If you want your software to be used in the social sciences, esp. psychology, I recommend two critical features - the first involving simple slopes and the second involving hierarchical regression.
In general, the ability to follow up an interaction involving a continuous variable to conduct a "simple slopes" or "simple effects" analysis in which, at a minimum, the user can either probe the slope of a line at different values of another predictor (and obtain slopes with SEs) and even better, provide contrasts between these slopes. The other predictor may be continuous (and thus the ability to plot plus or minus 1 SE would be necessary) or categorical.
A second very common technique in psych is hiearchical regression in which two models are compared with one nested within the other. The only way I can do this using JMP is to run both models, extract the model dfs and SS values and MSE, compute an F, and then look up the F in a table (or compute the associated p-value using a JMP Formula). Very clunky. This is super simple in R, SPSS, and even Jamovi.