A common beginner mistake in Pandas is writing explicit loops to process rows, which cripples performance. This approach forces the library to handle data row-by-row in Python, bypassing its core columnar, vectorized engine built on NumPy. The mental model stays stuck on individual records instead of entire columns, leading to code that doesn't scale.

The performance gap is stark. On a dataset of 500,000 rows, a standard loop took 129 seconds to label sales tiers. Replacing it with a single vectorized operation using `np.where()` completed the same task in 0.08 seconds—a 1,600x speedup. This happens because vectorization pushes computation into optimized, compiled C code operating on the whole column at once.

Even cleaner is boolean indexing. Create a default value, then use a condition like `df["sales"] > 1000` to generate a True/False mask. Apply the override with `df.loc[mask, "tier"] = "high"`. This method is both fast and readable. The `apply()` function, while tidy, still executes Python code per row and should be a last resort for logic that cannot be vectorized.