Advertisers often use multiple publishers for their online campaigns, but may not use the best metrics to decide which ones they should compensate. Should companies compensate the publisher who showed the last ad to a consumer before a purchase? Or, should they pay publishers every time they show the ads to prospective consumers?
Research by Wharton marketing professor Ron Berman finds that the “last touch” or “last click” method advertisers typically employ to compensate publishers is the wrong way to go about it. Berman’s research shows that the “last click” method entails a moral hazard in driving “adverse selection,” where publishers show ads to consumers who would buy the product anyway. The resulting competition between publishers “creates a race to try and cheat the advertiser out of money,” he says. It also encourages free riding by some publishers on the efforts of others, just so that they can collect the credit and compensation for showing the last ad to the consumer making the purchase.
Berman advocates using the compensation metric that rewards “effort” over “performance.” In other words, “paying for exposures is more efficient and more profitable for the advertisers than paying a commission on sales.” In his research paper titled, “Beyond the Last Touch: Attribution in Online Advertising,” Berman used data from a large-scale online campaign. In an interview with Knowledge@Wharton, he discusses the major takeaways of his research, the findings that surprised him and what he plans to study next.
An edited transcript appears below.
Scope of the research project:
My research deals with online advertising attribution. Let’s suppose you went on a brand’s website – say, Gap — put into a shopping cart a shirt, [but] you didn’t buy the shirt, and you didn’t continue surfing the Internet. Typically, you will start seeing ads for the same shirt [repeatedly] on every website. Now the question from the advertiser’s perspective is, “If, in the end, the consumer did buy the shirt, how do you measure the effectiveness of each one of those ad exposures the consumer had?” This is the “attribution problem.”
Most advertisers today use something called “Last Touch Attribution” or “Last Click Attribution.” If I, as a consumer, saw 17 different ads on 17 different websites and purchased a product, the last ad I saw before purchasing the product [would] get all of the credit for the sale. My research [attempts] to determine [if] this method is efficient or not, and its impact on different types of advertisers and publishers in the online marketplace.
In my research, I ask: Is this Last Click method, which is used by the majority of advertisers, efficient? Why do they use this method? It is counter-intuitive. Why would they look only at the last ad that the person sees? I compare two types of compensation scales that are used online. One is paying for every experience of an ad being seen, and [the other] is paying just for every sale, like a commission on a sale that the publisher gets from the advertiser.
Exposure vs. sales:
My key takeaway is that paying for exposures is more efficient and more profitable for the advertisers than paying a commission on sales. Most people think [that] if you compensate a salesperson, you should pay a commission only when they make sales, and not when they show up for work. But if you compensate a publisher, say Facebook, you can pay Facebook for just showing the ads, or [for] generating sales.
If you have more than one publisher — say, Facebook and Google are competing — that [will] be free riding by one of them on the other. Because of the Last Click method, they are going to compete [with each other], and one of them [will] claim they are very efficient, although maybe they just showed the last ad.
The second [takeaway] is that the Last Click method creates competition between publishers that [in turn] creates a race to try and cheat the advertiser out of money. What would you do if I told you that the last ad you show gets the commission? You know [of] a consumer [that] comes to your website, and there is a high probability the consumer [would] buy the product. There is another consumer that comes to your website, [but] you don’t know if that consumer [would] buy the product. To which consumer will you show the ads?
“Paying for exposures is more efficient and more profitable for the advertisers than paying a commission on sales.”
You will show the ads to the consumer that is going to buy anyway. You just bomb the consumer with ads to try to be the last one. So, the other key takeaway is that, when you’re using the Last Click or Last Touch methods, all of the measurements and all of the metrics you’re getting [would] be maximized. You [would] get a lot of sales through [many] channels and publishers. But maybe they didn’t cause any of those sales that just happened to be the last ones.
