Credit Management meets computer automation

A necessary balancing act. A risky business

The date beneath

For businesses looking for the optimum approach, Equifax is one of the two main UK organisations (the other is Experian) who offer a range of support tools. First and most important is the underlying data. Since legal changes in 2002, organisations may no longer use the full electoral roll and must restrict marketing activity to individuals who have not opted out (currently around 60% of the roll). However, the entire roll is available for credit checking purposes - and it is used, in conjunction with data pooled from a large number of independent organisations in respect of individual payment history, to provide credit screening services against potential new customers.

It is important to distinguish screening services from predictive modelling. The former focus on how well individuals have dealt with credit in the past and takes the (reasonably) rational view that those who have incurred County Court Judgments or failed to pay bills on time might tend to do the same again in future.

Predictive services combine public data with other data held in-house by businesses to come up with a view as to how likely an individual is to be a future credit risk. This difference is at the root of many mass credit card mailings - and the cause of a certain amount of consumer anger. Mailshots go out explaining that an individual has been pre-screened for a particular card: All they have to do is fill out a form to accept. And when they do, their application is then rejected.

What is happening here is that the individual is a good risk on the basis of past data: But once more detailed information is provided, they do not fit the profile that the credit card operator is looking to recruit. This is also the root of a great deal of angst in face-to-face dealings: an individual turned down for credit may have little or no adverse history in their past at all. Rather, a statistical model may simply have said that whilst the overall likelihood of anyone defaulting is 0.1%, the risk relative to that individual is 0.2%. In relative terms, double the risk: In absolute terms, almost certainly a good risk.

Companies such as Equifax offer generic segmentation models, which sub-divide the UK population into groups constructed using a range of variables from the census and other sources: They also offer statistical modelling, to create the predictive scoring outlined above. A range of techniques may be used - although logistic regression and CHAID analysis are two of the most popular.

Sponsored: 10 ways wire data helps conquer IT complexity

Next page: Recession shakeup