Market Segmentation Blog

The Bridging Model:

Connecting A Segmentation To Customer Databases

As part of most segmentation research initiatives, an algorithm is developed which assigns future research participants into one or more segments.

The algorithm is programmed into what is known as a typing tool. Behind the tool is a set of mathematical equations that use the responses to a smaller set of survey questions, say 10 to 20, to predict to which segment(s) new research participants belong. Because the typing tool is based on a subset of the total number of original survey questions, there is some error in segment assignment (typically, accuracy ranges from 70% to 90%+). The typing tool is considered an acceptable alternative to refielding the entire survey. After all, asking all of the original survey questions on top of any new ones would make for a long, tedious respondent experience!

Typing Tools Classify Research Respondents Into Segments

Typing tools are intended to be implemented in primary research—both qualitative and quantitative. They identify key groups that are considered targets for new marketing initiatives or product development efforts. Yet, typing tools have limitations. To ensure the highest possible accuracy of classification, the questions included in the tool must use the original survey question wording and answer choices. In most cases, respondents must provide an answer for each question. These limitations are not problematic in primary research (question wording can be exact and responses to all questions can be required). The challenge arises when a company wants to implement the tool in non-research settings.

Understand The Challenge

Consider this scenario: your company has conducted a robust segmentation. The deliverables include a report, detailed personae, and a typing tool that allows future research participants to be classified into the segments. Management is delighted with the outcome and instructs the ad agency partner to generate a suite of online creatives (email, social media, banners) for each of the target segments. The agency suggests some primary quantitative research to obtain consumer feedback prior to launching the campaigns. Management agrees with this approach, but also wants to hear from their current customers. The company maintains a large database of buyers, including demographics and transactional history. It should be straightforward to assign each customer in the database to a segment and move forward with the campaign evaluation. Easy peasy, right? Not really.

Companies want more from their segmentation research investment. Ideally the segmentation would be leveraged throughout the company, which includes not only customer insights, user experience, product, and R&D teams, but also digital marketing and the sales force. Marketing wants to craft the right content to send to the right recipients. The sales force wants to identify which customer segments need different relationship management approaches to maximize retention. To be useful for the entire company, the segmentation must be applied to the “respondents” that each division cares about. In the case of marketing and sales, “respondents” are customers and prospects whose demographics and transactional behaviors are usually warehoused within large databases.

The typing tool would be the easiest way to classify customers and prospects in the database. However, the typing tool questions—a mix of attitudes, behaviors, and demographics, likely do not exist in these data sets. Obtaining the needed information to populate the answers in the typing tool often is impractical. Developing an online version of the typing tool and requesting customers to “type” themselves is an option (but not always feasible). The typing tool would not be a good way to move forward.

Bridging Models Classify Actual Customers Into Segments

The concept of a bridging model is simple. Imagine a deep chasm. On one side of the chasm is the survey data that was used to produce the segmentation solution (i.e., the segments). On the other side is the repository of information about customers and/or prospects (i.e., the company’s database). The bridging model literally connects the two sides of the chasm.

Set The Stage For Success

Although the concept is straightforward, the actual bridging can prove more challenging. The model must be built using information that is shared by the original segments and the database to establish the appropriate connections. This requires advanced planning. Starting the bridging model discussion in the early stages of the segmentation research process is key.

• Shared variables. Conduct a thorough review of the information contained within the customer database. Companies often augment their databases with third-party purchased variables (household size, annual income, ages, magazine subscriptions, and even hobbies and interests). These same variables can be appended to the respondent segmentation survey data. Once the segmentation analysis is complete, the bridging model can be developed using only the appended variables that are also available in the database. For B2B segmentations, third party data is unlikely to exist. An alternative is to include survey questions that mirror the database variables. For example, if company size is a variable in the database, a survey question can be constructed to capture the same information.
• Sample composition. If possible, include a subset of current customers in the segmentation survey. Having respondents with known transactional behavior (which cannot be purchased from 3rd parties) can help enhance the model. These appended variables also can be tested for inclusion in the original segmentation model.
• Multiple internal databases. The various departments in some companies maintain their own databases, each with different information housed in different formats. Furthermore, the customer database is often separate from the prospect database as the information contained in each database is likely to differ. In addition to the potential chaos these can generate, multiple data sets represent a unique challenge for the bridging model. This might call for building more than one bridging model.

Evaluate and Refine

Once the model is developed, it needs to be tested. If the company maintains close relationships with its customers, the sales force can review a subset of the classifications for face validity (does it make sense for Customer X to be in Segment 3?). Conducting A-B testing is another method to check the bridging model and the segmentation as well. Provide the segment-specific creative (such as a marketing email) to half of one segment and a standard creative to the other half. Observe the outcome (clicks on email links, actual purchases, or calls to the sales force—whichever metric is most relevant for the business). If the customers who received the segment-level ads demonstrated significantly more engagement with the brand or products, then the bridging model and the segmentation framework overall are likely working.

Occasionally, though, bridging models require some refinement. The threshold probability to assign a customer to one segment versus another might need to be adjusted because more customers than expected are being placed into a particular segment. Or certain variables need to be dropped from the model because of low fill rates (few customers have values in these data fields). The model and its implementation can be continually tested and updated to maximize utility for the organization.

Final Thoughts

Segmentation is a very powerful tool. When possible, leverage that power by applying the segmentation to your company’s database(s). Organizations that successfully classify their customers into segments increase the likelihood that their brand communications and new products will delight and retain them. This makes the time, effort, and investment to build the “bridge” worthwhile.

Elizabeth Horn

Senior VP, Advanced Analytics

Beth has provided expertise and high-end analytics for Decision Analyst for over 25 years. She is responsible for design, analyses, and insights derived from discrete choice models; MaxDiff analysis; volumetric forecasting; predictive modeling; GIS analysis; and market segmentation. She regularly consults with clients regarding best practices in research methodology. Beth earned a Ph.D. and a Master of Science in Experimental Psychology with emphasis on psychological principles, research methods, and statistics from Texas Christian University in Fort Worth, TX.

• 
•