scaling features (e.g. w/ z normalization or min/max scaling) to a comparable range is an important step so that the training works effectively and all features are treated equally within your training process. E.g. this can help to reach the optimum more quickly.
This thread could be interesting to check out:
If you think about what would happen if you do not scale at all: the algorithm would neglect the features with small magnitude and focus on the „large magnitude feature(s)“ too much, even though they might not have not best predictive capabilities among all features.
Please let me know if you have any open questions!
Best regards
Christian