The following is the first in a series of interviews conducted with OptiMine CTO Rob Cooley. In this series we will dive deep into one of the biggest challenges facing advertisers today; measuring the cross-channel value of their digital ads. Each installment will focus on a different aspect of the issue and include Rob's insight and analysis. In our first interview, we layed the foundation through a discussion of the importance of measuring cross-channel value. Rob also offered two ways advertisers can think about the problem--indirect marketing & direct marketing--both of which are rooted in traditional, offline, advertising strategies. To be sure, online advertising is a different beast, with characterisitics that fit in both paradigms and Rob explores both in this segment.
Q: How does OptiMine define cross-channel value?
A: What we’re really after is cross-ad value.
Early funnel ads such as display, Facebook, and other social ads do not generate many direct conversions. On the rare occasions when there is a click that results in a conversion, measuring that value is very easy. What is more difficult to measure is the effect upper funnel ad impressions have on the conversion performance of lower funnel ads such as search and retargeting.
But the impact of one on another does not always happen across channels. In fact, intra-channel effects are not unheard of, so what we’re really measuring is cross-ad value.
Q: So are we talking about view-through conversions?
A: It’s analogous, but not the same. View through by definition double counts because the view through says I’m going to count every time this ad was viewed and then they bought somewhere else.
For example, a consumer views five ads before doing a search and clicking on an ad and making the purchase. In this case, the paid search ad is given credit for the conversion, but so are each of the five ads seen prior to the search that served the sixth. So a single purchase now shows up as six in the advertiser’s conversion data. This happens because each publisher is tracking conversions and reporting them to the advertiser. But this causes a couple of problems.
First, the advertiser doesn’t know to what extent they are double, triple or quadruple counting because you don’t know the exact length of the purchase path, but you know they are at least double counting. Second, publishers expect the advertisers to divide the conversion value. If the advertiser think’s they are quadruple counting they’ll give each view through a quarter of the value of a click through conversion.
What we’re doing is far more sophisticated. We are asking, “What is the incremental value of this view?” We’re not double, or triple, or quadruple counting because the value is anywhere between zero—meaning it had no effect whatsoever—all the way up to the value of the conversion. Because when you’ve sold a hundred things, you only sold a hundred things. If your tracking told you that you sold 500 things, you know that’s wrong.
Q: Are there advertisers who aren’t interested in measuring cross-channel value?
A: I have yet to find one who doesn’t care. I have met a lot who have gotten frustrated and given up. They’ve decided it’s a dream that will never be realized.
Q: Do people accept the numbers as good enough because everyone is working off the same flawed numbers?
A: You still have to take action off the bad data, but there is a range of actions you can take. Some advertisers ignore view throughs assigning 0% weight. Others think view throughs are important and assign 50%. Ultimately, each decision leads to very different behaviors.
The critical assumption of view through is that all ads behave the same; a view through of ad A is equivalent to a view through of ad B. We’ve found that to be demonstrably false. Some are very valuable for cross-channel effect and some are horrible.
If you’re selling women’s shoes and it’s [the ad] in a place where women shoe shoppers hang out, the cross-channel effect can be high. If it’s in a men’s health section, you’re probably not going to get a cross-channel effect. But both will be counted as view through impressions. What the cookie doesn’t know is the computer is shared by a husband and wife, so the publisher is serving women’s shoe ads to a man and giving them equal credit in the purchase path when they had no effect on the purchase.
Just because the ad is on the path, doesn’t mean it had any relevance to the purchase.
Q: Are today’s digital advertisers less inclined to accept that some online channels are primarily for branding since any of them can be used as a purchasing channel?
A: With a lot of display I think it’s the reverse, looking at display as branding, buying based on volume (CPM) not on value. Several large retailers spend to a budget because they can’t measure the value, and they are not going to try. But they can measure CPM, so they get as many impressions as possible for the budget.
Advertisers who try to do display based on performance are disappointed because their tracking tells them it has little value. And if you are all about performance, you either stop spending money in display, because your reporting tells you it isn’t worth anything, or you decide display is for branding and shouldn’t be tracked.
