Machine learning techniques have fundamentally altered how oil and gas industry practitioners design fracture operations. In this paper, we perform data analytics utilizing response surface methodology (RSM), a group of statistical techniques that develop a functional relationship between an output variable of interest and several associated input variables, to optimize the output. We apply RSM to optimize horizontal well production based on initial production (IP) of horizontal oil wells for 180 days (IP180 Oil), as a function of five input variables: reservoir type, fracturing fluid (gal/ft), proppants (lbm/ft), cluster spacing, and stage length (ft). The RSM model correlates the initial production of each well to the input variables via a single equation, thus allowing for exploration of the fitted response surface in order to maximize production. Although the choice of the five inputs is made based primarily after consultation with industry professionals, we validate our selection by also applying an assortment of data-analytics-based methods that attempt to rank variable importance and thereby identify completion variables that may be predictive of initial production. The findings rank all five variables above the 50th percentile, thus indicating that the chosen variables have merit. This procedure is applied to a dataset of 201 horizontal wells from the Wolfcamp formations. The model fits reasonably well, with R2 = 61%, a very significant F-statistic p value, and a predicted versus observed values scatterplot indicating a good fit. The RSM analysis suggests that, within the feasible space defined by this dataset, maximum values of IP 180 Oil may be obtained by setting the fracturing fluid in gal/ft at approximately 1972, while simultaneously maximizing the remaining input variables (proppant loading, cluster spacing, stage length). The outcome indicates the possible directions to be taken in seeking a global optimum initial production for the setting of completion variables. Iteration of this scheme may lead to a near-optimum global solution. The real utility of this work may be indicating the way different studies may be designed to optimize production, each with its own selection of inputs, and ultimately be combined in a meta-analysis.