We don't have as much experience with small molecules as we have with peptides, where there are 7 or so standard peptide mixes (Biognosys, Pierce, Waters, SCIEX, Sigma Aldrich, etc.) you can order with the necessary properties of having 8-15 molecules (peptides) spread across much of a C18 gradient. Once you have a set of standards you can use as "anchors", you would measure them on your gradient, preferably in you sample matrix of interest, since the sample matrix can impact elution times and then assign them time-independent values, known as iRT values, e.g. 0 and 100 for the 2nd and last peptide in the tutorial, which then determines a linear equation from measured retention time to iRT value (iRT = mRT * slope + intercept), which you use to assign iRT values to the rest of your standards.
Once this is done, you acquire a method which includes both your standards and any unknowns you want to add with full gradient chromatograms. The important part is that this method needs to be good enough for you to identify a measured retention time where your unknowns elute. If they are truly, truly unknown and you have no way of pinpointing where they elute in this context, then you need to work out your method of identifying their measured retention time. Without that you can't calculate an iRT value.
Assuming you can identify a measured retention time in your full gradient method, then you simply run a regression between the measured retention times of your standards and the iRT values you assigned them in the prior experiment. This gives you a new linear equation (iRT = mRT * slope + intercept) and you use that equation to assign your unknowns iRT values based on their measured retention times.
Once this is done and you have iRT values for everything that interests you, then in any sample that contains your iRT standards (landmark molecules), once the times of your standards are measured, the times of the rest of your molecules can be predicted by performing a regression of the standards between iRT values and measured retention times to produce a linear equation the other way this time (mRT = iRT * slope + intercept), then you use this to convert the iRT values for your unknowns in your iRT library to predicted measured RT values.
Skyline does all this for you and the tutorial you mentioned is intended to teach you how for peptides. The tutorial page should also point to a webinar and there are other recorded lectures where the concept is explained.
Hope this helps. If you feel you are meeting all the requirements but failing to achieve the desired results for small molecules, please post screenshots of what you are doing and seeing and we will try to help more.