The first conclusion that surprised me was that paying for impressions or exposures is more profitable than paying commissions Twitter . If you ask any advertiser online, they [would] say, “I want to pay only commission. I want to pay only when sales are being generated.” This was my intuition [as well] when I started this research. When I tell this to managers and executives, they don’t believe me, although I can show them the math and show them it just works.
The other surprising finding was that Last Click or Last Touch is not necessarily a bad method. Under some conditions, it increases the profits of the advertisers. But these conditions are limited. If you ask advertisers, “Why do you use Last Click and Last Touch?” typically they say it’s the best method they’ve known so far. But I can show that, if it happens that your campaign has these specific conditions, it’s better to use Last Touch. When the first ad has the biggest impact … and not that [much with] the others, and consumers don’t visit the website too often, then the Last Click method is very efficient.
[Another] takeaway is, realizing that whatever attribution methods you use, over time, the publishers will learn how to maximize this metric. So, even if you told the publishers, “I’m going to compensate you according to who runs the fastest,” someone would be the fastest runner, and they’re going to [use] the best-maximized method.
The idea is to sit down with the publishers and with the company you’re using for doing the attribution, and think, “Does this model that they’re using make sense to my consumers?” Also, design an experiment and ask, “What would happen [in] using an attribution method, and what would happen without using it?” You can’t just move to a new method, see all of the metrics maximized and say, “Oh, I just made more money” — because, probably [that] isn’t correct.
“The Last Click method creates competition between publishers that [in turn] creates a race to try and cheat the advertiser out of money.”
One interesting misperception [relates to] ads called re-targeting ads. These [are] very specific ads that you only start seeing after you put a product in the shopping bag. [People] in the industry believe these ads are very useful. They are good at reminding the consumer to come back to, say, Gap’s website, and purchase the shirt that the consumer was wondering whether [or not] to purchase.
My research shows that most of these re-targeting ads are probably inefficient and have no impact on consumers. They are shown last before a purchase because, if a consumer went to a website and put something in the shopping cart, there is a much higher probability the consumer [will] buy the product anyway. If you [showed that] consumer [your] ads, compared to consumers who didn’t just visit the website and didn’t put anything in their shopping cart, it would appear you have a huge increase in sales [from] those consumers you [showed] ads to. The misperception is … you’re going to think that re-targeting is a very useful channel. Typically, [re-targeting ads are] not that good.
My method is unique in that I use game theory. I did an analytical and theoretical analysis of this problem, and only then did I test it on the data. All of the other research on this topic [uses] an empirical approach [where the researchers] took a set of data, ran algorithms on it, and tried to figure out which publisher is better.
What I ask is, “What is the market using in measurement and compensation? What would be the consequences? Can we observe that in data?” I found that a lot of the empirical applications and measurements are incorrect. They assume that the market is efficient and behaving correctly.
But if you do the theoretical analysis, you can figure out from the data if the market is inefficient or efficient. The publishers are basically showing ads to consumers that will buy anyway. This is called “adverse selection” or, essentially, trying to cheat the advertiser. Other empirical applications cannot differentiate between showing ads to consumers who would buy anyway, and you stick them with ads to get the commission.
“The publishers are basically showing ads to consumers that they will buy anyway.”
After working on this research project, I started getting a lot of communication from companies saying, “Hey Ron, there is this approach that you’re suggesting. But there are at least a hundred companies that claim their attribution product is better, or is the best. How do we compare? How can we even tell which one is better, et cetera?”
One idea I’m trying to push — probably with a few companies to experiment with — is to try to build a system where you generate advertising data, try different attribution methods, and see which one matches a company’s type of customer.
This is interesting [because] most attribution companies don’t publicize [their work]. They’re not transparent about their algorithms. They just say, “We have the best algorithm, and if you give us data, we can give you the results — but we won’t really tell you what’s going on.”
This approach, where you test through many of them, and run a competition between all of them, will hopefully tell us which method works better or not so that we can understand this process much better.