Retail Facility Location And Staffing Optimization Through Machine Learning

Humans exhibit striking regularity in their daily interactions with their environments.These frequent interactions are captured and maintained by retailers and businesses in large databases.Extracting useful insights,however,remains elusive for medium and small companies without large departments for data mining.This body of work attempts to help such businesses by providing workable solutions that are within reach.In that,the challenges of facility location and staffing policies are addressed in this research.The challenge of selecting a facility location has been the subject of many works of literature.Nevertheless,the approaches proposed range from the cumbersome to the out-of-reach to most small and medium enterprises.Likewise,staffing policies constitute a significant cost to businesses and have been studied thoroughly in previous literature works.This thesis aims to use publicly-available tools such as Python and MS Excel to address this gap and introduce a method that is easily accessible to these businesses.Telecommunication companies store terabytes of data of individual’s travel habits,the results of this thesis demonstrate that by exploiting these datasets,medium and small companies can identify,not only established areas of considerable density but also previously unknown or unexploited clusters.Furthermore,despite the lack of personal information such as age,gender,occupation,etc.,this thesis was able to create dense demographic grids based solely on geolocation data.From that,we were able to predict the type of individual who is more likely to visit that area,thereby confirming previous researches’ assertion that over a long period,humans tend to show consistent regularity in their behavior.Moreover,we introduced demand uncertainty into our model to ascertain the managerial implications of this research.Our methodology was shown to reduce the absolute forecast error by 92% over other widely-used location selections models.Building on the predicted frequency of visits,we also optimized the staffing policy under various scenarios based on a newsvendor model,where first,a service level was identified and then building on the projected number of items sold,a staffing policy was implemented.