Let's face it, we all want due credit for our work. That's why marketers so eagerly embraced the death of the presumptive "last click." That mythology forced a paradigm of clear winners (the content that ultimately converted) versus losers (everything else). Today, we know that's patently untrue. In fact, Microsoft's Atlas Institute estimates that between 93% to 95% of audience engagements with online advertising receive no credit at all when advertisers review campaign ROI.
Enter attribution modeling. Perfect science it is not, but it is a welcome departure from the last-click paradigm. What we're really talking about here is giving credit to what happens before that last click or, more accurately, before the conversion or sale takes place. We don't need this information to establish credit or place blame; we need data to refine our strategy and maximize each dollar we spend. If an online retailer sells 50,000 pairs of shoes in Q1, the million-dollar question, literally, is how or how much each marketing strategy contributed to the campaign.
So this brings us to some practical lessons on attribution modeling. There's no universal model, but there are a few tried-and-true principles:
Lesson #1: Don't fall in love with one or two metrics. An attribution model that looks only at site visitors, for example, may reveal important data on the value of that last click, but it reveals nothing about the placements upstream that pulled those eventual buyers further down funnel.
Lesson #2: Ad servers can help, but they're only one tool. On its own, an ad server can only track upper-funnel outcomes such as click-throughs and viewthroughs. This adds a piece to the attribution puzzle, but doesn't complete it. How do you, for example, determine how many clickers abandoned an ad in favor of a Google search, which would then become their final conversion point? This information is quite valuable.
Lesson #3: Integration of data sources is worth the investment of time and resources. By integrating data from multiple sources, let's say server and site data, marketers can create a single map that charts all paid media exposure and their link to site outcomes. More sophisticated advertisers also include other "events" in their maps, like direct marketing e-mails and organic search referrals, which they can then attribute to specific outcomes (e.g., order size or lifetime value). Yes, it requires some extra thinking to build this integrated model, but the performance reward is worth it.
So integration isn't easy or inexpensive. In fact, gaps will still remain that you'd be wise to close with other available techniques and technologies. Panels are one example. More and more, marketers are relying on panels to better assign attribution. Here are four ways panels can help:
It may sound clichéd -- but in our industry, experimentation is key to lasting success. Still, without a clear methodology for attribution, you'll likely be among those who mistake winners for losers. If, on the other hand, you reject the winner/loser paradigm, an artifact from the days of "last-click" thinking, you'll discover a secret many now realize. Attribution isn't a zero-sum game; it's your best way to learn more and earn more.
"•To avoid confusing cookies with consumers, a panel can provide a constant view of consumers to expose anomalies"
Ye-es, but a counterview is that a panel can provide a constant view of a small sample of atypical consumers, while cookies provide realtime analysis of all actual computer visits and interactions. But hey, I don't work for a company that supplies panels for a living! ;c)
As always, enjoyed your article (love your one about Behavioral targeting last year also).
My challenge with posting this comment, is how not to pimp out what we do. Data integration is VERY easy and inexpensive now. But hey, I work for TagMan, a vendor neutral universal tag solution that works seamlessly with everyone out there, including panels and cookie based data tied to other services ;c)
Companies that can tie panel to tag based measurement have the truest view of all, such as comScore's Unified Digital Measurement methodology, bringing together both the real person based insights of a panel with mass consumer usage of ad tags. It works for audience measurement and campaign measurement too.
Why on earth would you ignore robust one-to-one attribution data all the way to the point of transaction origination...in favor of TV-like panels which Neilsen developed in the 1950's??
This is a great article and very true. However, the business case for getting a lot of extra people involved becomes questionable. Why not use a centralized service?
Data analysis is extremely expensive and the people capable of doing it are few and far between.