What we’re offering performance-based advertisers is an opportunity to get back in the pool. We’re saying there really is value in display and we’ll show you where it is. Every time we’ve done it we find the best ads are 10-50X better than the worst. View through assumes all views are created equal. What we’ve found when we done the modeling and taken action is that that is absolutely not true. There is 10, 50, 100 times difference between the best ad and the worst ad with respect to impressions. So if you assume they are all equal you make really bad decisions. Performance based advertisers can now identify the value and justify their spend.
Q: Are people who aren’t measuring better off than those who are attempting to measure?
A: Because the state of measurement is so bad, I actually think the ones that don’t try are better off.
The brand advertisers are treating this like indirect marketing, which it is. So I think they have the better paradigm. Whereas those who are trying to shoehorn it into the direct marketing paradigm are making things worse, because display advertising is not direct mail. And trying to pretend that it is direct mail leads to all sorts of really bad decisions.
People who aren’t measuring are closer to doing the right thing, because at least they are looking at it in terms of what audience should I get in front of. They are looking at it as branding, which it is, so they are aligned with the purpose of the ad; getting it in front of the appropriate audience for my brand.
The performance advertiser for display is shot-gunning stuff out there and taking the measurement they are getting at face value.
Q: Talk about the different ways of approaching the problem?
A: There are two paradigms in marketing. There’s direct where I know who I’m advertising to; direct mail, telephone, email. And then there’s indirect where I’m just trying to reach a particular audience; television, radio, newspaper, and billboards. Those are the two major paths.
Then the Web came along and confused people because it’s not 100 percent indirect, but it’s also not 100 percent direct. It is a hybrid. For example, paid search is indirect. You do not get to tell Google who you want to advertise to and Google will not tell you who the search results to.
The same is mostly true for display, but the promise of real-time-bidding is that it makes display more like direct advertising. That’s not exactly true, though. With RTB you know the cookie ID you put your and in front of, but you don’t know who. What the cookie ID does allow is for you to be far more precise in your targeting. The mistake advertisers make is treating that precision like direct advertising and measuring the value that way too.
What advertisers need to get their heads around is that the Web is half and half, direct and indirect. Where you can target specific cookie IDs, but you are still limited in what you know about that person behind that ID.
I can go direct advertising route where I assume the cookie ID is a unique identifier that I can track all the way through to the purchase, or if I don’t see that ID again I assume that individual did not purchase. And that’s the biggest problem with trying to map it to a direct marketing paradigm. We know that assumption is wrong a significant portion of the time.
Attribution (path analysis) vendors are setting this up as a direct response. They build binary response models predicting the probability of a cookie ID responding and then, if the ad was in the cookie ID’s path, they attribute some value to that ad if it bumps the probability of response.
Path analysis vendors are going to dramatically underestimate the value of the ads because a ton of the views are going to be associated with cookie ids that don’t convert. It’s not that the person didn’t convert, but it came through a different cookie id.
Shared devices is getting more rare, but multiple device usage is growing at high rates. They are going to underestimate the cross channel value of ads because they will look at it with respect to the cookies, see which converted, and they will be missing conversions because they are showing up on a different cookie ID.
Q: Are there times when direct is better than indirect?
A: t’s really about late funnel vs. early funnel. With a late funnel ads—search & retargeting for example—you don’t have to mess around with view throughs because the actual click to conversion rates are strong enough to justify spend.
The ads you want to watch for cross-channel effects are those with low click-through rates. In our experience, these are the ones that drive value further down the funnel. Ironically, they may have higher value cross-channel, but you don’t pay much because of the low CTR.
I would say path-based attribution works fine for late funnel ads. But if conversions are dominated by last click, you don’t need multi-click attribution to begin with. Where attribution falls apart isn’t because of the last ad, it’s because of the five ads leading up to the click. And those ads are, almost by definition, early funnel ads. If you’re struggling to understand the value of your early funnel ads, attribution will not work.