Panel One: The New World of Digital Advertising: Technologies and Business Models
Moderator: Emily Bazelon, Slate Magazine and Yale Law School
Scott Spencer, Google Inc.
Publishers who work with Google fundamentally want to provide the content free online: there is a tradeoff between ads and the price of a magazine, on or offline. What has changed: in the past, ads were bought based on the demographics of the readers. Newest change: buying process. You can now buy the audience independent of the publication, and that’s more efficient. If someone goes to a car site, I can target them with a car ad later, or if they abandon a shopping cart, I can remarket the items. Increases publisher revenue/advertiser’s effectiveness.
Drugstore purchases: items I buy may be very personal, but I let CVS scan my card and use my info any way it wants. Same with credit cards. Many offline parallels, but online is just more obvious.
Michael Blum, Quantcast
Quantcast as case study. The only way that people have tried to reach an audience beyond the existing audience is by using demographics. But demographics is always a guess wrong in both directions. Target market is 50% right, but we don’t know which 50%. So: take existing customers and find how they behave; then find people online who behave like that, and target those folks.
Jesse Pesta, Wall Street Journal
WSJ did a project: What They Know. This is about more than advertising. The ability to crunch numbers—mix databases including those that have existed for many years. Insurers are experimenting with the data pools and finding they can be as revelatory as a blood test.
Q: how to back up promises to consumers?
Blum: there are technological answers to problems like security breaches. There is still a distinction between personally identifiable and not identifiable information; scariest involve the former. Quantcast collects 2 quadrillion pieces of info a month, in a room of servers that no one at Quantcast could open and understand. It’s a big pattern recognition machine. It knows that Scott went online and searched for boots, but it doesn’t know what boots are or who Scott is.
Spencer: There’s a difference between speeding and having a law against speeding. Going 3 mph over speed limit is different than 100 mph over. The key is to make sure that things that are bad are curtailed, not the technology. Can a credit card company sell my data to an insurer? He’s not sure. It’s more transparent online.
Pesta: My name is personally identifiable information. If all you know is that I shopped for cowboy boots, that’s not personally identifiable. But these things meet in the database. If you know gender, age, zip code, shopping preferences—that probably does identify me uniquely. So is that personally identifiable information? Substitute “has diabetes” for “shops for cowboy boots.” What then?
Blum: personal health/other sensitive information should be treated differently, there’s a consensus on that.
Bazelon: so what happens once that information is collected to prevent its use to deny you health insurance?
Blum: maybe it shouldn’t be collected in the same way. EU has a rule: opt in before even collecting it.
Bazelon: with do not track, are we relying on companies not to collect such info? How do we ensure it’s not collected?
Spencer: many questions about what tracking means, etc. Useful to have some bright lines. We make an opt-out persistent for the user.
Bazelon: what are the unclear lines?
Spencer: resell data: if you’re comfortable with a first party seller using the data, why not allow resale?
Bazelon: is this a moment where we need government regulation?
Spencer: no official answer, but both government and self-regulation could work.
Blum: The NAI, mentioned above, represents advertisers who are very interested in showing that consumer choice is a top priority—not sitting around trying to avoid regulation, but trying to find something that is consistent, easy for consumers to understand, and could be adopted by market leaders as well as smaller companies.
Joe Touro, Annenberg School: heard rumors that Google is starting to desilo its data: Gmail v. contextual marketing etc.
Spencer: he doesn’t know. We take data privacy very seriously; consumers can look at what’s stored and can opt out, and we try to make the opt out work across browsers.
Linda Greenhouse: how concerned should she be that, after researching individual health insurance, ever since then, 5-10/day she gets invitations to buy health insurance? That, after researching odd health conditions, she gets ads related to those? Who thinks of her as an uninsured (Lucinda Williams-loving, per Amazon anecdote), multiply diseased person?
