How do the FMCG companies forecast volume for the new product?
There are many factors to consider before being able to forecast volume for a new product with reasonable accuracy, and even then many experienced marketers get it wrong as the variables are many and some of them are simply unknown.
The factors to consider are :
- Product features and quality vs competition
- Brand Strength if it is a new product from an established brand
- Advertising Investment to generate awareness and educate consumers on the features of the new product (the level of media investment needed changes depending on how complex your product offering is)
- Price and Promotional strategy vs competition
- Product test scores vs. competitive products based on consumer research
- Retail or distribution capability or customer (retailer) buy-in of your products
- Timing of launch if product is seasonal
- and more factors which are category specific
Most organizations take one or more of 3 approaches to work out the volume for a new product over a specific time frame.
- Market Share based: brand managers take all the product and marketing features of the new product and estimate a potential market share. This requires experience and analysis of historical data.
- Retail store based: this is a sales based approach where they estimate numbers of stores distributed, the initial sell-in units, the potential unit sales per store per week to arrive at a volume
- Advertising Awareness based : this is a pure above the line based approach which makes assumptions on how many consumers will be made Aware (Effective Reach) of the product, of that how many will try (Trial) the product (this percentage is based on research of previous similar new products) and how many will come back and buy again (Retention)
When seasonal patterns exist without trends, simple smoothing methods work well with deseasonalized
data. Intrinsic forecasting methods are those that use the past behavior of series to predict the future value of that same series. The term intrinsic denotes that the information used to forecast the series is internal or within the series. The terms time series analysis, univariate forecasting methods, and smoothing methods are often used as synonyms for intrinsic
methods (De Lurgio and Bhame, 1991). The moving average forecast uses the average of a defined number of previous periods as the future forecasted demand (Carbonneau et al., 2008). Moving averages model is useful when the patternless
demands exist. Patternless demand defined as one that does not have a trend or seasonality
in it. A patternless series is random with either smooth or erratic variation.
Weighted Moving Averages (WMA)
It is normally true that the immediate past is most relevant in forecasting the immediate future. For this reason, weighted moving averages were used, when the more important period (sales, or return data) were considered. For example, sales amounts are high before Christmas, so in January expected sales amounts are low, and then lower index was used to calculate the January’s forecast. The most important reason for not using moving averages is that exponential smoothing is as accurate as moving averages while at the same time computationally more efficient. When using moving average, it is more difficult to determine the optimal number of periods to include in the average. In contrast, it is somewhat easier to find the optimal number of periods using the exponential smoothing model.
Exponential Smoothing refers to a set of methods of forecasting, several of which are very popular – Brown’s double, Holt’s two-parameter and Winters’ three parameter EXPOS. EXPOS is the most widely used of all forecasting methods. The trend and seasonally adjusted exponential smoothing methods are used in many computerized forecasting systems for production, inventory, distribution, and retail planning (De Lurgio and Bhame, 1991).
Single Exponential Smoothing
Single exponential smoothing is easy to apply. Uses – the most recent forecast, the most recent actual demand, and a smoothing constant (alpha). The smoothing constant determines the weight given to the most recent past observations and therefore, controls the rate of smoothing. It must be greater than or equal to zero and less than or equal to one.
Optimal values of alpha should be used in exponential smoothing. The function of the smoothing constant is to give a relative weight the most recent actual and forecasted values. With high(low) autocorrelations use high(low) alpha. With high(low) smoothness in demand use high (low) alpha. Determining the alpha values is critical to get more accurate forecast values for the data. If you try to forecast in seasonal trends, you have to calculate in the high alpha level (Finch, 2006).
Regression analysis is a general approach to modeling the causal relationships between one variable, such as product demand, and one or more other variables, such as price, industry sales, or other economic variables. Regression analysis distinguishes between the variable that is being predicted, called the dependent variable, and the variables used to predict that dependent variable, called independent variables (Franzese and Kam, 2007).