Well the first thing I'd do is, create a scatter chart:
x: date (week)
y1: cost
y2: unit

If it's efficiency, then you're probably looking for units / cost.

I don't know that you can assume a normal distribution is even applicable. And then, and then, I don't know why you need that.

If it's finding efficiency, first define efficiency, and then see what numbers push farthest in that direction.