies should be able to use statistical forecasting to predict demand fairly accurately, but, as the Terra Technology study shows, they can’t. When we consider the three month digital camera lifecycles and six months cell phone lifecycles, the relevance of the second bullet about the forecast accuracy of new products becomes very apparent. New product launch is typically the time when the most margin is captured, an yet it is also the time when the forecast is most inaccurate, meaning that a lot of margin is not captured. And therefore anything Terra describes in the report is significantly worse in industries with shorter product life cycles, which automatically leads to more volatile demand. In fact our anecdotal evidence from speaking to companies in consumer electronics is that their forecast accuracy is very seldom above 50 percent regardless of the life cycle stage of the product largely because the short life cycle means that the product is either being introduced or being phased out.
In the Aberdeen study referred to above, the author notes that Best in Class performance
- Average percent forecast accuracy at product family level (across a three-month time period) is 87.1 percent
- Average percent forecast accuracy at individual SKU item level (across a three-month time period) is 70.8 percent
In the Terra Technology study the author notes that
- During 2010, the average weekly error was 48 percent with a slight difference by top performers, who came in six points lower at 42 percent.
- Meanwhile, monthly error averaged 33 percent with a five point spread between top performers and the average.
With leading CPG companies struggling to get forecast error below 30-40 percent after years of trying, with product life cycles shrinking every year, and with market differentiation leading to product proliferation, what is the likelihood that forecast accuracy will improve dramatically over the next five years? In my opinion very little. So where do you think your next breakthrough in supply chain performance will come from?
- Learning to forecast and plan better?
- Learning to respond profitably to actual demand, or plan variance?
The third bullet from the Terra Technology report introduction states that “Demand Sensing continues to provide a consistent step change in forecast accuracy for all scenarios, including promotions and new products. For the combined 2009-2010 period, Demand Sensing reduces average weekly error by 40 percent.” hints at the importance of response management, but does not go far enough since it only indicates how to get a better understanding of demand in the short term. Response management is about satisfying that demand in the most profitable manner. Demand sensing is about knowing sooner about demand shifts. Actually I find this standard definition of demand sensing a bit funny and it is a good illustration of how planning is seen from the wrong perspective. The term demand shift implies that the forecast is correct and somehow demand has shifted in time or location, which is of course completely wrong. The customer didn’t buy the ‘wrong’ stuff in the ‘wrong’ amount at the ‘wrong’ time. What actually happened is that we predicted incorrectly what they wanted, how much they wanted, and when they wanted it, even where they wanted it. Nevertheless demand shift captures the concept that actual demand occurs at a time, quantity, price, and/or location that was not expected. While demand sensing is really important, even perhaps the most important aspect of supply chain execution, it only describes part of the response management story. Response management is about knowing sooner about a broad range of supply chain disruptions and acting faster to provide a profitable response.
Stay tuned for part two of “The importance of Response Management” tomorrow!