Predictive Analytics, Enemy Tanks and Plum Customers

So you've developed your customer response model, and are about to spend your marketing budget on a massive customer outreach program. The goal is to increase your response rate to offset the costs. Is it going to work?

The reality is it may not offset the costs. Some predictive models demonstrate 'lift' when applied to the real world. Others do not. What makes the difference?

A lesson can be learned by some of the early pioneers of predictive analytics, albeit in a different industry (defense) and under a different moniker (artificial intelligence or AI).

In the 80's, DARPA was an enormous funder of AI applications. Don't let the name scare you, AI at its core uses lots of data and computing power to produce statistical models, not too dissimilar to the customer response model that you are developing now or planning to develop.

During this time, a lot of research funding was directed to very elaborate, black-box algorithms, such as neural networks. One project of particular notoriety within predictive analytic circles was the development of image recognition system to detect enemy tanks within a picture.

Read on…