Pricing and Promotion in an E-tailing Store
Author: Sudipt Roy, Ph.D.
Introduction: Before finding out the actual price in a store, buyers form expectations. This is called price expectation. Retails can influence buyers’ price expectations by changing the store prices as well as by advertising and other communication strategies.
Objective: The objective of this study was to understand the impact of different factors that change the price expectations and to recommend pricing and promotion strategies that help retailers grow the top line.
Key Findings: Analysis uncovers two customer segments with distinct shopping behaviors. The purchase decision of the first segment, about 73% of the total, is greatly influenced by price expectations that they form before they any store. The other segment relatively more price-sensitive one, makes its purchase decisions based on expectations it forms about prices in other competing stores. At constant cost, we show that a promotion scheme with deep discounts (e.g., 5 % for 3 weeks or 15 % price cut in the first week) boosts sales more than the one with shallow discount (e.g., 3 % for 5 weeks).
Methodology: The analysis was carried on using data from an online grocery store. Latent class modeling technique helped uncover the two classes that are otherwise hidden in the data.
Citation: Price Expectation and Purchase Decisions: Evidence From an Online Store Experiment, with Tat Chan and Amar Cheema, Customer Needs and Solutions, 1(2), 117-130, 2014.
For more click here.
Working with Data that is Discontinuous, Censored and Multivariate
Author: Sudipt Roy, Ph.D.
Count data (1, 2, 3…) are “discontinuous” as they do not admit any number between, say, 1 and 2. Often we are constrained to discontinue counting before the process is over. This leads to “censoring” where we know that the actual number is “more than 10” but do not have the actual count. “Multivariate” analysis allows us to look at numbers that are related (No. of cookies, cups of tea and coffee sold in a café).
Objective: Censoring is a real-life challenge. The objective of this work is to draw meaningful insights from that data that is censored. In addition, the analysis identifies how customers behave differently from one another in their engagement with the company.
Key Findings: Higher censoring introduces more bias. Moreover, the model identifies three separate sub-groups of buyers. Two of these are heavy buyers of company’s product but one sub-group engages with a wider set of offerings than the other. The third sub-group is a light user of company’s product.
Methodology: Finite mixture of censored Poisson regressions approach is used to model the count data that is from the publishing industry. The Expectation Maximization Algorithm is used to estimate the model.
Citation: Finite Mixtures of Censored Poisson Regression, with Dimitris Karlis and PurushPapatla, forthcoming in StatisticaNeerlandica.
For more click here.
