How do Models Work

Sales Forecasting Models represent the much desired but seldom attained Silver Bullet for most large and small retail chains. If a retailer knows ahead of time approximately how well a proposed new store will perform in a location it makes the decision process about opening that store much easier because the biggest problem, risk of failing, is largely resolved.

The chart below shows a logical approach to sales prediction based on understanding the potential of the location to attract customers from various sources such as home, work or shopping. Logic-based modeling simulates the decision process of real estate executives who analyze a location based on experience and information about the location's potential to attract customers.

Statistical forecasting models take information on the demographics of the trade area, daytime activity, retail synergy and other local factors such as visibility or accessibility and combine these into an equation to predict sales. Although widely used, traditional statistical approaches such as correlations or regression analysis tend to produce very weak, unreliable models. The forecasting landscape for retailers is littered with hundreds of these models that have resulted in billions of dollars in mistakes and many unhappy executives and investors. It takes a rare combination of an experienced real estate analyst and a statistician with expertise in noisy, real world forecasting to use statistical methods properly. Analog models are often simple statistical models.

Spatially-oriented models (often described as GIS Models or Gravity Models) often work with a combination of demographics, to describe the potential value of a household, and the position of the household relative to the store's location. The closer a household is to the location the more likely the pull of the store's gravity to attract people living there and bring them into the store. Many models combine some aspects of both the statistical approaches and spatial approaches.

Since most models are developed by back-predicting sales for current stores then using the equations developed to predict forward they are prone to statistical bias. This is further complicated by the fact that approximately 50% of store performance is due to non-location factors such as management, marketing and market presence. With a couple of hundred vendors, ranging from quality organizations to academics, in the market place and an average price tag for models in $200,000 range finding a good company to do sales predictions models and using a model for site selection is not for the faint of heart!

How Logic-Based Sales Forecasting Models Combine Intelligence on Customer Origins to Predict Sales

How Logic-Based Sales Forecasting Models Combine Intelligence on Customer Origins to Predict Sales