Before a model can compare features, the features have to be put on a common footing. One might be measured in single digits and another in thousands, and left alone the larger-scaled one dominates purely because of its units rather than its importance. Normalization rescales them so that comparison measures what you meant it to measure. It is an unglamorous preprocessing step and skipping it produces confident nonsense.
The same step is missing from most informal financial comparison. Two costs quoted over different periods, two returns measured over different windows, two percentages taken against different bases: in each case the raw figures are not comparable, and comparing them anyway produces an answer with the shape of a conclusion and none of the content. The arithmetic runs fine. It is the units that fail, silently, which is why the error is so easy to miss.
What makes this hard in practice is that normalization requires deciding what the common base should be, and that decision is a judgment rather than a calculation. Per year, per unit, per dollar invested, per month of use. Different choices are appropriate for different questions, and a comparison is only as good as the relevance of the base you chose. Picking a base that is easy to compute rather than one that matches the decision is a real failure mode in both contexts.
The discipline, then, is two steps and not one. Convert everything to a common base, and state what that base is when you write the numbers down. The stating matters as much as the converting, because a normalized figure with the base left implicit is one step away from being compared to something else on a different base by a future reader, including yourself. Machine learning pipelines record their preprocessing for precisely this reason, and a note next to a number you worked out accomplishes the same thing.
When you last compared two financial options, did you actually convert them to the same base, or compare them as quoted?
Educational only · Not financial advice · Results not guaranteed.