Finding Your Stability Threshold With AI

Many FWA detection systems use volume to overcome a lack of ability.  But throwing more and more medical claims data at a system to find fraud patterns is not a silver bullet. Adding data might improve a model’s performance in the short run, but quickly hit diminishing marginal returns.

On the other side of the coin, there are many simple questions that can be answered and modeled with a very small training set.  These questions have very few variations in the outcome and are based on simple yes and no answers with no shades of grey.


Traditional rules-based systems involved with medical claims processing offer a good example of an application with a low-performance threshold. Most medical claims share key fields: the parties involved, the number of charges, standard procedure codes, time frame, etc.  Most medical claims share standard forms to comply with standard regulations.  Algorithms that automatically process standard documents usually only need a few hundred examples to train to an acceptable degree of accuracy.  These systems are stable but extremely limited in scope.


True AI models in medical claims FWA detection systems should train on real-world examples. If medical claims conditions change, gradually or suddenly, and the model cannot change with it, the model will fail, and the system’s predictions will no longer be reliable.


Keeping the model stable requires ingesting fresh training data at the same rate that the environment changes.  This is called the stability threshold.  True AI systems like the Centaur continually train and adapt on new data at real-time rates, providing the stability levels required for dependable FWA detection.

Request Your Claims Denial Impact Calculator