Screening for bad customers: the difficult balance between proof and intuition

I’m back from the Academy of Marketing conference, in Coventry, where I presented on the challenges of modelling undesirable customer behaviour. The topic of profiling dysfunctional or undesirable behaviour is a key interest of mine, that I investigated in my PhD, and addressed in a previous posting here and a range of publications and presentations. But, today, I would like to share a very brief overview of the presentation at the Academy of Marketing conference. As usual, your thoughts and feedback are mostly appreciated, so do contact me.

Undesirable customer behaviour is very costly to organizations, be it directly by hurting the bottom line through foregone revenues (e.g., when customers do not repay their loans) and/or increased costs (e.g., in the case of fraudulent insurance claims), or indirectly by affecting staff’s morale and productivity (e.g., as a result of abusive or threatening behaviour from customers) or the organisation’s reputation (e.g., when the organization’s name is dragged into the front pages because it was used to transfer the proceeds of crime or finance terrorist activity).

When the organisation spots undesirable behaviour among its customer portfolio, it may try to demote said customer (e.g., decline further loans or offer them on less favourable terms) or even terminate the relationship altogether. However, there are numerous technical and legal barriers to overcome. Moreover, such initiatives also have a negative impact on customers’ spending and loyalty, and often generate negative word of mouth. So, this really is a case where prevention is better than cure. In other words, it is critical for firms to correctly identify and avoid customers likely to engage in undesirable behaviour.

Profiling undesirable behaviour is very, very challenging. Experts deal with speculative, temporary modelling scenarios as a result of the factors depicted in figure 1. Consequently, the resulting decisions are made under conditions of heightened ambiguity.

Figure 1. The Challenges of profiling undesirable behaviour
(c) Ana Isabel Canhoto, 2010

In my research, I address how organisations cope with these difficulties by looking at both the signs of (un)desirable behaviour and the process of capturing and making sense of these signs.

Regarding the signs, I distinguish between those with low levels of abstractedness (e.g., income) and those loaded with assumptions about the drivers of the profiled behaviour (e.g., intentions). I also distinguish between static and dynamic behaviour drivers, with the latter being more difficult to model than the former.

Regarding the process, I consider both automated screening tools and manual screening, noting the relative advantages and disadvantages of each one. For instance, while automated tools are superior in terms of collecting, processing and disseminating data, they usually need a large database in order to outperform manual screeners which, by default, is not available in a novel situation as the one encountered by banks in the 2007-08 financial crisis. Also, as demonstrated by colleagues analysing the 2007 sub-prime crisis (e.g., here), soft data often improves prediction of loan default.

In summary, profiling undesirable behaviour requires a sophisticated approach that balances automated and manual screening, and hard as well as soft data. My presentation at the Academy of Marketing conference explored these issues using the specific case of screening for credit worthiness during the credit crunch.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s