Trying to assess the effects of marketing on business performance is a daunting task, celebrated long ago by John Wanamaker’s famous quote regarding not knowing which half of his advertising was wasted. The root of this problem is that, for most companies, as much as 40 percent to 80 percent of their business results cannot be attributed to any specific marketing activity.
Numerous reasons exist why some business results are non-attributable, data capture errors aside. In brief, they may be the consequence of a company’s prior brand building or represent the halo effects from competitor advertising, which might be stimulating demand generally in the category.
Another problem with attribution is that some consumers are stimulated in several ways, often simultaneously, and they may simply choose to respond through one channel rather than another. A prime example is when consumers are drawn to a company’s Web site to investigate available products and services, then opt to purchase through an offline channel. With more companies encouraging consumers to contact them in more ways, this has become a bigger issue.
The obvious concern is that, without knowing how to handle non-attributable results, assessments of marketing’s effectiveness are clouded and decisions regarding how to spend new marketing dollars may be seriously flawed.
Apportioning business results across multimedia marketing activities. One way to solve the attribution problem is to use an econometrically based performance simulator to identify the various advertising and non-advertising factors that are causally related to all business results, including the non-attributable results.
Econometric techniques, especially time series regression, provide a well-structured means of evaluating the drivers of business performance by isolating key explanatory variables and holding constant certain variables that may mask the effects of advertising. For instance, sales may be related to the amount of spending in several media such as television, print, direct mail and online advertising during a specific time period.
Special events and even public relations activities may also affect short-term sales. With adequate data, the performance simulator can estimate these influences as well.
In essence, the performance simulator explains the up and down variations in business results, say, on a weekly basis by looking at variations in marketing spending and other factors that occur in proximity to those results.
To do this, the simulator estimates the carry-over effects of advertising as well as saturation effects. Though more difficult, it also may be able to isolate the effects of brand equity on results.
The performance simulator also can determine which media types directly affect business results and which “interact” with others synergistically. Separate effects are typically quantified for non-advertising influences as well, including economic trends, market factors, seasonality, competitive actions and product changes.
Taken together, these latter effects may be considered “the baseline.” So this way all results, both attributable and non-attributable, are ascribed to one or more of these marketing investments or baseline factors.
Reapportioning results by media and other causal effects. An important advantage of using a performance simulator is that it can look retrospectively at what happened after a certain amount of marketing spending or can look prospectively at what is likely to occur based on planned marketing spending.
In either case, the beauty is that the models can account for the full range of attributable and non-attributable business results.
Though we believe this is the best approach to understand what is driving non-attributable business results, it can be limited by certain factors. The approach requires a couple years of reliable historical data and a high level of statistical expertise to develop the underlying models used in the performance simulator. There also is no guarantee that the simulator always will find statistically significant relationships or remain stable when radical changes in the marketplace disturb longstanding relationships.
Despite these potential problems, we have found that this approach adds important information to the marketing decision-making process.
A brief example. Consider the case of an Internet company that serves as an online marketplace. This firm uses a multimedia marketing strategy, centered largely on television and online advertising. It also does a fair amount of outdoor advertising. Though it has an active Web site with large volumes of daily traffic, it is uncertain what drives this traffic or the ensuing online transactions.
Banner click-through rates are modest, and few individuals who click through transact during that same session. As a result, most key business results are non-attributable, forcing the company to make its marketing decisions mainly on intuition.
Using a performance simulator it was shown that the company’s site activity, though highly influenced by various non-advertising factors, is also affected in the short term by television advertising and even those dubious banner ads. The simulator also detected an interactive effect caused by the combination of television and online advertising, as well as other contributing influences.
With these findings, the company decided that its major marketing efforts in general are worthwhile and that online advertising specifically should be expanded. It also decided to conduct more advertising tests and experiment with alternative media mixes now that it had a way to measure their effects.
Having a way to account for non-attributable business results is extremely valuable for companies because these results often represent a significant portion, if not the lion’s share, of total business activity. Marketers need to understand what factors are responsible for producing these results since they may determine whether their marketing activities will achieve their business goals and be cost-effective.
An econometrically based performance simulator can be a powerful tool to help marketers not only understand what caused certain results to occur, but also to estimate what results will likely occur based on new marketing spending.
