Marketers from other fields have often disparaged online marketers for measuring program performance by counting clicks, rather than "hard" metrics such as sales. Direct and database marketers, in particular, have been some of the heaviest critics because they are used to a direct relationship between marketing stimuli and effects: We mailed X pieces that generated Y sales.
Many people assume that finding the relationship between marketing and sales is relatively straightforward; surely we can build a regression model (or something similar) that finds the relationship between the two. But this is a very difficult problem for a few reasons:
- The problem really is a time-series one, rather than a discrete event. Each medium exhibits its own time-series behavior. For example, direct mail can have an immediate effect, but usually has a rapid decay. Television advertising, by contrast, takes time to have an effect and takes longer to reach a saturation point. In other words, there are "lag" effects that must be considered.
- For most companies, there are marketing campaigns going on all the time, plus a host of confounding effects, including competitive spend, seasonality, price promotions, changes in media mix, etc. All of this activity can both reinforce and diminish the effects of a campaign at any given time. For example, in auto sales, an aggressive promotion by a large company can actually stimulate sales for competitors, if it is effective in getting enough people shopping for new cars. So as we measure our sales during a period when a competitor is advertising heavily, our marketing may look extraordinarily effective, even though it is really getting a boost from the competition's spending.
- Most models assume that variables, such as marketing spend, are independent of our target variable, which in this case is sales. But sales and marketing spend are intertwined in ways that are often contradictory. For example, if sales are up, we may have more budget for marketing, but this added spending may not provide a lift directly in line with the new spending level. Conversely, sales may be below expectations, and we may spend more money on marketing to make up the difference. In other words, causality runs in both directions.
As a result of these factors and others, finding a set of variables that enable us to attribute sales in response to marketing stimuli can be difficult. However, a time-series analysis concept known as co-integration is one of the best methods for finding these variables. Co-integrated variables exhibit specific kinds of relationships to each other over time, although at any given time, they may appear to be moving in different directions.
A great example is the price of gasoline in relation to crude oil prices; over time they are related, but sometimes move out of sync. For instance, crude prices could fall, but gas prices could rise (perhaps due to a large refinery going out of production). If gas prices rise too much, they may cut demand, leading to a reduction in crude prices. Similarly, it is well-recognized and long established that store traffic, sales, and store staffing levels form a long-term equilibrium.
This all leads us back to the concept of clicks as a metric. It may very well be that for some applications, click rates can be a useful co-integrated variable. Or perhaps something even more abstract, like the number of online searches done on your brand, may turn out to be just as important. But these variables can only be discovered through rigorous time-series analysis.
The critics of click counting are right in pointing out that most of what passes as analysis of online marketing effectiveness is so basic that it is at best naïve and at worst misleading. But these critics often do not have an adequate solution to address this measurement problem. We also should remember that attributing sales to online marketing too "tightly" misses some of the less tangible benefits, such as a website that may serve as a self-service site or information portal.
Time-series analysis is one of the most difficult types of analysis, so it is not surprising that neither the new- nor the old-school marketers have used it much. We use it extensively for all sorts of analysis, so we know how challenging it can be, but we have also seen how it generally yields very useful -- and sometimes surprising -- insights. We hope that these techniques gain wider adoption; after all, customer relationships unfold over time.
On Twitter? Follow iMedia at @iMediaTweet.