Data Smart: Using Data Science to Transform Information into Insight

4

Optimization Modeling: Because That “Fresh Squeezed” Orange Juice Ain't Gonna Blend Itself

Business Week recently published an article about how The Coca-Cola Company uses a large analytics model to determine how to blend raw orange juices to create the perfect not-from-concentrate product.

I was discussing this article with some folks, and one of them blurted something like, “But you could never do that with an artificial intelligence model!”

They were right. You can't. Because Coca-Cola doesn't use an artificial intelligence model. It uses an optimization model. Huh? What's the difference?

An artificial intelligence model predicts the result of a process by analyzing its inputs. That's not what Coca-Cola is doing. Coca-Cola doesn't need to predict the outcome when they combine juice A with juice B. It needs to decide which combination of juice A, B, C, D, and so on to buy and blend together. Coca-Cola is taking some data and some business rules (their inventory, their demand, their specs, and so on) and deciding how to blend a product. These decisions enable Coca-Cola to blend juices with complementary strengths and weaknesses (maybe one is too sweet and another not sweet enough) to get exactly the right taste for the minimum cost and the maximum profit.

There's no one outcome that needs predicting. The model gets to change the future. Optimization modeling is analytics' Arminianism to Al's Calvinism. Free will, baby! (Sorry, that's the last historical theological joke in this ...