Pesta: Gets to the “creepy” place of reaction. Amazon is different because you have a relationship with Amazon, which you agreed to when you started buying from it. Amazon is paying attention within its own ecosystem, just like the guy behind the counter at the local bookstore. Other examples get closer to the creepy factor because you don’t know who surmised the existence of your status/diseases. There’s also collaborative marketing on Amazon (users like her).
Touro: but underneath the hood, Amazon is doing the same sort of thing that other sites do.
Pesta: but you did agree to the first-party relationship, like with the local donut shop.
Blum: it’s easy to ask “who’s watching” or say “it’s nobody’s business,” which is part of the creepiness factor, without understanding the technology. But in a lot of this tech there aren’t other people who know these things, and “we” aren’t being watched. No humans are involved. It’s just pattern recognition.
Bazelon suggests that this isn’t making anyone feel better.
Blum: but it doesn’t understand that boots function like shoes.
Greenhouse: but a human will make use of the knowledge.
Blum: not in many of these cases, though consumers should be given choices.
Greenhouse: but they could find out.
Blum: mostly they’re just interested in doing it by automation.
Pesta: follow the money—there might not be money in aspects of this business. Example from WSJ: man in business of online ads/tracking, and he said, in regards to whether he knew a real name—“no, and I have no interest, because it would be too costly and there’s no money in it.” But on the other hand, another company is associating people’s real names/addresses with these databases; they thought they’d found a way to make money.
Blum: the examples you gave are creepy in part because they’re health information, which should be treated differently.
Lee Tien, EFF: The right to forget: users being able to expunge data held about them. How valuable are those data as they age? Consumer advocates want to know: are there places where natural degradation of value becomes clear? A day, a week, 30 days? Mitigate potential costs to privacy from civil litigants/government access while not necessarily damaging advertising uses of those data.
Spencer: we try to limit retention, but he doesn’t know of any study of lifespan.
My thoughts, listening to this overall discussion: the fantasy of perfect control (if we just add enough tech, we will control everything we want to control) has shifted from copyright owners to advertisers, which isn’t a great surprise.
Blum: could distill data over time and keep only the most important parts for longer.
Tien: but what are the times?
Blum: 13 months—hard to see greater value.
Bazelon: 180 days for stored email is generally thought to be the line past which it should be harder to subpoena.
Amy Kapczyinski, Berkeley: Aren’t we worried that Greenhouse is going to get charged more for health insurance? Either they’ll offer you a higher price, or they won’t put you in the “discount” group. Some of this tracking is about price discrimination, notoriously unpopular with consumers. Do we mean that Google shouldn’t track diabetes searches when we say sensitive information is different, or what?
Spencer: many aspects to this. I search for a computer and I’m offered 20% off if I go to a particular site. That’s great price discrimination for me. There are positive and negative use cases for the tech. Self-regulation attends separately to sensitive data, avoiding storage and use.
Kapczyinski: how do we operationalize that for Google? Do you store those searches?
Spencer: we don’t store health related information, though it may work differently for things like Gmail—he only works in one area. We don’t use sensitive information in behavioral ads. (Though they do in contextual.)
Bazelon: per Greenhouse, clearly somebody is.
Pesta: nontraditional advertising: banks have started using flavors/varieties of this data to decide what sort of potential customer you might be, in realtime. Capable of offering the bank a quick general judgment, and the bank can then decide which credit card offers to show you. Is that price discrimination?
Wendy Selzer, Princeton: we’ve heard about opacity to the end user and hearing that advertisers don’t want to disclose algorithms/get too close—incentives are against good notice to user who might be creeped out in the moment. How could we shape incentives to require notice given through the tech closer to the way consumers are using it? What might that look like?
Spencer: we believe in greatest possible transparency. You can go see what data we have about you. You can adjust them. Also, every ad, we say this is targeting and you can click on it. Almost nobody does, but they can, so it’s transparent; we can’t force them to click.
Blum: why don’t people opt out? Maybe they aren’t worried.
Bazelon: what happens if you opt out? Is it forever?
