One objective of a smarter forecasting process is to identify and evaluate systematically all factors, which are most likely to affect the course of demand for products and services. What can we do for new demand for which there are no historical patterns? The demand forecaster should identify measurable variables for factors affecting the quantity demanded for a market or region of interest, using comparable products or services for which there are historical data

The exhibit below is a predictive visualization of a fictitious GLOBL product discussed in my book, Change & Chance Embraced: Achieving Agility with Smarter Forecasting in the Supply Chain. The product, called iHearBuddy, is a learning app that translates lectures for foreign language students. The plot shows historical data values, a trend/seasonal forecast profile with prediction limits based on the Additive Holt-Winters exponential smoothing model, coded ETS (A,A,A). It clearly shows a dominant seasonal pattern (reflecting consumer habits?) and a less pronounced trend pattern (reflecting consumer demographics?).

In next exhibit below, the three ETS components are depicted in pie- and cone charts to help visualize the relative contribution of the trend and seasonal variation to the total variation in the data. Using an exploratory year by month decomposition method calculated in Chapter 5 in my book (using the “Two-way ANOVA without Replication” tool in the MS Excel Data Analysis add-in), we can get the following insight into the data: iHearBuddy variation is made up of 51% Seasonality, 4% Trend, and 45% Other.

Pie charts and cone charts are alternative ways to depict the total variation in the demand variable in terms of (1) seasonality (51%), (2) trend (4%), and (3) other (45%). Exploratory decomposition display of GLOBL product iHearBuddy was created with Excel Data Analysis add-in using a Two-way ANOVA Without Replication algorithm.

Based on the forecaster’s domain knowledge and the most recent history of product iHearBuddy , the forecaster may be able to advise that if the business can capture 7% of the educational market, it will yield a monthly internal rate of return of 21.5%. The business planners have designed the daily production requirements, equipment, manpower, and facility requirements based on these sales targets. Recent recessionary trends have been a concern, and now the planners would like to backtrack and see what the outlook of the consumer electronics industry is before embarking on the new venture.

The information on the industry would indicate whether this product is still a wise investment. The demand factors and consumption trends that need to be investigated include price, income, demographics, advertising, and regulation.

As an exploratory step, the demand forecaster could make some insightful assumptions for product iHearBuddy. The dominant seasonality can be quantified by consumer habit factors, such as number of holidays and School openings/closings, driving the demand for iHearBuddy. The trend relates to the underlying growth of the student population and can be quantified by the age-cohorts of student consumers, for example.

Uncertainty is a certain factor in characterizing change and chance

The “Other” component (45%) still contains information about everything else not attributable to consumer habits (seasonality) and consumer demographics (trend). This could include promotions, economic cycle, unusual events and random error. In a modeling environment, we first characterize the trend/seasonality with a time series forecasting model with a trend/seasonal forecast profile (e.g. Holt-Winters exponential smoothing or an ARIMA (011) (011)12 “airline” model; these are in the same family of the State Space Forecasting models discussed in Chapter 8 and Chapter 9 of my book.

The model residuals (Actuals minus Fit) can be analyzed and used to quantify the remaining factors with causal (regression) models, for example. This iterative process is like peeling an onion approach to modeling change with chance. The models give insight into the forecast profile (change), while the forecast errors (Actual minus Forecast) from hold-out samples can be used to get insight into the uncertainty factor (chance). In this sense, we can characterize uncertainty as a certain factor.

Factor Impact Matrices

The impact of the dominant drivers of demand can be summarized in a Factor Impact Matrix, displayed below, along with a predictive visualization of the historical and future impact of a driver of demand. To monitor a time history for the impact of drivers on demand on an ongoing basis, we can create Factor Matrices for the Immediate Past, Present and Future. The time periods should be representative of the situation. For example, one could use lead-times for inventory or production. Illustrative Factor Matrices for trend and seasonal drivers of demand for GLOBL product iHearBuddy.

The direction would indicate the impact a factor would have on demand if it were increasing or decreasing. For example, price increases would tend to result in decreases (-) in demand, while increases in holidays and festivals would tend to increase (+) the demand for electronics products. The quantification of the resulting impact is based on informed judgment and domain expertise and signifies the intensity on a numerical scale, say 1 to 5.

A predictive visualization of the impact of a driver of demand.

To summarize the impact of a driver on demand, we create a predictive visualization of the driver or factor by relating the score to the size of a dartboard surface. Use the relationship of Area = π (radius)2 to determine the size of the circles shown in Figure 1.19. For instance, for a score of 5, the circle should appear five times larger than the one with a score of 1. To get the correct visual effect, you can accomplish this as follows: Score 1 = radius 1, Score 2 = radius √ 2. Score 3 = radius √ 3, Score 4 = radius 2 and Score 5 = √ 5. If you have deeper domain knowledge, you can extend the scale to 7 or 11. 

Hans Levenbach, PhD is Executive Director, CPDF Training and Certification Programs. He conducts hands-on Professional Development Workshops on Demand Forecasting for multi-national supply chain companies worldwide. Hans is a Past President, Treasurer and former member of the Board of Directors of the International Institute of Forecasters. He is group manager of the LinkedIn groups (1) Demand Forecaster Training and Certification, Blended Learning, Predictive Visualization, and (2) New Product Forecasting and Innovation Planning, Cognitive Modeling, Predictive Visualization.

I invite you to join these groups and share your thoughts and experiences.