Comment by wasabi991011

Is that even remotely possible?

You'd have to ensure a positive expectation value over not only every item, but every combination of items a consumer could by. You could focus only on the most likely possible orders (assuming you have the data, I don't know how many stores actually track combination of items bought), but it's not obvious to me that there's a tractable top n most likely orders that gives a reasonable enough estimate of expectation value.

On top of that, you would be interfering with whatever system you already have that sets the cents of each item (whether marketing with 99¢, or % discounts, or a system that tracks that 97¢ means lowest sale, etc).