The other day I was flipping through the channels and landed on an episode of CSI.  I forget which flavor, it looked sunny and beachy so it was probably not the one in New York City.   Anyway, in this episode a murder took place and the only evidence seemed to be a passerby’s recollection of a license plate and hair color of the driver.  Not much to start with but as we all know that is enough for the CSI team.  

Armed with some pretty outstanding technology the team’s first challenge was creating a pool of people to interrogate based on the passerby’s account.  Not surprisingly a partial plate and a hair color provides about 200,000 candidates and this crack team only has an hour.  So how do they breakdown this size to determine who to interview?  They segmented the candidate pool then, using behavioral and demographic data of past murderers, matched those segments to predictive models to see who was most likely to be a murderer.  Like all perfect prediction models, the computer reduced the candidates from 200,000 to 5.  Good work HAL.   And wouldn’t you know it, one of those 5 was the bad guy.

After watching this I did some research and found that this application of predictive analytics is based on some truths and criminal best practices.  In the crime-fighting world the models are normally used for dispersing cops and optimizing coverage.  The episode is obviously fictional but their process:

1. collect historical data,
2. build a population of subjects, 
3. analyze for trends, patterns and similarities (segment), 
4. test and investigate
5. act

This shows, in a more entertaining way, the best practices approach to building a useful predictive analytics model.  If Hollywood and your local police force are using predictive analytics to entertain and protect then shouldn’t the marketing department be using it to increase revenue and understand?