Retail Customer Acquisition

This course teaches you to understand how to value customers financially, and how to recruit financially valuable customers in situations where you do or do not have access to vendor databases of potential customers, time to test acquisition promotions, or access to multivariate techniques such as regressions and neural nets.  You will learn how to use the programming/analytics language R to accomplish this.  The point is to recruit as many of the most valuable new customers as you can afford to, so as to improve the value of the firm as rapidly as possible. You should have already under your belt: Some knowledge of the retail decision environment (bricks and mortar or on-line). Some knowledge of the programming language R. Some knowledge of statistical methods up through OLS multiple regression. Some knowledge of customer privacy law in your local environment, or access to expertise on the topic.

Describes typical structure of customer data files and introduces a sample master customer file for an imaginary retail business in .csv format Excel knowledge required), reading it into R using the R Studio graphical user interface, and summarizing it. Also introduces a products file.

The various sources of customers data, strategy of what to purchase, and a discussion of costs.

This is part 1 of a discussion on how to calculate Customer Lifetime value, In order to cherry-pick good future customers from among your non-customers, it is extremely helpful to have a measure of “good.” That is, what is the value of a customer to you now, and what is your best projection of his or her value into the future? The best available measure of “goodness” is Customer Lifetime Value, variously known as LTV or CLV. Under "Resources" you will find an incomplete Excel spreadsheet you can use to follow along with the lecture and build your own CLV calculation spreadsheet. Part II (module 5, called "4b") fills in the remainder of the discussion of how to use CLV and why it is the preferred measure when it is possible to calculate.

This short module finishes the calculation of CLV and addresses why CLV is the superior measure of customer valuation and how it can be used.

Customer Lifetime Value, the subject of the last modules, is sometimes difficult or impossible to calculate. So, here is a collection of other measures of individual customer value that can be used to identify “good” non-customers for acquisition – not as connected to firm value, but workable. In future lectures, when we use CLV as our measure of customer value, you can, if you must, use one of these.

How you go about acquiring new customers will depend upon whether or not you have a good customer database (and can thus calculate and use a measure of customer value), 2) whether or not you have access to a good vendor database of noncustomers at reasonable cost, 3) whether or not you have time available for running a test and data-mining the test results, and 4) the which analytic techniques you have the software, data, and competence to use. This module discusses which of the following modules will be of most use to you, given your situation.

Here you are still in a bit of a bad spot. At least you have a reasonably large customer data base, but you can’t afford to access vendor data bases of non-customers, you can’t afford either the time or expense for testing, and you don’t have the software or the expertise for multivariate data analysis. So, we’ll do the best we can with what we’ve got, which may not be particularly effective.

If you haven’t already, you should review module 7, because this one is just a variation on it. Here, you don’t have access to a vendor database or multivariate methods, but you do have time and resources to run tests to find the most effective means of new customer acquisition among those available to you.

This situation is somewhat better than the previous ones, because we can use vendor databases to both enhance the information we have on our current customers, and also to cherry-pick the consumers from the vendor databases that look like our best customers. The cherry-picking is less efficient, because we can’t use multivariate methods, only logical “IF” type statements. Furthermore, since we can’t spend the time for testing, we can’t refine our approach to cherry-picking. But, the use of vendor databases represents a significant improvement in our efficiency of valuable new customer acquisition.

This situation is somewhat better than the previous ones, because we can use vendor databases to both enhance the information we have on our current customers, and also to cherry-pick the consumers from the vendor databases that look like our best customers. The cherry-picking is less efficient, because we can’t use multivariate methods, only logical “IF” type statements. Furthermore, since we can’t spend the time for testing, we can’t refine our approach to cherry-picking. But, the use of vendor databases represents a significant improvement in our efficiency of valuable new customer acquisition.

Even better, now we have the time to test, which will allow us to do some data-mining of the results of that customer acquisition attempt to refine our customer acquisition strategies when we “roll out” the full promotion. Our approach here will be to: Send a test promotion to a large random sample of non-customers in our vendor database, and Mine the results of that customer acquisition attempt to try to maximize the return on our costs of customer acquisition. However, the lack of multivariate methods means we could still do much better.

Continuation of Topic 10a.

We have another big improvement in our fortunes, which is the availability of regression methods. OLS regression (or some other form you might choose to use) can be used to predict the CLV of a non-customer, as part of the decision of whether or not to attempt to acquire him/her. That will make our customer acquisition more efficient, in terms of return on investment, than the cruder forms of data-mining seen in earlier modules. However, in this module we have assumed there is no time for testing, so we have no opportunity to mine the results of a test and refine our decisions about whom to approach in vendor databases.

Now, we are really in the catbird seat. We have vendor data, we can use regression methods, and we have time to test. This will make the customer acquisition activities a bit more complex, but far more efficient. Furthermore, we can predict the financial results of the final roll-out of the customer acquisition attempt with reasonable accuracy.

Continuation of 12a.

In this case, we will learn a new tool, neural nets. Unfortunately, without time to test. we lose a good opportunity to make the new customer acquisition really efficient, but we will be testing in the next module.

The most powerful situation. We have a powerful and flexible multivariate method, access to vendor data, and time to test.

Continuation of actual module 14a.

Continuation of actual module 14b.

Teaching all of the software necessary to achieve automatic marketing and machine learning in the customer acquisition problem would be well beyond the scope of this course. However, we can present the conceptual framework for that approach here.