In the marketing research community there's an ongoing debate about the "best" statistical approaches to building marketing models. Champions of different methodologies debate findings, claims, and counterclaims back and forth like teenagers arguing about the best new rock bands. Logarithmic. Bayesian. Additive-linear. Pooled. Store-level. Every technique has passionate proponents.
Unfortunately, this rhetoric often springs not from clients or objective research, but from vendors attempting to differentiate themselves from competitors.
These confusing competitive claims obscure the more important facts: Marketing models are tools, built to address specific issues. Depending on the goal and specific circumstances, they can be finely crafted precision instruments or blunt objects. It doesn't really matter as long as you have the right tool for the specific job. The real art of marketing models lies as much in their proper interpretation and application, as it does in the particulars of construction.
The most important thing to remember about marketing models is that they provide fact-based points of reference for decision support, not objective answers. Models must be applied along with context, experience, and judgment to be used effectively. Models should provide visibility to factors that really drive the business. They should never be used as a "black box" that provides an unexplained answer. When we hear clients say "but the model said...," we know that more education and support is needed.
If the insights indicated by the model can't be easily explained or supported by other analysis, then it's necessary to challenge the model and the analyst both. Different marketing models are constructed with different emphases and for different purposes. When addressing multiple issues, it's important for client and analyst to work together to balance conflicting modeling objectives, and be aware of the trade-offs.
For instance, the best approach for modeling price promotion elasticity will not yield the best results for evaluating media mix. It may be worthwhile to build separate models for the two purposes and triangulate the results.
One of the most important areas is the impact and return on marketing investment (ROMI) of brand equity advertising. Since marketing mix models measure the short-term sales impact of advertising, clients can be left with a confusing information gap when trying to balance short- and long-term views of effectiveness. For example, in a short-term view, advertising romi might be weak compared to trade promotion, suggesting continued migration of dollars to temporary price reductions. Without an objective, long-term view, the marketer might be seduced into an approach that delivers short-term sales at the expense of overall brand health. On the other hand, clients might look at weak short-term advertising impact and dismiss its significance because they are focused on a strategy of long-term brand building. This would be equally unwise.
We know from many long-term studies that short-term ad impact is the best indicator of the strength of long-term ad impact. Since the ratios of short- to long-term impact are pretty consistent, we can be pretty sure that weak short-term results will turn into weak long-term results.
As the demand for marketing modeling increases, experienced analysts will become more of a commodity. These analysts, who combine strong technical competence with a deep understanding of business issues and client strategies, are essential to delivering the promise of improved marketing effectiveness. Since much marketing modeling is based in data and analyses, there's still plenty of room for misjudgment and error. The best path to a great analytic result, is to be sure that an experienced practitioner melds the art and science of marketing in a way that is both statistically valid and makes good business sense for the client.
John Nardone is executive vice president, product development and marketing, for Marketing Management Analytics. (firstname.lastname@example.org)