Precisely wrong or generally accurate?
Editor's note: Jim Kenyon is director of IT services for Optimization Group, an Ann Arbor, Mich., research firm.
For as long as businesses have been advertising, they have been asking “Is my advertising working?” John Wanamaker, the 19th century retailer, is credited with the oft-quoted quip: “Half the money I spend on advertising is wasted; the trouble is, I don’t know which half.” Fast-forward over 100 years and businesses are still looking for ways to understand whether their advertising efforts are working and, if so, which ones and in what combinations.
These are difficult questions to answer, primarily because it is hard to know what advertising has reached a given consumer, much less in what order. In an ideal world, a business would know what advertising, in what sequence, each consumer, transactor or not, has seen. With this data, one could try multiple techniques to suss out the contribution of each component. But the world of commerce is not ideal. A business does not know the entirety of ad exposure of potential customers.
Some businesses – notably those that operate primarily online – like to use last-click attribution for sorting out the utility of advertising efforts. Google, the hegemon of online advertising, even promotes last-click attribution in its Google Analytics platform. It sounds sexy, the tool looks high-tech and, heck, Google is suggesting it, so it must be good, right? Not so fast. Just because something is easy to do, has a high-tech-looking interface and comes from a market leader does not make it accurate. In fact, one should consider whether the tool promotes more AdWord spending or delivers useful results.
A nice Marketing Science Institute article, “Attribution modeling: understanding the influence of channels in the online purchase funnel,” by Hongshuang Li and P. K. Kannan, provides some insights into the issues with last-click attribution. Avinash Kaushik takes things a bit further in his blog post, “Multi-channel attribution modeling: the good, bad and ugly models.” Both of these articles should be on your must-read list if you are tasked with understanding “what marketing has done for you lately.” Using last-click or last-action attribution is a bit like giving the phone company full credit for a phone-based sale.
Our firm has been working on return on marketing investment (ROMI) for nearly 10 years. We took a different approach than many by not focusing on just the readily available data from online channels and we did this for some pretty simple reasons: digital ad spending is still less than 25 percent of all ad spending; and advertising, in general, does not operate in a vacuum.
Not considering the effects of issues beyond your control, but still very real, and not considering “the other 75 percent” of your ad spending seemed like problems to us. Yes, it is difficult and tedious to include external factors or useful proxies for the same. Yes, it is difficult to aggregate spending across all advertising channels – those guys in accounting use complicated spreadsheets and databases. Yes, it is difficult to know precisely when a particular channel was active (i.e., in the field). No, the results are not precise to the last penny, but, yes, they are generally accurate. And that’s the point. Would you rather your decisions be based on precisely wrong or generally accurate information? Would you rather your conclusions and actions be precisely wrong or generally accurate?
In today’s metric-driven business climate, there is a desire to be as precise with measurements as possible, sometimes at the expense of relevance of said measurement. While a car’s fuel gauge can predict velocity with extreme precision, it only is accurate in one instance: when it indicates that the tank is completely empty. The same can be true of Web site metrics. While the number of visits to a particular page is relatively easy to measure, it doesn’t always tell the whole story about how or why a person arrived at that page.
Last-click attribution risks committing a Type II error when the consumer does not continue to purchase from a particular page. Did she go to a physical outlet (Lands' End comes to mind) and make a purchase? Did she take out her smartphone and complete the transaction? Did she forward information via e-mail to her mother-in-law so “Grandma” would be able to pick up the Lego in question at Target? All these cases end up underestimating the value of said page. Beyond this, simple last-click attribution risks committing a Type I error by assigning purchases to this page that were really generated by other sources. Perhaps a TV or radio spot motivated the transaction and the page visited is just the one that the consumer managed to find (sorry, e-commerce developers – many of your sites are very difficult to navigate).
Further, marketers like to make a distinction between customer- and firm-initiated Web site activity. This seems specious as one has limited ability to understand what media an individual has encountered, particularly for non-online channels. Confounding this is online media encountered through different devices. Coalescing multiple (potentially shared) devices into one individual is problematic. Did I just save a link from an e-mail and dig it up as an entry point at some later date not because it was relevant but, instead, because it was handy? Did I ignore a firm-initiated link and go directly to the firm’s site, appearing to be customer-initiated when, in fact, I was responding to a firm-initiated communication? Am I accessing the Web site from a friend’s iPad? The possibilities boggle the mind.
Viable alternative
Fortunately, there is a viable alternative to last-click, last-action, path-sequence and other individual-level marketing effectiveness strategies. We sometimes refer to this option as “follow the money.” In the end, what CEOs, CFOs, CMOs and decision makers of all stripes want to know is “How did our various marketing activities affect sales?” Sales are measured, conveniently, in units and dollars. Advertising spending is measured, again conveniently, in units (of various types) and dollars. Finance folks, being who they are, like to track dollars and when they were spent or received.
Warning: Some complex data mining is condensed in the following paragraph to keep your eyes from bleeding – and to keep you from falling asleep. Not everyone appreciates the inner beauty of a multiple adaptive regression spline, neural network or support vector machine.
Given that marketing and sales activity can be measured in dollars and we can know when these dollars were spent or received, we can mine the data to find relationships, if they exist, between advertising and sales. We don’t have to look for particular Web site path sequences, who clicked on what just before purchase or other myopic measures. As an added bonus, the results are in terms the C-suite understands: dollars. Not Omniture page clicks, DoubleClick click-through rates or other cryptic measures: just simple dollars.
To be fair, we don’t know what caused Customer X to make a purchase. We only know that TV spending, in conjunction with direct mail activity, resulted in higher sales and that billboards did not help sales at all, for example. But the mission was not to determine what motivated Customer X to purchase; it was to determine how the various marketing activities were contributing to sales. Can we predict, based on this data, what sales for next month will be if we increase TV and direct mail spending to the penny? No, certainly not. But we can predict ranges for sales given these spending levels that let you run your business. We can also understand the interactions between marketing activities – and whether these activities will make any difference if economic or other external factors change. Sometimes it is just as important to know when to “sit this one out” as it is to know when to “go all-in.” If the winter is going to be warm and dry it doesn’t make much sense to spend your entire marketing budget trying to promote higher sales of extra-warm mittens. As an old fisherman once said, “Save your bait; the fish just aren’t biting today.”
Velocity is slowing
Why do we give so much attention to Web sales? Is this where the majority, or even a large part, of commerce is conducted? According to census.gov retail sales data, e-commerce represents less than 6 percent of all retail spending in 2013, the strongest year, to date. Granted, there are some segments, like small consumer electronics, where e-commerce has taken hold. But this does not seem to generalize to all products, particularly not to large (in size or dollar value) items. Interestingly, the velocity of e-commerce sales is slowing over the 13 years of available e-commerce sales data. If it is going to make brick-and-mortar sales obsolete, it’s going to take some time; perhaps enough time that some other distribution channel will come along first and e-commerce will end up like catalogue sales: another distribution channel that set out to conquer the world and ended up as a bit player in overall commerce.