A global financial services and insurance company based in New York City minimizes risk, increases fraud detection and gains deeper insight into claims losses when it implements IBM Watson Explorer software to provide a faster, more efficient data capture and analysis solution
Client Background:The client is a multinational insurance and financial services company. Through numerous subsidiaries, the company provides commercial, institutional and individual customers with a broad portfolio of products and services, including property casualty insurance, life insurance, retirement services and asset management services.
The business of insurance is simply about assessing the risk of occurrence of some action - damage to a high-risk structure from a hurricane, for example - and putting a dollar figure on the actuality of the action occurring. It’s a high-stakes bet, and insurance companies need every bit of information at their disposal to understand the true risk and limit risk exposure.
As one of the largest corporate and industrial insurance underwriters in the United States, an international property and casualty underwriter, and domestic life insurance provider, this client must understand and minimize risk to be profitable. For decades, the means for understanding risks and assigning a financial value to those risks before issuing policies has been the use of tables, charts and graphs based on structured data. Before issuing a life insurance policy, in a basic example, underwriting systems consult a table that determines the likelihood of a person dying and the heirs collecting on the policy before the amount of monthly premiums are collected to cover the payout. The processes for determining risk have become increasingly sophisticated as more and more data is used, including information about a person’s occupation, lifestyle, race and other demographics. This client and other insurance companies today employ vast databases that contain this quantifiable data and use it to make underwriting decisions.
The insurer faced the challenge that although structured data is critical to understanding risk, much of the information that could improve accuracy and reduce risk was not typically stored in its structured databases. An existing business intelligence solution used by the insurer did not capture critical data from unstructured sources, such as claims adjustor or underwriter notes. An example of this data would be if a business wanted to change its property insurance policy on a building based on a new addition to the building. An underwriter might have been able to note that an advanced sprinkler system was included in the new structure, and that a fire suppression system was retrofitted to the existing building at the same time, but that information may not have been captured in a way that would change the cost of the policy or document and reduced the risk to the insurer.
Additionally, assessing the accuracy of exposure over the life of a policy posed a challenge. As a result, when a customer wanted to renew or extend coverage, the wealth of information about the policyholder and the existing policy was essentially locked in multiple applications and systems. Information about the customer and the business environment that could affect premiums, terms and reserves was not known.
To complement traditional structured data sources and use the rich information captured in unstructured documents to understand and reduce risk, the insurer sought a solution that would capture and analyze policy-related information from a wide variety of data sources to support timelier and better-informed decisions about issuing policy.
IBM Watson Explorer Solution:As much as 80 percent of enterprise data is unstructured, captured in notes fields and scattered among systems. This global insurance firm is using natural language processing to comb through 15 years of structured and unstructured data to glean entirely new insight into the true risk faced when issuing or renewing policies. The same analytical technologies that power the IBM Watson computing system enable a better understanding of the complex factors contributing to claims losses and support ongoing examination of existing policies as they mature. For example, the solution can discover specific words or phrases such as “fire-suppression system” or “not up to fire code” within a claims adjustment report that can be used to adjust policies or rates, and even discover potential fraud when conflicting statements are given.
To use the policy-related information captured in unstructured documents such as underwriting and claims notes gathered in the field, the insurer deployed IBM Watson Explorer software. The initial implementation of the solution is being used to search 15 data sources, both structured and unstructured, going back 15 years.
Key data elements have been defined for extraction, including policy number, effective date, limits of liability, payment history, claims, notes, reports, correspondence and related policies. Over time, the insurer plans to expand the number and types of sources to encompass databases, blogs and wikis that provide critical business, economic and weather-related information.
Once captured, the information will be used to create insights that validate actual policy information with the content of company databases and evaluate how this information, along with related, unstructured content about the customer or policy, could affect risk exposure. For example, does the company also insure any of the policyholder’s suppliers or subcontractors? If so, is the company exposed in multiple ways that are not reflected in existing premiums or reserves?
To better detect fraud and adequately fund reserves in the future, the insurer plans to implement additional IBM Watson Explorer software to build predictive risk assessment models. Trend data from the business analytics system will automatically populate these models, continually improving what-if analyses.
Benefits of the Solution:• Anticipated lower risk to the business by uncovering unexpected patterns and associations among existing data sources
• Expected reduced claims losses as a result of accurate assessments and setting of adequate reserves
• Saved millions of dollars in staff time by automating the risk assessment process
InstrumentedThe solution captures data from structured and unstructured sources such as text fields in online reports, web surveys, and handwritten reports scanned into document management systems. The databases that contain the structured and unstructured data can reside in systems throughout the organization, yet remain accessible to the content analysis solution.
InterconnectedThe content analysis solution reaches across organizational lines within the organization and presents a more unified view of policyholders and the risk exposure of the insurer. Data captured about property policies, for example, is presented in context with other policies, such as life and auto, giving a fuller understanding of the relationships between the insurer and the policyholder as well as relationships between policyholders.
The solution uses natural language processing to analyze millions of pieces of structured and unstructured data to find patterns and unexpected associations that create a more accurate risk profile. Without human intervention, the software combs through the data to discover entirely new insights into whether to issue policies and how much a policy will cost the policyholder.
Insurance companies face the ongoing challenge of remaining profitable while providing dependable coverage of losses for people, businesses and institutions. This balance is achieved through accurate analysis of the risks of an event occurring, which can be represented as a statistical probability based on structured data. Using IBM Watson Explorer software, the insurer is transforming its business by incorporating unstructured data into the decision-making process. The result is more accurate risk assessment that reduces the financial exposure for the insurer.
The solution delivers improved decision making for the client based on the ability to analyze data from a variety of sources and systems to support accurate assessments. It also results in faster and more productive assessments because underwriters do not have to hunt for data; they are presented information based on predefined queries and analysis.
For shareholders and investors in the insurer, the improved accuracy of risk assessment helps reduce exposure and ensures that accurate reserves are maintained to pay out in the event of major events, such as hurricanes. This accuracy preserves investments and supports ongoing viability of the business. And the solution also benefits policyholders covered by life, property and business insurance by providing accurate coverage based on their true risk and accurate pricing of policies that ultimately serve their needs better.