“It is said that the present is pregnant with the future” – Voltaire
Forecasting, therefore, is an attempt to deduce the future from the present. It is both, art and science. We will explore the practice of forecasting demand in the short to medium term. Within the constraints of economic and technology trends, demand forecasting drives the planning process in most businesses.
What is Demand Forecasting?
In simple terms, demand forecasting is the process of estimating the number of products that customers will be willing to purchase in the future.
Understanding the Art and Science of Demand Forecasting
“Science and Art” is used to describe disparate endeavors, often with different connotations. In this blog, “science” has to do with forecasting techniques, the “art” is in their application. In business, the practitioner must often pick particular techniques, without having enough information to determine its suitability.
Demand Forecasting Techniques: Combining Quantitative and Qualitative Techniques
There are qualitative techniques and quantitative techniques used in forecasting demand. The former leverages the experience of domain experts; the latter relies on the processing power of modern computers. The best results are often obtained by combining the two.
Domain experts can account for the total of all factors influencing demand. They can also forecast with a limited amount of data. However, domain experts have a limited span. One cannot expect an expert to forecast demand for hundreds of SKUs with any degree of accuracy. A practical approach is to provide the expert with a base forecast and let them pick which part to override. Aggregating the data to the appropriate level helps. For instance, an account manager might have a better estimate of the total revenue from an account at each SKU level. Similarly, a product manager may have a fair idea of sales of a product family but may not be very good at forecasting shipments per SKU location.
[You May Also Like: B2B Demand Sensing: 7 Things You Must Know ]Quantitative techniques assume that the unknown variable (variable to forecast) is related to another variable(s) whose values are known. They attempt to uncover the relation based on historical data and use it to project the value of the unknown variable in the future based on the value of the known variable.
When the known variable is the same as the unknown variable (the known part being historical values and the unknown part being future values), the techniques rely on the internal correlation of values within a time series. They are known as time series methods.
Time series methods look for the following in the historical data:
- Level: All-time series methods care about the historical value of the variable. The moving average method proposes that the average value of the variable over the recent past is a good estimate of its value in the next period. It is often enhanced by applying a set of unequal weights to values from the past.
- Trend: Most time series methods also look for an increasing or decreasing trend in the historical data and project it into the future. Double exponential smoothing is a popular method that detects a trend.
- Seasonality: Cyclical patterns that repeat over time comprise seasonality. Holt – Winters models and Box – Jenkins models capture level, trends, and seasonality. We discuss some of these methods in our What-if Wednesday series. For data with multiple patterns, there are decomposition methods such as Fourier
When the known variable is another quantity, for which we have a better forecast, and which has a causal relationship with the unknown variable, the techniques are called regression analysis methods. It is important to note that “correlation proves causation” is a logical fallacy, also called “cum hoc ergo propter hoc,” Latin for “with this, therefore because of this.” Causation must be established before regression analysis methods can be used.
A set of interdependent regression equations that, together, describe sectors of the economy are termed an econometric model. They are better at identifying turning points.
The art of forecasting is, unfortunately, not amenable to a similarly succinct taxonomy. We will continue this conversation in a series of blogs highlighting aspects of the art exposed in our work with data sets from various industries.
Enjoyed this post? Subscribe or follow Arkieva on Linkedin, Twitter, and Facebook for blog updates.
