Due to the rapid changes in both the health care market and more specifically the amount of fraud being committed in it, it is even more important than ever to maintain some degree of life cycle management to update the analyses used to detect and identify aberrant activity.
However, many organizations do not have the resources or the expertise to maintain the performance of their analytical methods. Quite a few organizations build transactional rules or models and then continue to use them for years (in some cases four or five years). And even though the analysis has degraded (sometimes by as much as 60 percent), they continue to use the same analytical methods year after year.
Without proper tools and methodologies in place, it is very difficult to manually maintain fraud analytical methods.
One of the few statistical truisms is that all models and analytical methods degrade over time, and this is even more true for fraudulent models. This is largely due to fact that the entities committing fraud or abuse adjust their behavior to avoid detection by said models.
When the models or analytical methods become less effective many more false positives are created and more money is lost.
The whole scenario very mush becomes an issue of the cart before the horse: because organizations are so busy chasing the leads they have, they don’t have the time to improve the quality of the leads they detect and identify.
As mentioned previously, many organizations cannot address this issue with their current resources, hence organizations should seriously consider acquiring either more resources or technology that can evaluate the performance of their analytical methods on an on-going basis.
None of this is easy, and there will be pain in introducing new processes to make sure that detection methods are more accurate, but the alternative is far worse and ultimately, it defeats the purpose of the having the analytical methods in the first place (if the methods become wildly inaccurate).
Be honest: when was the last time you updated your fraud detection models? And if you have updated your models, what strategies did you use to help carve out the time?