Keep in mind that a single aminoacid substitution in b1,4GalT leads to it being a GalNAcT. So structure based predictions of enzyme activities are going to have their limits. Generally (my knowledge extends to data as of ~ 2008), you can guess quite accurately whether a GT sequence encodes a fucosyltransferase or a sialyltransferase using sequence similarities. You won't be able to predict the acceptor, though. Acceptors can be glycosphingolipids and glycoproteins or even proteins. Then, for higher organisms (bacteria have been known to switch their linkage with a single aa exchange) you can probably tell UDP-sugar (Glc, Gal, GlcNAc): beta 1,3 glycosyltransferases, beta 1,4, alpha 1,3 and alpha 1,4 again by sequence similarity. Most glycobiologists agree that it's not going to be a beta 1,5 glycosyltransferase. What you cannot tell from sequence simillarities is the right donor (UDP-Glc, GlcNAc, Gal, GalNAc) or acceptor substrate (alpha or beta? Glc, Gal, GlcNAc, Fuc.... but probably not Sialic acid because that's always at the end of a sugar chain.) With alpha 1,6, beta 1,2 etc - forget structural approaches completely - (cazy61 family encompasses an O-GlcNAcT->EGF, beta 1,2 XylT->Man etc...)

Experimentally, you have to be lucky and stumble over the right donor and acceptor combination (acceptor in the right configuration on a pNP or benzyl moiety) all the while being sure that your metal regime (Mn2+ or Mg2+...) is the right one and that you aren't too far off with the pH. Be warned, you can easily waste years of your life looking for an activity and no activity can also mean that your GT isn't well expressed or that your tag kills the activity etc... For glycosyltransferases that work directly on proteins, there's only genetics approaches that I would know of (GT phenotype phenocopying substrate phenotype...). For these guys, you will need high end apperature to biochemically prove activity (MS, HPLC). And do start with cazy.org, but make your own blast searches for the closest known GTs, look at the expression patterns, formulate a hypothesis and check around it. For expression, use high expressing systems (baculovirus etc.) and agree with your supervisor on when exactly you are going to stop looking.