Show simple item record

dc.contributor.authorMishra, Arul
dc.date.accessioned2016-01-01T05:54:39Z
dc.date.available2016-01-01T05:54:39Z
dc.date.copyright2015-12-11
dc.date.issued2015-12-11
dc.identifier.urihttp://hdl.handle.net/11718/17155
dc.descriptionThe R & P seminar held at Wing 11 Committee Room, IIM Ahmedabad on December 11, 2015 by Prof. Arul Mishra, University of Utah on "Identifying the most and least promising customers through similarity kernels".en_US
dc.description.abstractResearch has demonstrated that identifying profitable customers and acquiring them is far more expensive than retaining existing customers. The higher cost emerges because firms are unable to identify the truly profitable customers or those that would convert to their brand. In their inability to accurately identify leads they spend resources on many leads that appear potential but do not convert resulting in a lot of waste. One way to identify the small number of promising customers is to recognize that promising customers appear like anomalies. However, traditional statistical methods cannot identify anomalous observations combining both numeric and categorical variables in a dataset. This task becomes complex in this age of BigData that contain variables that can’t be assumed to follow any statistical distribution, are at times sparse and are generated via a dynamic process where it is hard to fit a stable predictive model a priori. We present a method of detecting anomalous yet promising customers using similarity kernels that can handle mixed attribute data. We test the performance of six kernel based algorithms to detect anomalies using both simulated and real marketplace data. Our proposed method holds implications for research on word-of-mouth (WOM) to find out consumers who are most likely to diffuse a message versus those who are least likely to. In customer churn it helps companies to correctly identify customers who are more likely to churn so that they can invest resources on them and avoid wasteful spending on those who are less likely to churn. In sales data it helps identify heavy users of a product generating high sales versus non-users on whom offers have no influence. In market segmentation research the presence of such extreme observations can lead to consumers being classified into wrong clusters. Correctly identifying the most valuable, from the least valuable, customers can result in targeted sales promotion offers, accurate online advertisement delivery and even better direct marketing.en_US
dc.language.isoenen_US
dc.publisherIndian Institute of Management, Ahmedabaden_US
dc.subjectsimilarity kernelsen_US
dc.subjectleast promisingen_US
dc.subjectmost promisingen_US
dc.subjectBig dataen_US
dc.titleIdentifying the most and least promising customers through similarity kernelsen_US
dc.typeVideoen_US


Files in this item

Thumbnail

This item appears in the following Collection(s)

Show simple item record