A total of 60 peaks generated by forced degradation of three active pharmaceutical ingredients (APIs) were used for the column selection in the development of a stability indicating HPLC-UV analytical method using LC-UV and LC-MS peak tracking. Two mobile phase additives and two organic modifiers were evaluated while screening a list of carefully chosen chromatography columns. The column screening was utilized and the best column selected based on total number of resolved peaks, resolutions, peak widths, and peak shapes. 0.1% TFA in ACN/water was used for the initial screening and optimization of the gradient profile. Three different concentrations of TFA in ACN/water were also evaluated. The optimum TFA concentration, 0.10% (8.77 mM), was considered as optimum for further gradient optimization based on the resolution of critical pairs. After the selection of column, mobile phase and mobile phase modifier (TFA) selection, optimization of the gradient was achieved by a combination of automated chemometric peak tracking and software-based decisions in AutoChrom MS. The correct peak retention equations (i.e., retention time vs. mobile phase ratio) were generated by using first one-step gradients with a wide range of % B followed by optimization in multi-steps gradients. It was found that extrapolation, using quadratic retention models, can lead to large errors in retention time (tR) predictions, especially for poorly-retained components. We present the challenges in resolving the critical resolution pairs, including those with the same m/z, the overestimation and the prediction errors of the software, and why the peaks model (i.e., accuracy of predicted versus experimental) fail at the extremes of the gradient. By using this approach we were able to generate a suitable stability indicating chromatographic method for an extremely challenging sample comprising of three active pharmaceutical ingredients (APIs) and related degradation products with a wide range of hydrophobicity. The APIs were in-house compounds, their identity are blinded in this paper and are not relevant for purpose of the study. There were good matches between the predicted and experimental retention times of the tracked peaks. The peak model was used for the generation of an assay/potency method using the computational tool only.