Last month, news articles reported that Jon Venables, previously convicted of the barbaric murder of toddler James Bulger and released with a new identity 8 years into his sentence, had been returned to prison after infringing the strict conditions of his release. The second article, published soon after this one, revealed that a blonde, green-eyed, American woman, calling herself Jihad Jane, had agreed to carry out a murder overseas on behalf of a terrorist group. But what do these two cases have in common and how does this relate to marketing?
Both Jon and Jane failed to meet expectations that others had about their behaviour. Indeed, according to reports, social workers had considered that Jon Venables did not pose a significant risk to himself or others, while Michael Levy, a US Justice Department attorney, said that the Jihad Jane case “shatters any lingering thought that we can spot a terrorist based on appearance”. Jon and Jane are outliers who did not fit the relevant profiles.
Unfortunately, before Jon other former prisoners re-offended, and before Jane other westerners joined the jihad. That is, these occurrences were not unique. So, why did the officials in question fail to predict either case? Bringing the discussion to the business context, how could the 2008 sub-prime crisis occur, or why is it difficult to spot a money launderer? And, more broadly, why is it so difficult to prevent ‘undesirable’ behaviour?
Organisations routinely capture records of almost every aspect of our daily lives (see EPIC), incorporate them in a database and develop models of who we are and what we do. These models, the profiles, enable private and public organisations to assess risks and opportunities, subsequently informing decision making in a variety of areas, including non-commercial applications such as crime prevention and detection. In order to be effective, profiles need to be based on well-defined models of the behaviour in question, which means that there needs to be a substantial number of past events to build the models on, and/or there needs to a clear cause-effect chain. But profiling the type of activity described in the examples above creates significant challenges to organisations because of the secretive, ambiguous and dynamic nature of behaviour in question (e.g., ). Moreover, with such small samples, erroneous inferences may occur (e.g., ).
The lack of empirical support for the profile means that the resulting decisions are made under conditions of ambiguity. The problem is further composed by the possible consequences of false positives. For instance, data subjects may be wrongly allocated to categories that they do not belong to and/or suffer discrimination as a result of profiling – e.g., be prevented from boarding a plane or have one’s financial assets frozen. Furthermore, as the negative effects – or costs – of false positive and of true positives are, traditionally, felt by different agents, there is likely to be very low tolerance for profiling mistakes and poor support for intrusive profiling practices.
Profiling works very well in credit card fraud because, over time, the credit card provider develops a very good model of what is normal behaviour for the owner of that specific card. Moreover, as plastic money is increasingly popular, the data pool is ever increasing and more and more encompassing. Finally, the costs of false positives are, traditionally, negligible and far outweighed by the benefits, and benefit the same individual.
For situations where there is not a direct relationship between the model of the profile and the object being profiled, there is limited empirical evidence, and there are high economic and/or social costs of false positive errors, profiling remains a challenge. This is the situation increasingly faced by marketing and risk managers, as they are called upon to assist with initiatives such as e-borders or the fight against money laundering. This is not to say that profiling is not useful – on the contrary, it has a crucial role to play. But it needs to be based on a highly iterative approach to modelling, as indicators that were established at the outset of the profiling exercise may become obsolete very quickly.
 Canhoto, A. I. (2008). Barriers to segmentation implementation in money laundering detection Marketing Review, 8 (2), pp.163-181
 Risen, J. L., & Gilovich, T. (2007). Informal logical fallacies. In R. J. Sternberg, H. Roediger III, & D. Halpern (Eds.), Critical Thinking in Psychology (pp. 110-130). Cambridge: Cambridge University Press