Sales Forecasting Model
Research by Brian T. Ratchford
THE ABILITY TO RECONCILE SALES FORECASTS WITH SALES PERFORMANCE IS CRITICAL. WHEN IT COMES TO FORECASTING SALES VOLUME/DEMAND AND MAKING SOUND MARKETING DECISIONS, A MULTI-CHANNEL, MULTI-REGION SALES FORECASTING MODEL AND DECISION SUPPORT SYSTEM GO A LONG WAY TOWARD MEETING THOSE NEEDS AND CHALLENGES.
Recent research by Brian Ratchford, PepsiCo Chair in Consumer Research at the Smith School, suggests that sales volume can be forecast by applying established marketing science methods to solve managerial problems. Consumer packaged good companies like PepsiCo and Kraft Foods are increasingly faced with the complex and difficult task of forecasting sales volume and demand for goods sold through multiple channels such as grocery, drug, mass merchandise, and convenience stores in multiple regions. For this reason, it is important for companies to develop separate sales forecasts by product category, channel, region, and major customer account within each channel.
Forecasting has been less than successful in the past because multiple forecasts generated by different users within companies—such as the sales, finance or brand management departments—each based on different methods, were not reliable enough. Incomplete data, unavailable data and the need to capture the effects of past sales, trends, pricing, and promotional and seasonal variables combined to make multi-channel, multi-region sales forecasting an especially arduous task.
In his paper, "A Multi-Channel, Multi-Region Sales Forecasting Model and Decision Support System for Consumer Packaged Goods," co-authored with Venkatesh Shankar, marketing professor at Texas A&M, and Citigroup Senior Vice President Suresh Divakar, Ratchford discusses the development and implementation of CHAN4CAST, a sales forecasting model and a Web-based decision support system (DSS) for carbonated and non-carbonated soft drinks for a leading consumer packaged goods company. Using a dataset drawn from IRI Infoscan data, the company’s wholesaler shipment data, A.C. Nielsen’s Scantrack data, and Wal-Mart’s Retail Link spanning 149 weeks, the authors developed a forecasting model using the best available econometric procedures. Because they needed to consider a large number of variables and the tight timeline, stepwise selection was employed as a preliminary step to develop the initial models. These preliminary models were further refined and the final model was validated against alternative models, using holdout samples. The procedure includes a method for forecasting future values of variables that help in predicting sales, such as price and display activity.
To develop the DSS, Ratchford first
identified the users, what forecasting
information they needed to make
decisions, and what contents and format
of the outputs each user wanted. An
information technology consultant helped
map the model forecast outputs to the
desired outputs and the drill-downs of
the users and developed specifications
for the Web-based tool. The model
integrates all existing forecasting
approaches into one system with field
input, has a scientific benchmark for
determining forecasts, and offers
diagnostics when the actual volumes
deviate from the planned volumes,
placing accountability on the
appropriate managers to meet sales
targets.
| "Forecasting has been less than successful in the past because multiple forecasts generated by different users within the company, each based on different methods, were not reliable enough." |
"The model is being successfully used by a leading consumer marketing company for its major annual forecasts," says Ratchford. "The company estimates that the use of the model and DSS has saved $11 million for an investment cost of less than $1 million."
The model captures the effects of
non-traditional variables such as
temperature and quality of day effects
to improve forecasts and incorporate
several intricate adjustments to the
forecasts, for example, day-of-week
lifts for the cusp weeks, load-ins that
occur before special holidays (e.g.,
Fourth of July) as well as trading-day
adjustments that account for differences
in sales between weekdays and weekends
in a month. Key to the company’s needs,
the authors’ model includes an
appropriately derived quantitative
relationship between weekly retail sales
and
wholesale shipments. Taken together, all
these distinctive features of CHAN4CAST
have enabled the company’s top- and
mid-level executives in sales,
marketing, strategic planning, and
finance to develop accurate forecasts of
sales volume, plan prices and
promotional activities over a long-term
horizon, track sales response to
marketing actions over time and simulate
forecast scenarios based on possible
marketing decisions and other variables.
Divakar, Ratchford and Shankar’s work has been accepted for publication in the upcoming issue of Marketing Science.
To learn more about this research, contact bratchfo@rhsmith.umd.edu.
IN THIS ISSUE
--------------------------------------------------------------------------------
Research@Smith Fall 2005, Volume
6, Number 3