Josh Chasin is Chief Research Officer of Comscore, a marketing research company.
AdExchanger.com: Of the three silos of metrics – audience measurement, campaign analysis tools and web analytics – in your opinion, which one is the toughest to grasp for the end client?
JC: One the one hand, I think the most vexing thing for the client is that the three don’t match. In network TV, for example, all metric streams come from the Nielsen ratings– there is no analog to web analytic data (maybe Set Top Box data will get there) and ad campaign data is run off the ratings data. So the advertiser raised on TV is flustered by the fact of multiple metric streams.
On the other hand, there continues to be an issue with both ad server and web analytic data, in that they talk about “Unique Visitors,” but unlike audience measurement they base these measures on cookies. At this point, we all know empirically that cookies are not unique at all, and they can dramatically overstate and thus misrepresent reach. And this isn’t just coming from the audience measurement side of the divide anymore; web analytics guru Eric Peterson summarized the issue nicely right here.
I would refer all prospective Internet advertisers interested in parsing the definitional differences and limitations of the different ways to calculate Unique Visitors to the IAB Audience Reach Definition Guidelines.
How is Comscore quantifying the Long Tail?
How long is the long tail? 50,000 sites? 100,000? A million? Well, at comScore, for the US in May 2009, we show a total of 504 billion Page Views (across all sites, long tail included). Almost half (48%) were on the top-100 properties. 59% were on the top-thousand properties. I don’t have adspend data in front of me, but I’m certain the advertising revenues are even more highly concentrated in the head than the Page Views.
That said, we’re able to go about 35,000 sites deep in the US right now, and we’ll thoroughly quantify the Long Tail through Media Metrix 360, which involves hybrid measurement combining our 2 million person panel with site-centric server data; the server data provides robustness when you get far enough down the tail that the panel sample sizes become an issue. But right now, for buyers and sellers of online advertising, let’s not kid ourselves—the money is disproportionately in the head and torso, not the tail.
As part of its new Comscore 360 offering, the new Comscore beacon combines with the huge 2 million member Comscore panel to provide a supposedly better estimate on audience. Why is this better?
For a good 12-15 years, the dialogue in online metrics was about site-centric versus panel-centric data. Each is good at different things. Naturally the way the mind works, people began thinking, “What if we could combine the best of both worlds?” (A friend calls this “you got chocolate in my peanut butter.”)
Panels are great for measuring people; site-centric data are great for measuring machines, especially when those machines are servers. Internet measurement is essentially about measuring the way people use machines.
Panel data are good at providing true unduplicated reach; shared traffic across entities (do my visitors also visit my competitors?); demographic composition; and duration. Site-centric data provides a census of all server activity, which, when properly filtered (eliminating out-of-country traffic, non-human traffic, ineligible events such as redirects) provides an empirical count of the user-initiated activity occurring on the website. The integration of the two essentially results in the allocation of the person data—in all its richness—across the census count of activity.
It is important to note that not everything that generates a server call or “Page View” in site-centric data corresponds to an eligible Page View in audience measurement; the latter is by definition exclusive, because advertisers want a rigorous accounting of the actual pages that users in the target actually were exposed to. So hybrid and internal data won’t necessarily match. But they ought to move in concert together, and publishers, advertisers and agencies will end up with a keen understanding of what drives the differences. Hybrid measurement will remove a good bit of the uncertainty from online metrics, and that will prove to be good business for everybody.
How do you differentiate Comscore 360 with a company like Quantcast that has had an installable “beacon,” if you will, for several years?
In all honesty, I don’t tend to see Quantcast in our competitive set. Their business model is entirely different from ours, and is far more similar to those employed by ad exchange companies such as Platform A, Specific Media, Blue Kai, or eXelate.
Saying that there is too much data seems like a cop out. Do you agree?
Well, I’d probably say that a complaint of too much data is really a plea for more insight.
Data is ubiquitous. But knowledge is elusive. It is important for those of us in the metrics space to remember that ultimately we are trying to inform better business decisions. It isn’t how much data you have; it is how you facilitate the deployment of that data in the service of those practical applications.