Blum: allows you to opt out once or forever.
Spencer: you can adjust your profile or you can opt out, and you can plug it into the browser so that it carries across (I thought this was just Chrome). (As it turns out, Google thinks I’m a man for advertising purposes. Also it thinks I’m really interested in gymnastics. OK then. Given how much time I spend reading fan fiction—and not following gymnastics—I’m oddly disappointed by Google’s profiling abilities. On the other hand, were Google to de-silo its information by integrating Gmail, that bad profile would be much more accurate, and I don’t think I would like that.)
Pesta: we were aware of the ad bargain, but we weren’t aware of the personal information collection bargain we were making. Tools should make clear what bargain we are in fact striking.
Bazelon discussed having a Twitter impersonator. Even if you’re not a journalist, you’re probably participating in these social networks where you’re putting a lot of personal information out there. Maybe it just seems like a fact of life to her: you have to be careful and hope you don’t make a mistake you’ll be stuck with forever. She’s tempted by a right to forget, especially for teens/young people who aren’t thinking about the import of information they put online—they deserve more of a safe harbor than they get.
Pesta: He’s not the Jesse Pesta who hired a hit man to kill his family, if you Google him.
Hoofnagle: Costs of regulation? OBA looks a bit like a Ponzi scheme to him. Everyone is talking it up, but are the numbers revolutionary or incremental? Is there anyone doing good work to show the revenue gain over contextual models—which supported the web until this time?
Spencer: The cost of self-regulation is not necessarily in the checks we pay to the NAI. The real cost is that we look at the tenets of the initiatives when we’re developing a product, building in an opt-out. That has a much bigger costs, because it takes away time, but we’re committed to notice, choice, and opt-out. As to whether OBA works: studies he’s seen indicates that it does—the ability to use audience data does have the ability to get an offer in front of a consumer that has more of an impact (he said “more impactful,” but I refuse). Consumers do like ads that are relevant. Advertisers who can target their ads are willing to spend more of their budget doing so, which increases money paid to publishers.
Q: NAI has members that include Google and Quantcast, but some members will also target people based on their “social graph,” and others will target not just based on what disease you have but what stage you’re at. Some will take “I opt out” to mean “I opt out of receiving this ad via tracking.” So what kind of self-regulation/enforcement are we talking about?
Blum: you end up with a nugget of agreement, and then individual practices. Sensitive information requires opt-in consent, so maybe the company has that, but if it doesn’t, then he wants to hear about it.
Q: so self-regulation is a floor, but will it give consumers what they’re looking for?
Blum: the aspiration is to do that, and we have an incentive for members to comply because it doesn’t work unless all members do comply.
Daniel Kreiss, ISP: Maybe consumers need to get more comfortable paying for journalistic content, per NYT, if the bargain isn’t working. Especially if publications are no longer the go-to site for selling audiences to advertisers, and the intermediaries are now in charge of matching.
Spencer: There is a tradeoff: free content for ads. The more efficient that is, the more will be able to survive. Advertisers still want a strong, safe site environment where consumers will have a good contextual experience. They also want the ability to mine the relevant data.
Pesta: WSJ has a pay wall: the key issue is striking a bargain. Paying can be one bargain. (My questions: What kind of advertising does the WSJ do behind the pay wall? Is it contextual or behavioral? That is, can I get the privacy benefit of my bargain?) Newspapers made a blunder: after training people to buy their product for 100 years, they stopped asking people to pay. People don’t want to pay for the NYT because they haven’t had to for years.
Bazelon: in fact, a few big news organizations play a huge newsgathering role—the NYT might be in a category where the model could work.
Q: privacy concerns associated with a paywall? If the internet goes to an authenticated model, isn’t that just as privacy-risky?
Pesta: you enter into a deal with WSJ; the WSJ has your credit card number and your name, so there is a database. Traditionally that’s one of the most valuable assets of a newspaper: people willing to trade money for the product. Long tradition of renting the mailing list, though he doesn’t know if the WSJ sells its list.
