A Model Success Story
Af TER RoLLINg oUT PREDICTIVE MoDELINg for medical
claims management this year, Ahold uSA Inc.—a supermarket
chain with approximately 800 stores and 110,000 employees—
expects to see annual savings in the low seven digits from a
workers’ compensation medical costs.
the expected savings is based on results from a pilot
program in which the model was applied to 1,800 open
claims last year. “What the model does so beautifully for us
is that it identifies our problem claims in advance,” observed
James Snell, the head of predictive analytics at Mac risk
Management Inc., the division of Ahold uSA that manages
workers’ compensation and general liability claims.
by addressing cost predictors on a per-claim basis, the
model was able to project unpaid medical losses in 30-day
increments, which resulted in immediate interventions to
reduce claims severity. “the return on investment of a well-run
predictive modeling program can be exponential,” he said.
the effectiveness of predictive modeling depends on
vast amounts of appropriate data. Ahold uSA is in the
enviable position of having a “treasure-trove of data” in its
claims administration systems, said Snell, based on data
collected since the company started self-insuring and self-
administering in the 1980s.
the Ahold uSA model uses claim characteristics, medical
transaction details, and other data sources to reveal factors
predictive of potential claims severity, said rong yi, a consultant
in risk management and predictive modeling from Milliman Inc.
Such indicators include multiple visits to doctors and the use of
certain prescription drugs.
the model then prioritizes claims that need special
handling and medical case management, she noted. this helps
injured employees receive appropriate medical care to reach
maximum medical improvement and return to work sooner.
And frequency/severity-based predictive
models are now a best practice in personal
lines, necessitating new variables and interactions for competitive pricing, Stoll said.
According to the towers Watson
third annual predictive modeling survey,
released in February 2012, almost all—
97 percent—of u.S. personal lines executives view sophisticated underwriting and
risk selection as essential or very important. eighty-five percent of those said they
are using predictive modeling or plan to,
according to the online survey of 60 u.S.
carriers and nine Canadian insurers.
For pricing and underwriting, predictive
modeling is also picking up in commercial
lines. Approximately 70 percent of
respondents indicated they are either
currently using it or planning to do so
within the next two years, according to
the towers Watson survey.
the greatest commercial lines
momentum is in primary commercial lines
such as business owners, commercial
auto, business package, and workers’
compensation, for which more data
enable stronger models, said Wu.
Predictive modeling, however, is
taking longer to pick up in workers’
compensation because it is such a
unique line, Wu said, and has unique
challenges and opportunities. unlike
other commercial insurance, workers’
compensation has a uniform product
in coverage and pricing. “there are less
risk characteristics available for workers’
compensation for risk segmentation than
for other commercial business,” he added.
Workers’ comp also is lagging the most
because it requires multiple data sets for
which data availability for variables can
be shallow. Such is the case for applying
predictive modeling to the claims process,
which began about five years ago, said Wu.
the health care industry, meanwhile,
has been using predictive metrics for
medical case management that are also
applicable to workers’ comp, said rong
yi, director of the risk adjustment and
predictive modeling practice at Milliman
Inc. “there is a familiarity in the medical
data and health care system of high cost
claimant behavior that also applies in
workers’ compensation,” she said.
In health care, the term “predictive
modeling” was coined in the early 2000s to
describe models containing co-morbidities, yi
said. these include diabetes, high blood pressure, depression, smoking, and other factors
to detect potentially expensive claims.
Predictive modeling will grow only
more attractive for insurers and other
businesses. At the same time, actuaries
still need to secure their role in the
marketplace, experts say.
For starters, data and benchmarking
organizations could be viewed as either
collaborators or competitors. WorkersComp
Analytics, for example, is building a claims
database called the national WorkersComp
Clearinghouse, which also offers metrics
with results similar to predictive analytics.
other professions, which could be
collaborators or competitors, are pursuing
predictive modeling as well. to stay
competitive, therefore, actuaries will need
to show how their expertise is necessary
and more desirable. “If actuaries do not
embrace data-driven analytics and provide
the service, companies will secure it
from other professions,” said Stoll.