It occurred to me today, in a meeting where we were talking about a pricing model for a complex of services, that that’s another application for statistics.
If you have a complex thing that you want to give to your customers as a simple thing, you don’t want to have a complex price associated with it. Your cost to produce the thing is a function of many variables you know, many variables you have a vague idea about, and many variables you don’t even know about yet. You can’t hand that massive cost function (times 1.15) to your customer, you need to simplify it to 0-3 or so variables for them, while making sure you can still make money.
There are statistical methods around that can let you model a complex space with a simple one, while putting (probabilistic) bounds on the error. Good thing to think about; it gives a way to make a factored, principled argument about pricing that doesn’t suffer from overestimation (like interval-based methods tend to) or unbounded error (like ad-hoc methods tend to).