Thanks to customer relationship management systems many companies have a lot of data about their customers
and their shopping and spending history. Using that data and machine learning algorithms it is possible to implement the following example applications with little programming effort.
1) Campaign optimization
First, advertising post is sent to a random sample of the customers in the current customer data base.
One notices the customers reacting to the advertisement and those which are not.
With this data and customer data like age, sex, family status, spending history one constructs a machine learning predictive model for the probability that a customer will react to the advertisement.
Then this predictive model is applied to all customers of the current customer data base and for each customer his probability to react to the advertisement is calculated.
Finally, one sends the advertisement only to those customers with a high predicted probability to react.
Typically, with this approach the reaction rate of a campaign is increased by a multiple.
2) Churn prevention
Roughly speaking, it is said to cost about seven times more to acquire a new customer than to keep an already existing customer. Therefore, it is important to recognize those customers early on, which are likely to want to leave and go to the competitor. This can be done in an automatized way using machine learning.
From the past one has the information which customers left and which ones stayed.
One also has customer data about age, sex, family status, spending habits of these past customers.
With all that data and machine learning one constructs a predictive model for the probability that a customer will leave in the near future.
Finally, this predictive model is applied to all present customers in the data base. Those customers with a high predicted probability of leaving are dealt with separately in an extra marketing initiative in order to make them stay.
3) Estimate the buy-probability of a web site visitor on the basis of the click-behavior
In the e-commerce it is often wished to quickly have a good estimate of the buy-probability of a web site visitor. The only information availably is his click-behavior.
One takes past web site visitors, their click-behavior and the information which of these visitors finally bought a product and which did not. With this data one constructs a machine learning predictive model for the
buy-probability on the basis of the click-behavior.