Direct marketers have to understand the issues and choices involved in customer segmentation to be able to manage such a project effectively from conception to implementation. Any number of common pitfalls can stand in the way of success. We'll suggest how they can be avoided.
Goals
Several important considerations must be addressed before you start to segment your customer file. The fundamental goal is to identify groups, segments or clusters of customers (the terms are interchangeable) that from a marketing perspective are meaningfully different from each other.
Obviously we could create segments that differ demographically or in terms of behavior. But will these differences lead to a variety of marketing strategies that will be more effective than a single approach that treats all existing or potential customers the same way?
Intuitively, you know that your customer base, be it 50,000 or 5 million, is not homogeneous. In other words, at any point in time it will contain some good customers, some bad ones; some new customers, some old ones; some young, some old; some rich, some less rich, some poor; some price sensitive, some not; some extremely loyal, some not loyal at all. The list goes on and on.
And, because of this diversity, it makes little apparent sense to market to all customers the same way, i.e., the same level of marketing effort, the same offer, the same copy, the same creative. Yet this is what direct marketers do, perhaps without realizing it, when they search for the best control package in direct mail (the package that works best across all customer files) or the one best script in outbound telemarketing.
When pressed on this, many DMers will point to tests that clearly show that the control packages they've developed over time have out-pulled the segmented packages tested against them. So, assuming the testing was done correctly, the question is: Was there something wrong with the segmentation strategy itself, or are the weaker results due to how the strategy was implemented? We think it's more the latter.
So let's start our look at the way segmentation projects are initiated.
The first thing you need to decide on is the type or kind of variables you want to build your segmentation around. The three primary types are demographic, behavioral and attitudinal.
Demographic Segmentation
In segmentation systems based on demographic data the emphasis is on describing customers in terms of their personal characteristics (young singles, young families with children, affluent empty nesters, etc.), which through further research can lead to an understanding of their needs, wants and behavior.
While there are several ways to collect customer demographic information, the most common approach is to purchase demographic enhancement data from an outside supplier. The supplier matches your customer list against a large compiled consumer database using name, address and phone number, and then appends the demographic data from the compiled database onto your file.
Because compiled databases and matching technologies are not perfect, you will never obtain demographic data for every customer. Usually, hit rates will range from 50% to 80%. Records that do not receive this type of demographic data can be backfilled (or approximated) by appending geodemographic data — usually averages for the area in which they live (average age, income, home value).
Companies wishing to employ demographic, lifestyle and life-stage segmentation but not wanting to create their own segmentations will frequently turn to products such as Prizm or Cohorts to meet their needs. While a meaningful description of these products and their competitors is well beyond the space we have here, suffice it to say that these systems allow a marketer (through exact name and address matching, or based on just the ZIP codes these customers live in) to segment their file into 40 to 60 predefined clusters that may be useful for company-specific marketing segmentation.
Behavioral Segmentation
Segmentation systems based on behavioral data emphasize what customers have purchased (clothing, appliances, big and tall sizes, baby clothes, closeout deals). This information can be used along with a product cross-sell analysis to suggest other products they may be interested in.
Companies with a comprehensive customer database in place have the best possible source of data for a behavioral segmentation. Because all purchasing data is already in the customer database — or should be — the hit rates will be 100%.
Behavioral-based segmentation systems are the best fit for product-driven organizations where product line managers call the shots. This type of segmentation is not as useful for individual customer value development. When new customers are acquired, the organization has to wait for them to define themselves via their purchases. Depending on the type of business and its characteristic frequency of purchase and/or seasonality, it could take quite a while to obtain enough purchase information to correctly classify a customer.
Even after your customers have made a few purchases, all you know is what they've bought. People could be buying things that you sell from another source, but you have no way of knowing that. Also, this approach does not give you any insights into how you could influence a customer's purchase decision.
In a similar fashion, behavioral segmentation is not as useful in product development efforts because you only know what the customers have bought, not what they are most likely buying elsewhere or what unattended needs they may have.
Attitudinal Segmentation
Segmentation systems based on customer attitudes emphasize the nature of the customers' relationship with your company. Defining buyers in terms of how they feel about your firm helps you understand why people do business with you. It also gives you some understanding of how they position your company against the competition. From there, it's possible to anticipate which value proposition or creative slant might best appeal to each customer segment.
The best way to collect customer attitudinal information is to ask them — that is, do a survey. However, surveys tend to be quite expensive to conduct. And even if you had the necessary funds available it's almost impossible to comprehend what type of incentive would be necessary to get everyone to respond.
And the critical question: If you survey only a sample of the customer database, how will those not sampled be assigned to the proper cluster?
To begin exploring this question in detail, let's assume we're building a segmentation model for a large retailer with multiple outlets — a situation we all can relate to.
Again, it's intuitive that not all of the retailer's customers will feel the same way about the chain. Some shop there because of service, others primarily because of price; some because of the breadth of the product lines, others because of convenience; a few because they like the ambience of the stores, others because it's the hot new store in town. You get the idea.
Now let's assume we all agree with the above and we decide that we want to create a scheme that would segment our customer database along the following set of major dimensions: price, service, product depth, product quality, the overall shopping experience and brand image.
How would we go about it?
In DIRECT's March 15 issue we'll answer this and other related questions, such as: Given a segmentation based on a sample, how do we place all our customers into the proper segment? And if we can't, what should we do?
This is the first in a series of articles on customer segmentation. Later installments will discuss the tools of segmentation: factor analysis, cluster analysis, discriminant analysis, CHAID and logistic regression.
