Data mining and modeling: from survey response to customer insight

Survey responses can be a powerful source of motivational data in any customer profiling effort. When stored in a marketing database and merged with behavioral and demographic information, survey responses contribute to profiles that provide powerful insight - assuming the data elements are properly defined and transformed in advance.

Unfortunately, the converse is also true. Improperly defined or transformed survey data will contribute inflexible, incomplete or misleading information to the database, thereby supporting incorrect conclusions and their inevitable outcome: bad marketing decisions. This is not just a garbage in, garbage out data integrity issue. Even a valid response to a good survey question can yield unfortunate results if: The question is intended or phrased so the responses gathered cannot be applied directly to developing customer intelligence and marketing insight. The response is time dependent. The question is attempting to obtain information that is available elsewhere. The response is not mapped and transformed to a standard scoring structure that would allow marketers and researchers to use the scores as easily and consistently as they could any other element in the database. To achieve the desired results and avoid the pitfalls, marketing should take the lead in developing and documenting business rules for profile surveys that ensure: The right question is being asked in the first place. A valid question for one constituency can be rendered invalid for another on phrasing or emphasis alone. Take, for example, Rate your level of satisfaction with this product. While providing some information to the product team, it does little to help develop additional customer insight. For marketing purposes, it is better to base the question on both attribute importance and performance scores, i.e., ask how well the product met a brief- respondent-prioritized list of requirements. Phrased one way, the organization learns only how well received their mousetrap is. Phrased another, they learn 1- what makes a better mousetrap in the eyes of their customers and 2- how well this product lived up to those expectations. Responses do not contain a time-dependent element. Surveys designed to support a sales effort depend largely on dynamic time-critical information: when are you likely to buy, etc. But the volatility of this type of information makes it ill suited to a marketing database s customer profile. Maintenance requires frequent follow-up that could alienate customers and prospects. The alternative, a profile that provides dated information could damage the underlying database s credibility in the eyes of the user community. For reasons of both data integrity and marketing relevance, it is better to define a standard profiling survey featuring less volatile information related to who the customer is, as opposed to what they plan to do. If the same insight can be obtained through other means, with data that is at least as predictive as a survey response, leave the question off the survey. First, there should be an appreciation for the time respondents invest in completing surveys. For that reason alone, surveys should be limited to only those questions that cannot be answered cost-effectively- by any other means. Keep it brief. Second, behavioral data e.g., purchase history- is, in general, a better predictor than motivational data e.g., stated intent to purchase-. A marketing database needs motivational data only where 1- there is no analogous record of behavior upon which to base an analysis or 2- there is a need to validate the predictive ability of such data. Of course, a database designed exclusively around customers could have the ability to store both purchase history from sales data and stated intent to purchase from survey responses. If both were stored and made available to the user community, the result would appear to provide a choice to the user community in a bigger and better database. But this choice is an illusion that will expose itself as such the minute users realize they are selecting the purchase history in lieu of the stated intent every time. On the other hand, a database designed around prospects might include intent to purchase since there is no better predictor available.- The challenge in all such keep/drop decisions is to clearly and unambiguously deliver the insight required to define, measure and analyze marketing initiatives. If one data element delivers that insight effectively, don t muddy the water by introducing another only masquerading as a choice. Define a single response scale from your organization s perspective and map all responses to that scale.This scaleis, simply, a continuum of most favorable/least favorable, with responses mapped accordingly before applying them to the database. A very high consumer response to value would map to most favorable. The same response to price would map to least favorable. Each survey question should, therefore, come bundled with its own transformation algorithm. These transformations are often difficult to define and maintain, but the rewards in doing so far outweigh the pain. A consistent scale provides a framework for selection, discussion and analysis that would be impossible to achieve otherwise. It allows survey response data to be used as selection criteria, unit of measure and result as consistently and unambiguously as any other field in the database. In other words, it would enhance the database s capabilities and improve the results delivered by the professionals that the process was built to support - which are precisely the objectives of any business automation effort. Like marketing, its parent discipline, survey development is both a science and an art. But once developed, survey usage and measurement is pure science. Capturing, mining and leveraging all of the insight survey responses can provide demands a solid process, documented rules and the discipline to effectively use them. But the rewards, in the form of consistent, quantifiable, clean data ready to be easily converted to actionable marketing intelligence, are great. Source: www.dmreview.coma>