Phase II: Product realization and heuristics update
Starting from a generic basic skin moisturizing cream, identified as K0-N1, which solves the problem specified in Table 2 at a minimum cost, specific changes were made to comply with different attributes for the two evaluated references (Samples 1 and 2). For this, additional heuristics from Table 3 were included in the algorithm as restrictions that allowed searching for feasible alternatives, which increased or reduced the level of the aimed attribute. The resulting formulations are presented in Table E-1 of the Supplementary material.
From the results obtained in the consumer assessment phase for Sample 1, ease of absorption, freshness, and residues were the most important attributes by the composite index. Due to the high interaction between these attributes (see Table 5), and the fact that the base formulation left more white residues than the Sample 1 (Figure 3), the first change made to the algorithm was to implement one of the actions to reduce the white residues by adding up to 2% low HLB emulsifier (see Table 3). A series of experiments were performed to understand how each change affected the emulsion properties, especially the white residues after product application on skin and the product viscosity. Each one of these alternatives was generated using the optimization problem specified in Table 2. The name of the formulations was given according to the following parameters: The first letter indicates if the product is a cream (K) or a lotion (L), while the number corresponds either to the base case (0) or to the samples 1 or 2. The following parts also contain letters and numbers. The letter corresponds to the aimed attribute(s) - R for “Residues”, G for “Greasiness/oiliness”, V for “Thickness” and N if no attribute is aimed. The numbers correspond to a counter of the alternatives at this stage.
For the case of Sample 2, the starting point was set to be Formulation K1-R3, as in this case residues are the most important attribute too (0.34 in the composite index; see Table 6), and from the previous example, it was known that K0-N1 left considerably more residues than Sample 2, as it can be observed in Figure 3. The next action targeted the thickness, as there was no reliable action to correct and easily evaluate the change of the freshness attribute. Two different actions were taken, to observe the results of each change.