05.01 "Quantifying and Improving Promotion Profitability at CVS"
VIEW VIDEO
by Kusum Ailawadi, Jacques Cesar, Bari Harlam, David Trounce

Like many high-low retailers, CVS (a leading U.S. drug retail chain) makes approximately 30% of its sales on some kind of consumer promotion. However, a large proportion of these promotions are not profitable. In this large-scale study of promotion effectiveness we (1) decompose the immediate sales bump of each promotion offered by CVS during the year 2003 in any of its 4,400 stores into switching, stockpiling, and primary demand components and estimate its net sales and margin impact, and (2) relate this impact to a wide range of promotion, brand, category, and market characteristics to understand how promotion effectiveness varies with these characteristics, so that this analysis could be used as an input to decisions about which brands and categories to promote, how much, and in which markets. To validate our work, we design and implement a controlled field test to assess the sales and margin impact of discontinuing promotions in 15 product categories where our analysis show that promotions consistently have negative margin impact and only a small positive net sales impact. A projection of the results from the 13-week test in 350 stores to a chain wide 52-week period show a net sales loss of $7.8 million, which is less than 0.1% of total revenue, but a net margin gain of $52.6 million, which is highly significant for the company. CVS is now implementing these changes chain wide and investing the margin savings in resetting its everyday prices.

05.02 "An Assortment wide Decision-Support System for Dynamic Pricing and Promotion Planning"
VIEW VIDEO
by Martin Natter, Andreas Mild, Thomas Reutterer, Alfred Taudes

Reporting from a large-scale practical implementation of marketing science models, this paper solves problems in the field of revenue management. The prices of a large subset of articles from a do-it-yourself retailer are automatically optimized by the application of reference price models. A large positive impact on profits is achieved. The models used at a more advanced stage of implementation also lead to a considerable increase in sales. Currently, thousands of separate models for different countries are estimated at the article level every month and are provided to the store managers using a fully automated workflow.

05.03 "The Right Product for the Right Person: Product Recommendation from Infrequent Events"
VIEW VIDEO
by Marty Vrieze, David Freed, Brendan Kitts

We describe five years of work developing, analyzing, and running one-to-one marketing systems in a successful cross-channel retailer. The problem we set out to solve is one of targeting the right customer with the right product. Previously our participating retail chain had sent e-mails to opt-in customers featuring products chosen by the marketing department. In 1999 we ran a pilot study in which we replaced the static products with product offers selected by a probabilistic one-to-one recommendation algorithm we later called “cross-sell.” The pilot test was very successful. Clickthrough increased by 40%, revenue by 38%, and units sold by 61%. The full system was deployed in October 2000. At that time we also tested a family of additional recommendation strategies based around the hypergeometric distribution test. These tests generated a response rate of over 100%. From 2000–2005, the recommendation system has continued to be used each week, and the retailer continues to maintain control groups and report on performance. Incremental profit from the system has been substantial. The retailer even displayed the recommendations on their monthly catalogs. The hypergeometric recommendation method is noteworthy because it is a well-defined significance test (compared to almost all other recommendation strategies), and performs well on small numbers of events. Infrequent customer purchases are common in a great number of merchandising domains, including books, music, electronics, home improvement, and video games. A reliable solution to this problem is therefore of great interest to marketers.

WINNER: 05.04 "Customer Equity and Lifetime Management (CELM)"
VIEW VIDEO
by Abdel Labbi, Cesar Berrospi, Andre Elisseff, Giuliano Tirenni

The Customer Equity & Lifetime Management (CELM) solution is a decision support system that offers marketing managers a scientific framework for optimal planning and budgeting of targeted marketing actions (e.g., campaigns) to maximize the return on marketing investments. The CELM technology uses Markov decision processes to model customer dynamics and find the optimal marketing policy maximizing the value generated by a customer during a given time horizon. Markov decision processes are estimated using nonparametric estimation algorithms. Lifetime value optimization is achieved through dynamic programming algorithms that identify the marketing actions causing customers to transition to better value/loyalty states. These marketing actions can have the form of cross-selling, up selling, and loyalty marketing campaigns. Modeling customer dynamics allows us to consider lifetime value as an endogenous variable that both depends on and influences the optimal marketing policy. Heterogeneity across customers and across marketing actions is addressed by the model by tailoring the marketing action to the individual customer. The CELM technology can also be used to simulate the financial impact of a given marketing policy via Monte Carlo simulation; in this way marketing managers are able to simulate several scenarios and plan the monthly budget requirements to finance the chosen marketing policy. Monte Carlo simulation allows us to estimate the lifetime value distributions (i.e., the financial profile) of customers if a given marketing policy is applied. Thus, portfolio optimization techniques can be used to build a portfolio of customers optimizing the value/risk trade-off of the return on customer equity.