In order to remain competitive, organizations must institute a competitive pricing strategy. However, leveraging predictive analytics can help to solve the most challenging component of an analytically driven pricing strategy – estimating customer response. In order to estimate customer response a Market Response Model (MRM or price – demand model) should be used. This formula utilizes elasticity, though some people have preconceived notions about elasticity, and often are not clear what the term really means. Elasticity is the percentage change in demand over the percentage change in price. When using Market Response Models, you must specify a reference price to get a numeric value for elasticity. Only the constant elasticity model (below) has the same elasticity at all price points.
To avoid confusion, focusing your attention on the price-demand model, or Market Response Model is essential. Below are three commonly used Market Response Model forms.
The ideal situation is to examine experimentally controlled studies of actual customer purchases. Controlled laboratory experiments and survey responses can be unreliable. Consumer purchasing behavior is a better data source to provide insight into a strategic pricing strategy and customer price response. However, controlled ‘live market’ price experiments are difficult to automate. Organizations also run the risk of loss of goodwill if consumers catch them offering the same product at the same time to the same person at different prices. An alternative is relying on a company’s historical transaction data. Thomas Nagel, University Professor of Philosophy and Law at New York University, frowns on this approach, labeling it “uncontrolled studies of actual purchases.” [Nagel pp. 325], but there is a way to leverage transaction history to find pseudo-random, controlled price experiments.
When using historical transaction data it is essential to build Market Response Models using data mining to mimic a controlled price experiment by defining the conditions of the experiment such as lead-time, alternative offers and seasonality, then picking only observations that fit the criteria. Among the billions of historical transactions a firm may store, there are many millions of observations that are uncovered that meet the criteria of a controlled price experiment. The huge sample sizes provide a lot of power in estimating the Market Response Models. Power, in a statistical sense is the ability to detect a relationship to price, something where smaller live market or laboratory experiments fail. The best part is that this process is easily repeatable, amenable to automation, and in the long run very inexpensive. Running controlled experiments, while repeatable is expensive, not easily automated, and often lacks the power to detect subtle signals of consumer response to price, creating a bottleneck in your pricing strategy.