stolen, you might have been amazed at how fast the credit
card company identifies the different buying pattern and
alerts you. That is the beauty of modern predictive analytics. Learning to correlate not only your buying patterns,
but also those that occur after someone has stolen a card.
Marrying the two has allowed the credit card industry to
manage fraud differently.
Applications in Workers’ Compensation
When it comes to workers’ compensation program management, it is well known that the Pareto principle, or the 80-20
rule, consistently appears in workers’ compensation loss portfolios. This is the concept that a vital few claims, the 20 percent, will drive the majority, the 80 percent, of the program
cost. Most of the claims in that 20 percent category will not be
catastrophic. More likely they will be mixed within the large
inventory and will creep in value. Their durations will slowly
become longer while their costs grow higher. Being able to
predict the high value claims as close to the onset as possible
and manage these claims differently out of the gate is the key.
With today’s modern predictive analytics, multivariate mod-
els can consider simultaneously how various types of risk
characteristics of a claim, for example, age, gender, employ-
ment type, work hours, injury type, injury time, etc., affect the
outcome of an injury claim severity by assigning appropriate
weights to each of the risk characteristics. Well and carefully
designed claim models with 20 to 50 risk characteristics have
been successfully developed and deployed over the last sev-
eral years in the industry by several insurance companies and
self-insured organizations. The models have proven to suc-
cessfully predict more complex and costly claims.
Models can be built by using data acquired at first notice
of loss along with external data (such as unemployment
rates, worksite information for the policyholder, crime rates,
etc.) to effectively forecast outcomes for individual claimants. Such a model can transform claims management by
improving the allocation of claim’s resources, increasing
efficiency, improving the estimates of claims severity and
reducing claim’s cycles. Companies that begin to successfully integrate predictive modeling today may find that in
12, 24 and 36 months they will have learned things from
their individualistic experience and will likely be ahead of
those companies that wait. As in the fraud models, behavior
is always changing. Building on the knowledge in the model
everyday significantly can increases the accuracy.
However, the true savings doesn’t come from the model itself, but in the proper use of the information that the
model produces. In modeling workers’ compensation, early
identification of the vital few allows the opportunity for better