(A "symbol", in this context, is simply a group of some relatively arbitrary, non-zero number of bits, transmitted as one atomic unit.)
While in simple cases the two can be identical, as soon as one symbol encodes more than one bit, they diverge. And that is incredibly common in modern data transmission systems.
Therefore, in practice, the bit rate of a channel will almost always be higher than its baud rate.
The baud rate matters more when you are interested in the requirements imposed on the data transmission physical link layer (such as a radio link's spectrum bandwidth, or the transition times in a modulator). The bit rate, on the other hand, matters more when you are interested in how much data can actually be transmitted within a given time period. Both are useful metrics, but as they measure different quantities, they answer different questions.
As an example, consider quadrature amplitude modulation; 4096-QAM encodes 12 bits (212 = 4096) per symbol. Therefore, the baud rate for a 4096-QAM channel is 2-12 = 1/4096 of the data rate of the same channel.
In practice, some of those bits being transmitted will, in turn, likely be used for various kinds of metadata; things like frame headers, checksums, start and stop bits, on-bus addressing, forward error correction, and so on. In places, data is deliberately stretched out, not uncommonly to improve error handling or simplify synchronization. For example, SATA uses 8b/10b encoding on the physical link layer, and CDDA extends data multiple times for various reasons between the raw audio sample data and what's actually on the physical storage medium. Designs like these will reduce the rate at which useful payload data can be transmitted below that of the raw channel bit rate, and that is relevant for the amount of time it takes to usefully transmit some given amount of data, but they do not change the bit rate of the channel itself.