TOC Your data in jeopardy: what are the questions? Your Turn

Advertisers and governments should not be surprised to be lumped together as I am wont to do (05/30/06 fillip). After all political candidates have not waited for me to launch their election campaigns as companies their commercial products.

The tedious art of governing however must be distinguished from the bright science of capturing votes. Advertising itself is divided between the glamorous promotion of brands and the mundane act of selling products. It is with an eye towards governing and sales that I see a link between governments and advertisers. As they respond to the pernicious peril of terrorism, governments needs to identify dangerous fanatics among a mass of ordinary people. Looking among all consumers for the segment of one, marketing directors pursue as elusive a target. In both cases the issue is one of pattern recognition in the presence of overwhelming noise.

From this detached perspective, we believe we can find useful lessons for advertisers. Two weeks ago, we stressed that a key issue in pattern recognition is to draw the line between:

  • training a system on some sample of real life targets
  • using this system, as trained, to reach a decision on some random target, i.e. to perform a so-called classification
I have been accused of being hopelessly out of tune with the present times. While I do love history and epics, this should not be confused with disdain towards popular culture. The latter may be enjoyed in relief from boredom or despair, it can also be a source of inspiration.

In the instance a distinguishing feature of Jeopardy! is to turn the normal quizz show format on its head, as its contestants attempt to find not the answer, but the question. Like Jeopardy!, training searches for the good questions to ask. Classification looks for an answer, the same as an ordinary quizz.

Today's advertisers stand accused of the same ambiguity that taints the efforts of governments as they fight terrorism. They make frantic efforts to learn all they can about all of the people on the assumption that their target lies there, somewhere. As such they seem to forget what they are doing.

In pattern recognition, training a system on the whole set of potential targets is never done for the reason it is either impossible, too costly or at the minimum self-defeating. Training ought to take place on a small subset carefully selected to represent the whole. Pollsters do not query all electors for their opinions.

As far as training goes, there is no reason therefore not to pay for acquiring data for what is but a limited sample. Marketers compensate focus group participants for their time, marketers ought to compensate people for providing personal profile data. Give everyone clear title to one's personal data, organize markets on which participants will be encouraged to trade such personal data and advertisers will be able to perform their legitimate tasks without undue hardship neither to the general economy nor to individual data rights.

I am well aware of the potential danger of paying a consumer for personal data. No one wants to encourage people to make a living out of profile selling, not because it would be immoral but because it would terminally bias the training sample. One can check for such instances however and this is already done by focus group organizers. The real danger would be that besides undesirable ones too few other people would volunteer their data. Allow me to be sceptical and cynical both. Everyone has a price as they say. On a well organized market, with good liquidity in view of the potential number of participants, advertisers ready to pay market price will find their counterparts. But like the baseball players seen last week, more desirable people will fetch a much higher price.

Assume then that fair and balanced markets enable advertisers to learn what are the right questions to ask in order to recognize their targets. Classification can then be performed using the same architecture as described in my fillip about the airline passenger data. Since the stakes of selling commercial products are not as high, this architecture can be simplified by doing without tamper-resistant environments.
.
In the diagram above, let B be the advertiser. B puts all the good questions and desirable answers (aka selection criteria) in a wish list posted (p) in encrypted form on some transparent peer to peer network. Once it has been successfully matched (M) within a local confidential environment (1) to the profile of a consumer A , he or she may take the appropriate decision or, if the advertiser is so open, deepen the one to one interaction. To that purpose, the system mails (m) A's own wishes to be checked against the advertiser's profile, which in the instance contains data about the offer on hand. When verified, the resulting mutual match is a solid motive to try to close the underlying transaction.

In conclusion, whether they look for the right questions or for the desired answers, whether they train or classify, advertisers can fulfill their roles without any threat to privacy. They simply need to:
  • enforce the distinction between designing a marketing campaign and implementing a sales campaign
  • resort to appropriate technologies according to the constraints specific to each stage

Philippe Coueignoux

    • (1) for a description of a Confidential Environment, see US Patent 6,092,197 by Coueignoux
May 2006
Copyright © 2006 ePrio Inc. All rights reserved.