According to my opinion and scientifically one can not do this by closing the eyes and picking a random value from 0 to 1.
Maybe, maybe not. Tuning a smoothing filter is usually an empirical task, you pick a value that works "well enough" in practice. You're trading-off smoothness of result against speed (i.e. ability to respond quickly to changes in the input).
On the other hand, if there is a specific time-constant you're aiming for, then yes, just use that! *
You can analyse the filter expression to determine its frequency response. However, that basically requires the discrete-time Fourier transform (and maybe the Z transform as well). That's a little out of the scope for a Stack Overflow answer, I'm afraid!
* However, I'm not sure what you read from that datasheet that led you to believe that that particular value of tau is what you should be using...