*3.4. Improve*

As a result of the analyses performed and summarized in the previous section, the team responsible for this process was able to pinpoint a specific behavior in the process potentially related to the loss of purity/production volume of the desired product (*y*4, *y*6, and *y*8) when Premix 2 was fed into the reactor. When compared to one another, the average value for *y*8 for *x*15 = 1 was found to be 1.9332% lower than for *x*15 = 0, and the standard deviation for *y*8 for *x*15 = 1 was also 1.12 times that for *x*15 = 0, with both differences being statistically significant (*p*-values < 0.05). A preliminary confirmation of this was provided by the historical data itself, according to which Premix 2 was fed to the reactor for the first time on September 2014, matching the increase in variability and drop in average value for *y*8. While the technicians at the plant knew about this operational change (which they themselves implemented), they initially argued that such modification did not justify the change in the process outputs, but the unexpected behavior of variable *x*23 (see Figure 11) pointed to either a contamination issue with Premix 2, or to other, not registered process variables as potential causes. Consequently, they decided to further investigate them and, once detected, to correct the actual issue (the details on how this was done is not disclosed for confidentiality issues) and standardize the treatment of both Premix 1 and Premix 2 to make them equal. The solution was validated during the next two months, and remains a success to this day, with estimated benefits/savings above 140,000 €/year (40,000€ higher that the initial estimation in the 'Define' step).
