DATA MINING AS A PERSUASIVE TECHNOLOGY
by Ed Colet, Virtual Gold, Inc.

Computing technologies and products that are intentionally designed to influence a person's attitudes and behaviors in a predetermined way are collectively known as persuasive technologies. A strict definition of a persuasive technology is something designed to change and influence the behavior of the user of the technology itself (e.g. to eat more vegetables, to stop smoking, etc). Unlike the strict definition, the influence of data mining technology often extends beyond the user of the technology itself (i.e. the end-user analyst). Unless one is mining one's own individual data, the intent of data mining tools is to influence the behaviors of others - typically consumers. Two questions arise: How might analyst-users be persuaded to act upon such patterns that they discover? And once actions are taken, how might others (i.e., consumers) be persuaded or influenced to act in "desired" ways?

Persuading the end-user analyst:

In order for an analyst that has discovered a data mining pattern to be persuaded by the value of a data mining result, three points must be evident: credibility, actionability, and measurable results.

The analyst must believe that the information is credible, yet sufficiently interesting and surprising enough to pique his/her interest. The credibility of information from a data mining analysis is critical, and is heavily dependent on the quality of the data being analyzed. This is why clean data is critical for analysis. For the most part, a well executed and implemented system (i.e. extensive data scrubbing processes are in place prior to mining), with powerful and valid algorithms contribute to the credibility of discovered patterns.

Actionability refers to the notion that the interpretation of the pattern suggests clear and unambiguous actions to take. If patterns are well articulated, it translates into clearer actions to be taken. For example a pattern that reports "Yogurt sells well on Thursdays and Fridays when priced below $1.45" is better articulated than "Yogurt sells well on specific days of the week below a certain price point". The action suggested by the earlier pattern is clearer than that suggested by the latter.

The actions that are taken should also be tied to clearly measurable results. In this case, the increase in sales revenue is an obvious measure. Depending on circumstances, other measures such as ROI may be suitable.

Persuading the consumer(s):

The other dimension to consider is how might consumers be persuaded into acting in ways that are consistent with the patterns discovered by analysts; and expected in terms of the actions taken by decision makers?

If patterns are discovered as a result of mining data that reflect stable behaviors of consumers, then one would think that you really don't need to do anything to persuade consumers to follow one's marketing plans. Ideally, things should just happen naturally. But in actuality, things aren't so simple. The best laid marketing plans can have surprising and undesired effects. For example, perhaps a seasonal effect was overlooked and therefore consumers don't follow expected behaviors. As a result, strategic ways that can influence consumer behavior need to be considered.

A clearly useful strategy that's effective for consumers is to utilize personalization. For consumers that are loyal to retailers, explicitly gearing a promotion or marketing campaign to the individual shopper can be particularly effective. Personalization can also be applied to an appropriately defined peer group. This has the effect of being less personally intrusive to the individual. Techniques such as collaborative filtering, in which promotions are presented to an individual based on a match of their interests with their peers have proven to be effective. Personalization must be balanced with issues of privacy (for the sake of brevity not addressed here). Just as the credibility (and validity) of patterns are important for persuading analysts, the validity of discovered patterns must be equally applicable in order to influence consumers.

In conclusion, data mining can be a persuasive technology with respect to two aspects - persuading the end-user analyst, and then eventually persuading or influencing the behavior of consumers.

---

Ed Colet is the Acting Director of Research at Virtual Gold Inc., responsible for developing analytical methods for data mining and for investigating human factors and usability issues of business intelligence systems. At present, he is in the final stage of completing a doctoral dissertation in the Cognition and Perception program at New York University's Department of Psychology. Ed has also worked for IBM Research at the T.J. Watson Research Center. At IBM, Ed was a member of the group that developed Advanced Scout, the data mining application for NBA teams. His research interests focus on statistical methods and human factors.

For more information, see http://www.virtualgold.com.