So I guess I’m saying I agree.
Are reach numbers becoming irrelevant in Comscore’s monthly numbers related to ad networks? With clever frequency capping, it’s possible that Comscore reach numbers can appear more impressive than they really are from a frequency perspective.
Well, one thing I’ve thought about is something that one might call “effective reach”—what is the ad network’s reach among persons exposed, say, 3 times? (based on the old saw that effective reach is between 3 and 10 exposures.) That would go directly to the heart of the issue in question. Otherwise, though, it is a certitude of the math of advertising that reach times frequency equals gross impressions, so it’s difficult to get around the fact that lots of impressions with low frequencies will yield very high reaches.
What is important for advertisers to note, though, is that the reach of an ad network in total is a different thing from the reach of a campaign delivered by that network. That’s why we also do campaign-specific post-buy analysis work for many advertisers and agencies.
I would also point out that frequency capping is plagued with cookie deletion problems. The capping is per cookie, not per person. That’s another reason we see the importance of panels in being able to provide an accurate measure of a campaign’s actual reach and frequency.
Do ad exchanges present special challenges in data analysis?
Yes, naturally. Fortunately, there are solutions as well. Some ad exchanges are extremely proactive about combining the ad platform with a data platform, and in building analytics around both.
At comScore, as we migrate to Media Metrix 360, we are incorporating not just beaconing, but also a tagging protocol. This allows us to track delivery across websites by campaign, and to report on any aggregation of inventory in the fashion in which it is sold. We do a lot of work with ad exchanges and ad networks, and some of them are real power users of our offerings.
Are you seeing any good solutions for media buyers interested in better attribution models for their online buys?
The industry is learning that it is important not to attribute the last exposure with full credit for the action (click, conversion etc.) I think Atlas has done some nice work in this arena, as has comScore and others.
At comScore we’ve seen that, for example, the majority of the value of search is latent -occurring either in a subsequent user session or offline entirely. We’ve measured that 16% of the lift in sales from search may be observed within-session; another 21% occurs in a subsequent online session (making attribution tricky), and fully 63% occurs offline (trickier still .) We’ve used all sorts of tools to get at the true impact and effectiveness of advertising, including tracking eCommerce, integrating our data with third party sales data, and doing ad effectiveness research for times when the goal of the campaign is branding-oriented.
Indeed we have observed, and I personally continue to believe, that a significant portion of the value of online advertising exists in branding, and wholly beyond the click. This suggests, I think, continued creativity around new branding and advertising formats and platforms. But it also means that attribution modeling is more complex, because different impressions and campaigns have different objectives and should be evaluated accordingly. If I click on an ad for BMW, for example, how important is it that I’ve been exposed to brand building advertising for BMW, across media, my entire life? Literally thousands of touch points might have been involved in bringing me to that click.
What will real-time bidding’s (RTB) and demand-side optimization’s affect be on the analytics business?
I fear that the effect will be, good for analytics, bad for branding.
Broadly speaking, what are the measurement challenges that agencies are facing? How does this compare with publishers?
We’re hearing a couple of things from agencies. One, of course, is about quantifying the impact of advertising—effectiveness, results, ROI, branding versus direct response. There are so many new platforms for advertising—mobile, social networking, place-based video—and each is fighting for a piece of an increasingly fragmented ad dollar. So agencies need to be able to justify to the client the value of every dollar spent.
The second thing we’re hearing is about cross-media; how does advertising work synergistically across platforms? This is an area in which we’ll likely be doing a lot of work over the next several years. Clients are asking, for example, how does the Internet fit in with my TV buy? How should I deploy online video? How do search and display advertising work together?
While some traditional media are experiencing precipitous declines—notably newspaper and radio—the news of TV’s death has been greatly exaggerated. Many large agencies still spend—and make—the majority of their money in TV. So for publishers looking to break into new and under-developed categories, probably better to sell digital as enhancing a TV campaign than replacing it.