Moderator: C.W. Anderson, Yale ISP
Berelson, What Missing the Newspaper Means—studied responses to a newspaper strike in NYC. People initially said they missed knowledge of the public world. What they really missed: the weather report, the radio listings, and the ads. See also this piece.
Jason Kelly, AdMeld Inc.
His company: has 500+ publisher clients, from Fox to Pandora. Now there are demand side publishers and supply side publishers in between the advertiser and the publisher. (The advertiser is the demand side.) So you have a four-party chain. At least. There’s a lot of fragmentation. And a lot of money out there. Advertisers want to come online in a way that’s brand-safe and contextually relevant. (I just saw a mention that no one wants to advertise on 4chan.)
Realtime bidding on ads: will more than double in 2011. $1.4 billion in online ads (real and nonrealtime). “Thanks to the internet and digital technology, agencies are finding that the realization of their clients’ ultimate fantasy—the ability to customize a specific message to a specific person at a specific moment—is within their grasp.” – 2009 article. (Speak of the fantasy of perfect control!) Some advertisers are going to 100% targeted ads. Many major media companies are trying to cut the middleman to sell ads online—a private exchange. Google and Yahoo! have them, and AdMeld has one allowing specific publishers to curate their own supply directed to advertisers—gives the advertisers more protection.
David Ambrose, Scoop St.
Daily deals: group buying. We do neighborhoods: Manhattan, Brooklyn, soon part of Connecticut. Realtime bidding is neat, but a small business really just wants the customer to go through the door. So a deal is a way to do that. There’s a lot of education left for small businesses—a different demographic than we’ve been talking about. It’s also an older form of marketing: through email, not Facebook or other sites. Allows 2-way communication with consumers & small businesses. Lots of businesses have no idea what AdWords is or what it can and can’t do. They might hire a consultant to put that together. AdWords is really complicated. Daily deals seem like a good way to acquire new customers.
Many businesses ask: can you help me get the customer to return? Advertising as a dialogue.
Kate Kaye, ClickZ
Political advertising: done online to connect with donors. Creating a supporter base is really creating a database. They’ll ask for money, they’ll ask for help with your friends. Facebook ads are relatively cheap and can help later on with your lists.
Special election for Ted Kennedy’s seat: Coakley wasn’t doing a heck of a lot until a few days before the election. At that point, the only thing to do is get out the vote advertising. Can’t buy ads on CNN over the weekend. So they used self-serve platforms like Google and Facebook. They couldn’t do issue-based advertising; couldn’t use ads to counteract negatives. That has to be during the campaign when people are deciding who to vote for. Coakley’s campaign: a solid lead eroded by a great campaign that built momentum in the last month. Online ads alone wouldn’t have saved her, but it helps a traditional campaign.
SEIU was running anti-Scott Brown ads on BostonHerald.com; Brown’s campaign manager called up and said that he wanted to own the website. Made a direct buy a few days before the election. Rather than doing get out the vote, they used all 6 ad spaces on the homepage for days before election day. Many were video ads. Brown targeted 10 specific districts for GOTV and asked people to volunteer with the campaign. Even Google text ads were targeted in those districts. Their MO was translating what happened online to the real world. They sent text alerts to people whenever Brown or Coakley were on the radio—told them to call and ask a question about X. Brown’s spending was 5x that of Coakley on online ads.
Signing a petition online: she’s not saying it doesn’t help with the specific issue, but what it’s really about is recruiting you and finding you later. 2012 will be a real year for video ads. Persuasion message coupled with a clickthrough where you can learn more about the campaign/donate. A call to action, which can’t work as well on TV.
Anderson: raising money is about running TV ads. So is that the point of running internet ads to raise money?
