COMPANIES ACQUIRE CUSTOMERS through a variety of marketing channels and offers. Products and services are then cross-sold to increase product penetration, retention and profitability.
Ideally, a firm would develop unique targeting models or strategies for each combination of product, offer and channel to maximize results. However, this is usually not feasible — even with the largest analytical staff — because of the sheer volume of effort needed. And because products promoted through several channels and a mix of offers may appeal to different audiences, a one-size-fits-all approach usually won't work either.
There's another solution to this common marketing dilemma. By grouping combinations into logical offer segments — customers who look similar based on the items offered and product ownership — a marketer can more effectively target its wares while working within the limits of its analytical resources.
A direct marketer of music, for example, might have hundreds of thousands of titles it groups into categories including pop, classical, rap, classic rock and oldies. Besides music CDs, it sells DVD music videos. Historically, it acquired customers through direct response TV and Internet banner ads and currently markets by direct mail and e-mail, among other channels. Marketing's focus is on the sale of music and video series in which products are shipped within a category to customers until they decide to stop purchasing.
To develop separate models for each of the music categories by channel and product type (music or video) would require 60 individual models (15 categories times two channels times two product types) and would not be cost-effective. The company could opt to develop a single model for all categories; however, this likely would result in little differentiation within variables such as age.
Many marketers faced with this situation might choose to develop models for key products and services and buy vertical lists for the remaining products to test the ones that will pull the best. Segmentation also could be used as a quick and easy way to eliminate the “dead wood,” or models created for one situation could be applied to others. If limited analytic resources are available, some of the promotions might even be sent to random audiences. In certain cases products may not even be offered due to the cost of building unique models to promote them. While these methods are effective up to a point, none are optimal.
By combining categories into logical offer segments, the music company could minimize the number of models needed and yet show differentiation between the groups. These segments would include like-minded customers who generally would be as prone to respond to one set of offers as another.
To create these segments, the company would first overlay a demographic/life-stage clustering system that also could be applied to prospect lists. Next, the clusters would be ranked by the probability of individuals within them to own a product from a particular music category. This could be done by comparing the percentage of product owners in a specific cluster relative to the entire country.
For example, if there are 2,000 customers who own classic rock products and 100 of them are in cluster A, then their penetration would be 5% (100/2,000). If cluster A represented only 2.5 percent of the total U.S. population, then customers in cluster A would be twice as likely as customers in general to own those products.
The next step would be to select the top clusters for each category and group the categories by overlapping clusters (it's not necessary that all the clusters match for each category). Those categories that share similar clusters fall into logical offer segments. It's possible for categories to be included in several segments. For example, pop, classic rock and oldies-'70s might form a logical offer segment.
With logical offer segments identified, the company could then create a single model for this segment instead of three separate ones.
The process could be taken a step further by adding a dimension such as music vs. video products. Here the groupings might be very different and a logical offer segment might include classic rock music and video, pop video and oldies-'80s music.
Creating unique models by specific offerings is certainly an effective way to target. However, are there many situations where it makes more sense to use logical offer segments instead of creating separate models by category, product, offer and channel?
Consider these factors:
-
Will the item be promoted to a limited target audience?
-
Will the item be promoted infrequently?
-
Will the offer change?
-
Does the item have limited profitability?
If the answer to more than one of these questions is yes, then logical offer segments should be considered.
If gospel music is promoted only once a year with a new offer each time, building a separate model would be an inefficient use of resources; it could perform poorly due to changing offers. Combining the gospel items with other products would allow the marketer to build one model for all products.
How do you decide which dimensions (channel, product and offer) should be included in logical offer segments? Think about whether a particular dimension will add value to the process or if other constraints or options provide an alternative.
In many cases, the channel is dictated by the availability of contact information, appropriateness of a channel-offer combination, and the cost of using a channel. E-mail is one of the least expensive channels — but there may not be enough available e-mail addresses to meet your business objectives.
One alternative is to create models to predict the likelihood of response from each channel. For prospects or customers with multiple contact information, channel models can help predict which channel to use to market to a specific individual.
Consider if the dimension will interact with the other dimensions or stand on its own. Will it only add more complexity? Pricing, for instance, may not be a good dimension. It can be split into individual pricing offers, grouped into low/medium/high or full price/discounted categories.
While most people would prefer to buy products and services at discount, they don't expect to purchase everything that way. Sometimes they'll pay different prices based on circumstances. Because of these issues, pricing may add too much complexity as a dimension.
Similar to using demographic and life-stage information for clustering, credit data may be used to create logical offer segments. If a credit granter is continually adding new portfolios to its mix, it may be difficult to build separate marketing and risk models for each new set of customers in a timely manner.
Rather than treating each portfolio as a unique group, clusters can be developed with credit data so like-minded customers can be treated similarly based on creditworthiness.
JEFF NAGEL is director of marketing solutions for TransUnion's Analytic Decision Services division in Chicago.
