Leveraging Client-Supplied Data

Technological advances have enabled businesses to collect vast amounts of data, and many companies have some form of loyalty program to attract and track customers. Often complete analysis can be done using these data exclusively. Other times, these data improve the quality of survey data provided by market researchers. We can ask respondents to tell us what they have done recently, or what they believe that they will do next (stated behavior), and why. Companies often know exactly what people have done (actual behavior), but not why.

Stated Behavior What people say they do and what they actually do are not always the same. We can ask for stated behavior in a questionnaire and make inferences about what people will do in the future, but often the stated behavior does not match actual behavior. For example, how many times have you eaten in a fast food restaurant in the past 30 days? To answer this question, you might apply a heuristic. Typically, I eat in fast food restaurants one time per week. Thirty days is four weeks, therefore the answer to your question is four. The truth, most likely, lies somewhere between three and six. The data would be more accurate if we had a loyalty card that the restaurant scanned before each purchase.

Actual Behavior The ideal client-supplied data is accurate and up-to-date, with a well-populated client dataset--that is, records for a high percentage of customers or members are complete and correct (or near-complete), particularly in target segments--however, the data does not measure attitudes and perceptions. You need a survey to understand why. For example, you might know for a fact that a specific customer eats lunch in your restaurant every Tuesday, but the reason might be either that the customer loves the Tuesday special or that the customer has a standing appointment in the building across the street. This is where the ability to marry client database and survey data creates a perfect dataset.

Key Considerations Following are some important considerations when trying to marry survey data and database information:

1. What type of data do you have? Transactional level data may need to be aggregated before we can use it. For example, sales data for the past 24 transactions may need to be sorted by product type purchased, date purchased etc. How accurate are the data? Just because a string of digits is entered for zip code, it doesn't mean that it's accurate. It's at least as important with database information as with survey data to do outlier and distribution checks on the data. "999999" just doesn't happen that often in nature.
2. How well populated is the dataset? What percent of customers/members have all/most/enough records "filled in?" And when this varies from close to 100%, why and among whom does it vary? Missing data is prevalent in client databases. You need to be careful to interpret this correctly. Absence of information may not imply absence of behavior.
3. If the company has demographic data, how were these data acquired ? . Household level data, like PRIZM data, may not be as accurate as survey data (or company application form data). Purchased data typically represent the average profile for a person living in that zip or postal code and may not perfectly match the true demographics of the actual household.
4. Where are the data stored? Sometimes, the data may reside on multiple servers. In what format are the data? How will we export the file?
5. It will add more time to the project. For example, in one study, we had transactional level data for true online shopping activities that needed to be aggregated into category level data that provided an adequate number of observations for analysis. There are 100's of locations where one can purchase a shirt online. Some websites get more traffic than others and can be used, others need to be grouped into categories. Several iterations may be required before the data can be used.

The "holy grail" in marrying client-supplied data with survey-based market research is uncovering segments that vary robustly and meaningfully across both attitudinal and behavioral dimensions - the "ah-ha" moment that may not be apparent when segmenting on either dimension alone. Often the perfect dataset is elusive, but combining client-supplied data with survey-based market research with a "matrix" approach -- i.e., viewing populations together in terms of behaviors, and separately in terms of attitudes -- yields surprising and actionable results. The advantage of aligning client-supplied data with survey-based market research is that one type of data answers "what do people do?" and the other answers "why?"

Case Study An online shopping segmentation study was conducted to perform a multi-dimensional segmentation of the general adult online population (18+) that spent over $50 on e-commerce in the past six months, in order to:

• Identify segments with favorable affinity towards certain brands, and/or desirable profiles for tapping revenue opportunities for the client (and/or e-commerce partners).
• Identify high-leverage targeting, messaging, and positioning strategies to attract desirable and approachable segments.
• Support identification of meaningful linkages between these segments and behavioral patterns at the clients website and other behaviorally targetable environments.

The client supplied true online behavioral measures and allowed us to administer a 25-minute questionnaire to each person. The behavioral data were used to create the segments and the survey data were used to enhance the profiling of the segments. The client leveraged the data set they had collected by combining it with survey-based market research. The result was meaningful, accurate data -- both attitudinal and behavioral -- that could drive effective, targeted marketing and sales strategies.