The time step is an interval that you choose. In the article you linked it says, correctly, that the smaller the time step the more accurate the result but also the more computing resources required. As such the article recommends that you set your time step to equal the conduction timescale (as per the formula) giving you a trade off between accuracy and computing resources.
If you want to use this method then estimate the length of an element near a heat exchange interface (here the elements will be small and the thermal gradient high). After you finish the simulation, rerun it with a smaller time step. If the results are the same it means your time step was already sufficiently small.