Saturday, April 19, 2008

Online advertising: tutorials

The Law & Business of Online Advertising: Berkeley/Santa Clara

Tutorial on Online Advertising Technology

Kim Howell, Microsoft privacy expert

Among other things, Howell emphasized that 3d-party ad networks enable smaller publishers to monetize their sites (e.g., bloggers) and smaller businesses to participate in ad campaigns across multiple platforms. Intermediaries need to report to publishers and advertisers how many ads they served to figure out how much to pay, which means they record impressions and users’ IP addresses. If the advertisers only want to reach net-new users, then they need to get more information: thus they add a cookie to the user’s browser.

Financial/auto website inventory sells out very quickly. The advertiser may want to target users with specific interest, but use a general-interest site. The ad network server can check the cookies and see who’s been to financial websites and show them American Express ads. It extends the inventory of “people with financial interests” to nonfinancial publishers. This is important for general-interest websites, Yahoo!Mail, etc.

What if the advertiser wants to show different ads to men and women? The ad network server can have IP address, cookie, ads served, time/date of visit, and registration—if the user has registered with a site for various reasons, such as getting content. The user provides gender, age, income, etc. The publisher passes on that information to the ad server. The ad server can correlate the user account with the cookies and serve targeted ads. (You can find what servers are being called by choosing “view source” and searching for http.)

What if the advertiser wants to show an ad to a known buyer? E.g., people who abandoned a shopping cart or people who’ve made a past purchase. The advertiser can place a web beacon on a page it controls; the advertiser, e.g., Wal-Mart, can choose what information about the user will be sent to the ad server. So when the user is on another site, the ad server can check the cookie ID and see the user is a Wal-Mart customer and show her a Wal-Mart ad. The advertiser’s privacy policy will (or should) explain how the data is being used.

Data that could be in an ad profile: IP address, unique cookie ID, all publisher sites visited where ad server serves ads, all advertiser sites visited where there’s a web beacon, any publisher/advertiser info passed to the ad server, and geographic info derived from reverse IP lookup.

Question: is there any way around cookies? What if a user deletes/disables? Answer: there are new technologies in the offing, but the major technologies rely on them. Cookie management creates “orphan” information that exists, but is unusable.

Q: Are there ways to discourage cookie deletion? A: Other than education, not really. They can say the website won’t work as well without cookies, which is true. But other than that, no.

Q: How is this connected to personally identifying information? A: The publisher/advertiser can pass info to the ad server—the ad server, e.g. Doubleclick, doesn’t have any direct relationship with customers. It’s against National Advertising Initiative (NAI) principles to use web beacons to pass personally identifiable information. So legitimate/large players don’t do this.

What is the customer’s choice? (1) To clear/block cookies; you can even create custom block lists. Those two options conflict with each other, though. If you set your options to accept only first-party cookies, most sites should work for you. (2) To set opt-out cookies with NAI members. If you choose (2), the ad server knows you’ve opted out of targeted ads and won’t show you one. When you opt out, the data is still collected in order for publishers and advertisers to know which ads were served where.

Q: If an ad network were subpoenaed, could it turn over your information? A: Assuming they’ve retained the data, yes, though they should have a retention policy keyed to business purpose.

Comment from audience: if you have a dynamic IP address, as is common in the US, your IP address won’t be tied to you absent that cookie. A: If you want to track users, going to the ISP directly would be a lot more valuable.

Tutorial on the Business of Online Advertising

Hal Varian, Google

There are only a few hundred major players in the brand advertising space: big advertisers, big publishers.

Advertiser/agency contacts publisher, negotiating a price starting from a rate card—nobody pays retail. Aspects of negotiation: cost per thousand impressions $1-20; $10 CPM=1 cent per impression. Advertising requires scale, which is why there are so few major players.


The process: Advertiser puts creative on ad side server; publisher programs publication side server according to conditions (# of repeats, targeted info); ad is fetched and displayed according to terms; payment is made. Cost of serving is 2-5 cents per CPM, paid to ad-side server. Any pub server can talk to any ad server and vice versa. Doubleclick, Aquantive, Real Media etc. just provide serving technology, not creatives/ad pricing; some large publishers and agencies have their own pub servers.

Ad networks: about 300; an aggregator that reduces the need for separate contracts, separate sales forces. Network, like Google AdSense, sells ads to publishers on behalf of a number of advertisers. There are quality issues, though—the ad might not be as good. Valuable for small publishers and remnant inventory of large publishers when they haven’t sold out all the ad space.

Ad exchanges: a marketplace where buyers/sellers can meet. It’s a competitive market with commissions, like a financial market, but participants have less control over where and how ads appear.

Why cookies? Because http is “stateless”—it has no memory. The web would fall apart without a way to maintain memory. A common use of cookies in ads: frequency capping.

Website cookies are sometimes also called 1st-party cookies, used by publisher to sustain conversation with user. 3d party cookies are different—ad-side or pub-side.

Contextual ads, e.g., AdSense: advertiser uploads text creative, which could be the same as for a search ad. Publisher puts a bit of javascript on page. Users are shown ads contextually related to page content. CPC pricing: every time user clicks on ad, revenue is shared at a fixed rate between the ad network operator and the publisher; CPC price is set by competitive auction. Variations: image ads; CPM ads. Major plus: reduced transactions costs. It’s popular with small publishers and used as filler by large ones.

Search ads: On Google, advertisers can decide how to partition ads—whether they only want ads to show on Google, or also on search engines with which Google has agreements; geo targeting; day splitting; other options.


There are millions of keywords which can’t be priced separately. Instead, there’s an auction: each ad submits creative, keywords, bid. Ads are ranked by bid x click quality, as estimated by Google (the “secret sauce”). The best-scoring ads get the most prominent positions. The reserve price (the price you pay if you’re the only bidder/the only ad on the page) depends on ad quality—the higher the quality, the lower the reserve. Really good ads get promoted to the top of the page. Really bad ads get disabled and not shown. You’ve got to keep users as well as advertisers happy to keep users from developing “ad blindness.”

Now watch Varian set up some competition rhetoric: Publishers want max revenue. Effectively, all forms of ads are perfect substitutes. In practice, they often have a banner ad, some image ads, and some contextual ads. There’s no exclusion; on a large publisher, there can be 3 or more companies serving different types of ads, e.g. on the NYT page, plus what they sell directly. The switching costs are relatively low: all you have to do is change the html code on the page.

Another intermediary: SEOs and SEMs, which help manage search engine campaigns; he thinks that in the long run they’ll be incorporated into ad agencies.

Economic forces: pricing power. Prices are set by advertisers in competitive auction for search and contextual ads; set by competitive negotiation for display ads and ad serving.

Switching costs: For publishers, low: change code. For advertisers, they already use multiple channels, and switching is easy, or even not required: 95% of Google’s big advertisers also advertise on Yahoo! Users read a variety of content, and find it easy to switch search. There’s no exclusivity anywhere.


Supply-side economies of scale, or, what about entry? There are several players with appropriate scale, large data centers. What’s important is not tech capability but know-how.

Demand-side economies of scale/network effects. Varian’s never heard a convincing story about search engine use.

Q: Third-party trademarks?

A: His understanding is that the acceptability of this varies by country. Nobody’s quite sure what the policy is. As an economist, he likes competition! Google has a policy against misleading ads, but more choice is good.

1 comment:

Greg said...

Thank you for your post. I was not able to attend the conference, but I enjoyed your summation.

I started my own blog recently and discussed on it the FTC's proposed principles for online behavior advertising. If you have a chance, please check it out, My Two Pennies Worth.

I would appreciate your professional opinion since I just started it a few weeks ago. In addition, I am going to add you to my blog roll if that is alright.

Regards,

Greg
http://gregharding.wordpress.com/