Intelligent Enterprise

Better Insight for Business Decisions

RSS
Webcasts
Whitepapers
Subscribe
Home

December 5, 2001

Never the Same

Using value state models can help you classify customers to make better business decisions

by Erik Thomsen

Quality, unlike death and taxes, is not a constant. This variability applies in all areas — customers, products, suppliers, workers, or managers. One year's profitable customer or product is another year's money loser. Suppliers who are reliable one quarter may be unreliable another.

Because quality varies, you can track changes in quality over time and classify entities or processes (whatever you are measuring) by the way their quality changes or remains the same. Recording these changes is a necessary first step to analyzing the causal factors that promote increases, decreases, or stability in quality over time. The most systematic way to look at quality changes is by creating what I call a quality or value state model.

Tables 1A and 1B illustrate the difference between tracking quality per entity or process over time and tracking the ids of entities or processes by their quality changes. In the following sections, I describe some of the main steps for building customer value state models. You can use the same techniques to create value state models for product lines, suppliers, manufacturing methods, distribution channels, or business processes.

At its simplest, a customer value state model requires some collection of customer- and time-specific quality values. However, because the most common and fungible quality indicator is "earnings," I will concentrate on earnings-based customer value state models.

CREATING THE MODEL

The information requirements for building an earnings-based customer value state model include access to sales transaction data containing fully discounted revenue per item per customer (the stated revenue net any discounts) and the total costs (including all attributable overhead) per item per customer. Otherwise, deriving earnings per customer, transaction, and products is impossible. These earnings derivations are necessary to create any earnings-based, customer value state model. (See Tables 2A and 2B.)

Creating an earnings-based customer value state model involves the following steps:

1. Calculate and assign a total cost to every item sold.

2. Subtract any discounts, explicit or hidden, to determine a net value for revenues in the transaction detail table.

3. Sum the costs and revenues from the detail table into the summary table.

4. Calculate earnings per transaction per time per location.

5. Sum the costs and revenues along the time dimension of the summary table to that level of time required to see a single time-specific value state of the customer.

Step 5 requires business knowledge. You must determine the time granularity over which a customer value state is applicable. For example, typical catalog customers make purchases on a seasonal basis. For a catalog company, judging the performance of a customer on an hourly or daily or even weekly basis wouldn't make sense. Instead, a catalog company might sum customer transactions to the quarter level before looking at earnings and assigning a value state. In contrast, a luncheonette owner might define customer quality by the amount of earnings generated per week and so define a week-level value state model.

At the end of step 5, you have an intermediate customer value model that uses customer id and time as dimensions and customer value as its measure.

6. For any n time periods of interest at the granularity chosen in step 5, create at least three ordering bins that correspond to high, medium, and low earnings per customer per time period; assign each customer to an earnings bin. (See Table 3.)

Keep in mind that you can choose several different types of ordering bins. If you pick a relative ordering such as ranking, you will guarantee that some customers are always high valued, some medium valued, and some low valued. Furthermore, the numeric definition of the boundaries between the different bins may change over time. In contrast, if you define fixed earnings amounts as the boundaries between low, medium, and high, you may create earnings categories for which no customers exist.

7. Build a dimensional model out of any two-to-n, time-specific earnings bins used as dimensions.

Inversion occurs at this step. Up until now, earnings have been treated as a derived variable. And earnings bins were groupings of a derived variable. To create a value state model, you must treat two or more time state-specific earnings bins as low cardinality dimensions relative to which you will analyze a variety of customer behaviors. In my March 30, 1999 column, I wrote about the need to treat so-called dimensions and measures or variables as distinct use cases of the same underlying structure; dimensions and measures have no inherent differences. Value state models are a concrete class of applications that depends on using measures as dimensions. Table 4 shows an example of a customer value state model.