We want a demand forecasting model from our sales data. Where do we start — try a few algorithms and pick the most accurate?

First I check whether the data actually exists and is reliable, what business metric this moves, whether a simple rule would do, and who will use the output. I do EDA before anything else. I once found campaigns couldn't be traced to products, and we had to fix the pipeline and wait for new reliable data before modeling.

I also don't automatically run a baseline then gradient boosting, since we usually already know the baseline will perform worse, and I pick the most explainable metric rather than the 'best' one. For a regression problem I've used the ratio between mean absolute error and the IQR, to show the model's error sits below the typical spread of the values it predicts.