I love my clients, really I do, bless their little geeky, pointy-headed selves. However, they are not so good sometimes at explaining what they do to non-geeks such as myself. (I may have a geeky tendency here and there, but I am definitely no statistician; in fact, I haven’t had a math class since high school, although I did pretty well on the basic math GREs.)
So right now I am working on a green card case for a guy who applies stochastic statistical techniques to marketing, and he is remarkably laconic regarding what this means in terms of what he actually does every day; the most I’ve been able to get out of him is that he analyzes marketing data. I’m going to need more detail than this to do a proper job on his case, but getting him to expand on his job description is like pulling teeth. How might stochastic statistical techniques be useful in the analysis of marketing data? If y’all can help me understand what stochastic statistics is all about, maybe I can formulate more targeted questions for my pointy-headed client. Because lemme tell you, if I can’t understand what he’s talking about, then the Agency Formerly Known as INS definitely won’t understand what he is talking about. Here is the least pointy-headed explanation I’ve found so far:
Stochastic methods cover a range of techniques applicable in various fields, which I guess doesn’t help you much!
I would assume he’s trying to either model behaviour (by means of the application of stochastic methods to historical data to make predictions) and/or evaluate alternative strategies. The ultimate purpose of both would be to provide information to aid decision making. for example, he might model buyer behaviour in certain situations in order to inform the decision making process regarding future marketing strategies.
I think that he means that he can build a probabalistic model for how consumers will act based upon arbitrary market conditions. He analyzes behavior in the past under specific market conditions to build this model. This allows the end user to fiddle with those conditions in the model, to see how consumers would act in a given set of circumstances, even if those particular circumstances had never happened all at the same time. If the model is correct, they can plug in current market conditions or estimated future conditions to make a plan for what to do.
“Stochastic statistics” sounds a little redundant.
Stochastic is often used in “stochastic processes” meaning processes whose outcomes can only be described probabilistically – in this case, maybe, the behavior of who he is marketing to.
I head up a marketing analysis group for a database marketing company. What he is basically doing is using statistical techniques for a variety of analytical insights, but two key things for marketers are profiling/segmentation of customers and statistical models that “predict” customer behavior better than just randomly marketing to the general population. The idea is to craft a product’s message and then to aim that message to the customers that are most likely to buy that product. The idea is to generate the highest return on marketing dollars, e.g., get the most sales out of the lowest cost.
Typically the types of statistical analysis tools are regression, clustering, CHAID and CART.
I’ll illustrate with a simple example. I want to find all the people who might be interested in buying a Porsche. Based on the data I have about people who have bought a Porsche in the past, I could build a variety of segments that are most likely to buy. Segment one might be:
Male, aged 45, married, one child, income of $175k, net worth of $600k.
Segment two might be:
Male (getting an image here?), 32, not married, income of $110K, net worth ???.
And so on. Based on predictive modeling, we know that segment one is our best customer prospect, so lets go after guys that fit that description first, then go to segment two, etc.
Me, a geek? No way, I’m really cool, my Mom said so.
