The first attempt at solving a data science problem, using a simple, known and proven algorithm. The goal is to deliver in a very limited and predictable amount of time a first version of the algorithm, and compare further versions with it.
Good data scientists develop a baseline before working on a more complex algorithm.
Sometimes, a baseline is sufficient, and no further algorithm is needed.
If you cannot precisely estimate how long it will take you to develop a baseline, then it is not a baseline. If it is the first time you use an algorithm, you can’t call it a baseline.