The Next Revolution in Territorial Definition Is Here
By Marty Ellingsworth
In the past, analyzing historical claims within large geographic boundaries defined expected loss costs. States, counties, urban centers, and even map features were used to divide the country into approximately 1,000 different areas. Then, companies discovered a data-mining bonanza by including ZIP codes, thereby increasing the number of regions in the United States to more than 30,000. That analytic innovation resulted in decades of
benefits for early adopters, who remain competitive by frequently updating their territorial definitions as books of business and risks change. Today, a more refined territory definition process is at hand with the potential to create more than 200,000 segments.
Figure 1:
Territorial Definition Revolution
Click to enlarge
In this example, three maps of Milwaukee, Wisconsin, show increasing granularity of geospatial boundaries
as territorial definitions are refined.
Changing with the Times
Figure 2:
ZIP Code vs. Block Group
Click to enlarge
Within the green ZIP code boundaries, there is clearly more loss cost differential at the block group level (gray lines) that segments risks better. |
ZIP codes provide segmentation within larger territories, and
census block groups provide further segmentation within ZIP codes, with the added benefit that changes are better aligned with insured loss. The same technical approach is being applied to small-area geography in ISO Risk Analyzer® Homeowners perils-based pricing, with dramatic improvements in prediction of loss costs by peril.
The new process dramatically segments risks within and across larger geographies. ZIP codes provide mail delivery efficiency, but they are not aligned with insurance risk and are not designed to be homogenous insurance risk statistical units. The smaller physical size of block groups permits a finer resolution of the environmental factors impacting risk. The inherent mean-variance structure of large pools of claims data creates an environment
in which large swings in actual riskiness are bundled into an overly large geography. The old method compounds this by
using more geography to make up for sparse claims. The latest innovation leverages immense amounts of historical data and a statistically accurate geospatial modeling process powered by census small-area geography definitions. Filed and approved in many states already, it is the next revolution for competing on analytics in insurance.
Figure 3:
Powerful New Risk Segmentation
Click to enlarge
ISO Risk Analyzer Homeowners perils-based pricing adds significant differentiation of loss cost estimates within larger geographies.
Figure 4:
ISO Risk Analyzer Personal Auto – Indiana
Click to enlarge
Single-state writers and regional carriers may already be able to move their entire books of business into the new paradigm. |
While geospatial data powers this innovation, companies in the "have not" category for GIS can still implement and execute on this analytic using an address-parsed lookup table precalculated to map every known address. The polygons can be any predefined shape, so customization can be made for any requirement. Often a cascade of options is available using several mapping layers, such as U.S. Postal Service addresses, Tele Atlas, NAVTEQ, and others, to fully capture mapped points and their inherent and
surrounding data attributes. This agile process has been implemented successfully for almost a decade, and while not as
flexible as a full GIS architecture, it has a field-leveling effect
for competitors with bottlenecks in their IT departments.
The Genie Is Out of the Bottle
Territorial definition is now the minimum new investment to
be made in lines of business that historically paid little attention to the geographic and community effects of where risks exist. The environment is critical in understanding risk, and predictive analytic models that leverage geospatially derived data are ready now for widespread adoption. 
Marty Ellingsworth is president of ISO Innovative Analytics, a business unit of ISO.
|
