How data consortiums impact the accuracy of predictive analytics
Unseating an incumbent insurer while lacking the data required to make the most accurate decisions regarding risk selection and pricing is a challenge.
Competition across the insurance industry is growing, as companies chase profitability by focusing on narrow, segmented portions of the broader market.
This approach is sound but oftentimes creates obstacles for insurers looking to grow into new classes or geographies, since established carriers have a knowledge and expertise advantage.
Unseating an incumbent insurer while lacking the data required to make the most accurate decisions regarding risk selection and pricing is a challenge for both insurers that haven’t invested in predictive analytics, and oftentimes, even for those who have.
Extrapolating the best data
Despite the abundance of data being created in nearly every market, concern over a lack of usable information is becoming more prevalent as insurers come up against substantial challenges of legacy systems and a lack of IT resources to standardize and govern data in order to glean actionable insights.
An added strain applies for those seeking to modernize the customer experience and reduce the number of questions they ask potential policyholders on forms.
While in-house data may start to diminish over time, insurers will adjust to ensure they have the information necessary to make informed strategic decisions about how to grow.
The need for third-party data
An insurer has tremendous knowledge about the risks it writes, less information about the risks it quotes — and generally knows nothing about the policies it didn’t write or quote. As such, third-party data is becoming critical to insurers. Whether it comes from telematics, Census, Bureau of Labor Statistics (BLS), SAFER or any number of other sources, third-party data is proving invaluable to the development of predictive models.
One of the most valuable sources of third-party data comes from consortiums, particularly those that contain insurance transaction data such as recent premium, policy, billing, claims, and audit & inspections information.
ISO and NCCI are among organizations in the industry that pool data to set rates, and while information from those organizations is helpful in observing general trends and creating aggregates, it often lacks the transactional-level of detail that boosts the predictive capabilities of a model.
A consortium that captures transactional data can expand an insurer’s view by providing anonymized data with valuable detail relevant to assessing risk while removing all information pertaining to the insurer who wrote those policies, the PII of the policyholder, or previous information on rates and pricing.
There are many ways in which consortiums not only improve a model’s effectiveness in familiar lines of business, but also open the insurer up to more expansive, and tailored strategy around their analytics initiatives.
Expertise is power
One downside to building a model that’s based on a single view of the market is most apparent if an insurer’s growth business strategy involves branching outside of the geographies or classes with which it’s already familiar.
For example, an insurer with a particular expertise in long-haul trucking companies in Oklahoma, has become very good at assessing those businesses. So good, in fact, that their data may likely show that similar risks in new geographies are all good risks. That’s because, through their expertise in evaluating that group of businesses, they’ve actually created a selection bias in their data.
If they expand to long-haul trucking companies in Tennessee, a model built solely on their data would tell them that all long-haul trucking companies are good risks, which is obviously not true. What is true in this example, is that this insurer doesn’t have enough bad risks in their portfolio to create an accurate predictive model that applies beyond their current market share. Augmenting proprietary in-house data with consortium data protects an insurer from itself, providing the data necessary to fill in the blind spots when expanding into new geographies, selling up or down market, covering new exposures, or altering risk appetite. It bridges a significant part of the knowledge gap that is present when insurers look to expand.
Consortium data supplies a model with more variables to consider; that is, more predictive horsepower than a siloed data set. An insurer may have plenty of data, but only a small amount of that data might be predictive or actionable, which is where consortium data can be particularly important. Before a model goes into production, extensive testing should occur to find which correlations have the most predictive value. Consortium data ensures data scientists have access to a more diverse set of data and a wider array of predictive variables to align with an insurer’s unique business goals.
In fact, models that leverage consortium data have proven to add immediate value to an insurer’s profitability and loss ratio.
Valen Analytics recently released its third annual ROI study, measuring the evolution of data-driven insurers and their overall market performance in areas such as loss-ratio and premium growth.
Kristin Marr is president of Valen Analytics. To reach this contributor, send email to kirstin.marr@valen.com.
The opinions expressed here are the author’s own.
See also: