Profiles In Courage

Advertisers and media companies have long salivated over the gold-mine marketing potential of the Internet. To many marketers, the Internet is a medium where consumers can step onto their showroom floors with detailed purchase histories, credit records, and demographic data taped to their foreheads.

To this end, the online advertising industry has been experimenting with richer models of user data for improved ad targeting. For example, Engage AudienceNet promises the ability to deliver users to sites with a dossier of personal content interests. Thus for advertising purposes, Engage could tell a site that an incoming user scores a .84 on football, .68 on comedy movies, .55 on rental cars, and .77 on murder mysteries – and all without necessarily identifying the user as O.J. Simpson.

Privacy, of course, has always been the fly in the profiling ointment. Any effort to trace online activity back to a single source – from (customer profiling standard proposals to the Pentium III chip identifier debacle – must at least pretend to place consumer privacy concerns first and foremost if it’s to have any chance of survival.

If you ask some of the online advertising wonks, such as the Network Advertising Initiative (NAI), most consumers don’t mind walking into stores with things taped to their foreheads… particularly if it covers their faces.

But then people who dress up their dogs in sunglasses and Hawaiian shirts for kicks somehow always neglect to ask the dog’s opinion. Ask consumers about profiling, and you’ll probably get a different answer. Despite assurances of a “double-blind” information exchange, last week privacy advocates likened behavioral profiling to Orwellian subliminal advertising of the 1950s.

Music And Passion Were Always In Fashion At The COPPA Cabana

There are primarily three types of information businesses can gather about online consumers:

  1. behavioral information – or page request and clickstream information passively recorded through the use of user logins, cookies, and/or server logs
  2. declared information – or information typically obtained through users filling out forms and providing identifiers such as name, email address, zip code, and interests
  3. inferred information – or information that can be indirectly associated with users, such as through software that identifies similar likes and group behavior (e.g., collaborative filtering) or overlays of external consumer databases that can be matched up with a user’s declared information

Many advocates of these profiling systems promote the fact that only the first type of information is used and not the other two. The privacy argument here is that users cannot be identified beyond their behaviors and that any ad targeting is only a passive response to a user’s initiated actions.

While we think the subliminal advertising alarmists are smoking the same stuff as the Linux stock traders on Wall Street, serious privacy risks loom beneath the surface. The gap between these information types can be as thin as finding willing participants to match up two cookies on a web page. With such a connection, suddenly behavioral information could be actively used by businesses to initiate contact with users through media such as snail mail, email, and the telephone.

Granted, this scenario is unlikely if advertisers and businesses know what’s good for them. But there’s a fine line between spammers and stalkers. Until consumers are granted restraining orders, there’s been enough high-profile, egregious privacy violations in the past to keep everyone suspicious.

Escaping the Past

Privacy issues aside, there’s an implicit assumption that profiling is both effective and – for the viability of online advertising – necessary. But is it?

Behavioral data is certainly less subject to falsification than declared data. Taking Elwood Blues’ lead from The Blues Brothers movie, Greg has admittedly entered an address of “1060 West Addison” (home of the Chicago Cubs’ Wrigley Field) on more than a few Web sites.

But much of the logic behind profiling’s value is rooted in the passive mindset of traditional media, such as television and radio. Passive TV viewers offer little information about their current interests. Thus a detailed profile about their past behavior greatly enhances any ad targeting potential.

Contrast this with an active media characterized by user choice and interaction, such as the Internet. Instead of trying to extrapolate a user’s immediate interests from data collected weeks (if not months) ago, web users provide information about their interests at the very moment they click on a link or type into a search box.

It turns out that profiling is an improvement over crude run-of-site targeting, but its effectiveness falls far short of keyword targeting. In other words, if someone with a profile of a huge baseball fan types in “trips to Paris” in the search box on a web site, an ad for Air France will be much more effective than one for SportsLine.

Infoseek pioneered large-scale behavioral targeting with the launch of their UltraMatch (yes, this is the WWW and not the WWF) technology as early as 1997. But if it was so successful, why isn’t everyone following Infoseek’s lead years later? And why hasn’t Infoseek added anything new about UltraMatch on their site since June 9, 1997?

And Now A Word From The LAPD About Profiling

Of course, precisely matching advertising to a user’s content interests isn’t the be-all, end-all of online marketing. If you’re trying to sell or brand yourself for auto insurance, search queries and web pages about auto insurance shouldn’t be the only place to promote your business online.

Yet with today’s conventional web advertising wisdom, Milton Berle would have had to create slapstick skits about gas stations, automotive repair, and the petroleum industry for the Texaco Star Theater to effectively brand its namesake. It would be ludicrous to presume that the only businesses that should advertise on freeway billboards are in the transportation industry.

The good news, however, is that we’re asking good questions and experimenting with models to better serve business and consumers – interests we don’t believe to be mutually exclusive. However, getting to the right mixture will require rational thinking and open-mindedness on both sides of the profiling vs. privacy argument.

Because for now, for every Chicken Little who cries, “cookies stole my credit card numbers!” there’s some online entity caught with their hands in the HTTP cookie jar.

Related reading

Overhead view of a row of four business people interviewing a young male applicant.