Deborah's post reminded me of what I didn't have time to get back to
in chapter 4. See Figure 4.12 - Analytical Database Hierarchy. The most
powerful of the techniques is predictive modeling. Predictive models are
the basis for data mining. Here's a good short explanation about how it
works in banking (it's a Google doc).
Predictive modeling is essentially regression models (not necessarily
linear) that pair a probability of response (based on cases in the
customer database) with independent variables that are known to be
predictive of the phenomenon. If you want to know how the models are
developed (another place where you need a statistician who is an expert
in response models), there is a lot of material on the web.
What marketers need to be clear about is that with a small amount of initial information about a newly-acquired customer, they can predict the likelihood of specific types of behavior occurring in the future. This is the basic of credit-scoring models and many more product-specific models.
What marketers need to be clear about is that with a small amount of initial information about a newly-acquired customer, they can predict the likelihood of specific types of behavior occurring in the future. This is the basic of credit-scoring models and many more product-specific models.
How long have banks been using this? Predictive analytics are great because they can be used almost anywhere, although they should be paired with qualitative analysis too, as numbers can often lie.
ReplyDeleteAlthough this is in banking, the concept is similar to a project I worked on for about three years on sensor technology that did predictive modeling for vehicles. Sensors would be placed in locations in/on the engine, battery, and several other parts of vehicles.
The sensors would record oil viscosity, miles driven, temperature, fluid levels, and hundreds of other types of data that would be downloaded (or later sent wirelessly) to a central database where they could predict when the vehicle would fail based on analytics. Metrics like mean time between failure were used.
The data would improve existing records and offer warning signs in real time so that condition based maintenance could be taken before a failure occurred. I was responsible for collecting the benefits of the program, and one of the things that I found was increased battery life and reduction in misdiagnosis of batteries and electrical systems which resulted in cost savings for the batteries, costs of disposal, and time costs of diagnosing the fault in the electrical system, and discovered that we could also extend the average oil changes from about six to nine months, saving 33 percent in oil costs.
The airline industry (Southwest), locomotive engines (GE), Caterpillar, and other industries have been using this type of technology to reduce maintenance and operations costs, mitigate faults/failures, and improve reliability while enhancing customer satisfaction.