Comment on the implementation challenge of pricing decision support systems for retail managers

We are pleased to see continued academic interest in software for Demand Based Management (also known as Consumer Demand Management, or CDM). DemandTec's modeling approach has its roots in the research of several academicians [1] [2], and we continue to collaborate with members of the academic community in order to keep current with research and enhance our technologies. Montgomery's list of CDM providers is comprehensive. Our experience based on discussions with retailers is that the vendor offerings do differ substantially, both in terms of the underlying science and in terms of presentation (user interface, workflow, and embedded analytics). Some companies appear to aggregate data in order to deal with outliers and improve model stability. While this is occasionally necessary, our experience has been that Bayesian methods can successfully provide robust and adaptive modeling for the majority of retailers in a variety of verticals. Montgomery is correct that both the retail and consumer goods manufacturer markets finally seem ready for Pricing Decision Support Systems (PDSS), as evidenced in the 2003 surveys of two popular trade publications. The Retail Information Systems/Gartner 2003 survey indicated that 20% of retailers had already updated their initial price optimization technology, and another 27% planned to upgrade in the 2003-2005 timeframe [3]. And, according to a Consumer Goods Technology trends study [4], 30% of consumer goods companies had price/promotion/profit optimization technology in place in 2003, while another 21% planned on making this investment in 2004. While both surveys likely suffer from a bias in " planned " versus " actual " realization of software implementation and in sampling from their particular reader base, these numbers bode well for the CDM provider space. The list of requirements for a PDSS in Montgomery's article is fairly comprehensive. Perhaps the largest science challenge in developing CDM software is developing accurate forecasts in the face of incomplete information. Incomplete information comes in the form of sparse scan history (low volume products), sparse causals (products with little pricing and/or promotional information), missing observations (due to error, stock-out, or low volume), missing causals (e.g. display information is infrequently tracked by retailers), and a constantly changing product mix. Some additional required components that we would emphasize are: • A platform or mechanism for thoroughly cleansing and validating data, since implementing a PDSS is often the first time that much of a retailer's data is leveraged for analysis • Provision of infrastructure to process large …