SELF-HEALING CUSTOMER DATA QUALITY ISSUES THROUGH INTERPRETATION OF UNSTRUCTURED DATA: TEXT MINING AND MACHINE LEARNING APPLIED (PROJECT IRENE)

摘要

In traditional customer data management, it is rare that unstructured data, such as thosernstored in notes, is automatically and systematically interpreted to drive proactive data cleansing.rnAnd yet, it is a fact that sellers forming teams to manage their accounts in CRM keep exchangingrnverbatim through notes during the entire lifecycle of a customer relationship. The notes aim forrninstance at updating the players around the detection of new leads but also to warn about a keyrncontact changing job or leaving the company or about an upcoming modification of companyrncoordinates, in fact various comments that, if mined in an unsupervised way through machinernlearning, would allow an Enterprise data management group to operationalize a service of selfhealingrndata quality, thereby giving both time back and up to date structured data to sellers. Thisrnpractice oriented paper will discuss the results and method of a proof of concept developed andrntested for mining Microsoft CRM notes, in its first phase looking after the cluster of contacts leavingrntheir company. This paper will detail the business context and the technical realization of the projectrnand will demonstrate why machine learning is instrumental to go from prototype tornoperationalization. This paper will also illustrate the value of the process developed, for the businessrnon the one hand for whom data quality becomes ambient, and for the data management servicesrngroup on the other hand for whom this automated approach allows more relevancy andrnproactiveness in the data correction.