Along with warranty service abuse, dispatch frauds are a huge concern area for major IT companies. A spare part dispatch by a service provider to the customer is ideally complimented by a faulty part being returned back by the customer. Hence, every part that the service provider sends to a customer is followed by the customer replacing the non-functional part and sending it back. Unfortunately, the service providers lose over millions of dollars globally to non-returned parts and system exchanges. Neither are the companies tracking whether the parts or systems are being returned or not, nor are there processes in place to predict if a particular dispatch is fraudulent. A deep understanding of the dispatch process, with inputs from various stakeholders like Technical Support, Logistics Provider and Customer is required. The objective of the paper is to analyse dispatch frauds and develop a methodology that helps in avoiding and catching a fraudulent dispatch before it takes place, and hence, save millions of dollars in parts intake and sent. The steps followed to build this methodology are:

1. Recognizing and defining the metrics that help in identifying a machine on which a fraudulent dispatch has taken place

2. Carrying out primary analysis to gain a deeper understanding of the data and the fraud process as a whole.

3. Carrying out cluster analysis to group and characterize the features of such a machine and hence draw inference about possible machines on which fraud dispatch would have happened or can happen.

The propensity, of a machine to commit fraud, is used to flag the machine and serves as an alert when they contact the service provider for another dispatch. Subsequently, very high risk machines can be blocked for no future dispatches.