Source coding is exemplified as data compression--with obvious benefits: You don't have as many bits to send.
Channel coding fits the signal in the channel. For example, you'll have certain noise characteristics and certain SNR spectrum for a given channel. For a flat SNR spectrum and AWGN (additive white gaussian noise) you need a moderate amount of interleaving (smearing data over time so that no bit gets stomped-on too badly) and a combination of integration and FEC that is optimal. Current FEC works well when the raw BER is 1/50 or better. So you'll want to integrate up to a BER of 1/50, then put a half-rate FEC on top of that. That will give you a BER of (pretty much) zero.
Also, Shannon showed that you should NOT split a wideband channel into multiple narrow bands to send the same data. So this means you need to think more about the modulation in order to fit it snugly in the channel.
If you took the easy way and simply integrated until you had a good BER, you'd find that you would be wasting a lot of the channel.
Things get even more interesting when the noise is not AWGN, and/or the channel SNR spectrum is not flat.