Kaye: most campaigns spend maybe 5% on online advertising; most of the budget is to run TV and print ads. Given its cost and the audience they need to reach, they can’t use the TV budget—they could make awesome videos and websites, but in terms of online ad buys there’s only so much you can spend that’s valuable. They don’t have too much money online, but down the road they might. Online ad people have been sitting at the kids’ table and are just starting to get in on the talking points/daily calls. 5% will be the norm for a while.
Q: would Obama be president without the internet?
Kaye: hard to answer, but online organizing was crucial and revered. They did tons of interesting stuff and had a lot of money to play with. They spent $20 million online, which was nothing compared to print/TV, but way more than McCain spent.
Gangadharan, ISP: Does the daily deal constrain what small businesses you engage? A piano tuning business isn’t going to benefit so much.
Ambrose: true, some businesses work really well for this model—waxing at salons/manicures and pedictures—and others less so. Could consider packaging piano teacher with romantic dinner, though. See a lot of restaurants/spas/massages. Percent off deals lead to impulse purchases. Small businesses are becoming savvier: they want to cap the number of people, they want to do specific days; they’ve just introduced personalized targeting.
Kelly: we don’t buy/sell ads ourselves. We’re a service provider for media companies for their inventory sales.
Anderson: paper dollars to digital dimes: people in the news industry don’t understand exactly why that happened. Is it better tracking (efficiency means lower prices), greater inventory, what?
Kelly: was the predigital cost structure the right one? No doubt that the number of platforms has changed for reaching audiences. It is clearly moving from dollars, maybe to quarters. Look at TV consumption—distracted attention means people have a laptop open when they’re watching TV. Is TV advertising still effective? Spending isn’t going down. The ability to reach you while you’re watching the Superbowl having a conversation on Twitter or Facebook is a benefit, though. Advertisers can target all ad capability surrounding that event. It’s not a 1:1 tradeoff.
Anderson: a big switch in power.
Kelly: that’s what publishers have face when they move from an environment in which a NYT salesperson says to Ford “you’ll be on the front page of the NYT”—advertiser values knowing where that content is. Now, supply exceeds demand. NYT can still command a premium, but $1 is different from $20.
Q: publishers are then told that they’ll make more money if they supply their data; this gives up on publisher autonomy and encourages content farming. (He clarified that he meant that this drove the intrusions into privacy/massive data gathering that drive the presentations at this conference.)
Kaye: she agrees there are way too many low-level pages lowering prices. But data exchange isn’t totally a negative for publishers.
Kelly: the advertiser wants the targeting. The publisher has to decide whether to provide it. Some media companies want to have their own tech so they don’t have intermediaries in the middle, whether those are ad networks or data companies. This is an attempt to recoup some of the lost value. In response to a question, Ambrose discussed the idea of TV everywhere: if someone who pays for Time Warner can authenticate his identity, why can’t he watch TV everywhere? Another possible model.
Ambrose: if you can control the channel (be Time Warner) that’s definitely a promising model.
Q: the idea is “we represent a certain kind of subscriber, so you don’t need to know more data about them.” With Scoop St., the key information is locational—is that the key piece of data? How much additional value does additional information provide? Even if you want perfect price discrimination it might be too hard to get there.
Kelly: buyer may not know what site the ad is going to; advertisers are willing to pay to know the context. Then you get beyond that: what other info is valuable? We’ve seen everything from 30-300% return on knowing additional information about the audience. Everyone in this room is a publisher. Retailers like Amazon and Wal-Mart are also publishers—they have content, they have audiences, they have data they could potentially package. That data appended to inventory would be 3x-10x more valuable to the advertiser.
Ambrose: small business perspective is: I want to buy an ad on the NYT—cachet of the name was important to them. For our space it’s too early to tell what that key piece of info will be, though location is obviously very important.
Anderson: it’s less important that data be accurate than that everyone (ad buyers, publishers) agree that they seem accurate. Down the road: fights over accuracy.
Kaye: already happening! Publishers can’t stand Nielsen, because Nielsen doesn’t agree with data in publishers’ log files.