From predictions of business bankruptcy to potential property damage from hurricanes, data and analytics are providing new insights into everyday risks. There’s no question that insurers are increasingly using predictive models in their operations and relying on models to help plan for catastrophes. Geographical information about the properties they insure combined with catastrophe models can be used to produce an estimate of the number of claims they can expect. This information can help insurers plan to deploy claims professionals where they will be needed most. As new models and sources of data are brought into insurance processes, it’s becoming increasingly important to keep in mind applicable laws and regulations—as well as ethical principles and professional codes of conduct.
Members of professional organizations such as the Chartered Property Casualty Underwriters (CPCU) Society and the Casualty Actuarial Society (CAS) can provide guidance on an organization’s code of conduct to help understand ethical issues. For example, Canon 4 of the CPCU Code of Professional Conduct states, “Insurance professionals should be diligent in the performance of their occupational duties and should continually strive to improve the functioning of the insurance mechanism.” Thus, to the extent new data sources or predictive models can help improve the claims management function, it can be viewed as part of a CPCU’s professional standard to seek to use those data sources or implement such predictive models.
On the other hand, Canon 3 states, “Insurance professionals should obey all laws and regulations, and should avoid any conduct or activity that would cause unjust harm to others.” Thus, it’s a CPCU’s professional responsibility, under the referenced canon, to follow applicable laws and regulations with respect to new data sources or predictive models, and to not “cause unjust harm to others.”
Let us examine some hypothetical examples of how these principles may be applied in the field of claims management. As mentioned above, information from catastrophe models can be used to mobilize personnel in order to have claims professionals in place for increased claims volumes. It’s recognized that prompt handling of claims tends to increase customer satisfaction.
Some insurers are using data and analytics for bodily injury claims triage. For example, insurers might use predictive models to identify claimants whose bodily injuries have a potential to become severe. Helping those claimants get timely, appropriate medical treatment may speed their recoveries, which may allow those claimants to resume their normal activities more quickly and reduce the cost of the claims for their insurers.
Claims triage might also appear with respect to damage of insured property. For instance, provisions in some homeowners’ policies advise homeowners to take measures to protect property from further damage and to keep receipts for expenses related to such actions. Imagine the following hypothetical scenario: A storm hits an area, causing damage to many homes. With damaged roofs or broken windows, continuing rain may cause further damage to homes in the area, and it may take some time for homeowners to find contractors to make the necessary permanent repairs. Insurer A has a claims triage system, which flags claims that might lead to additional damage unless preventive measures are taken. This insurer may want to have its claims professionals remind the homeowners that their policy provides coverage for measures taken to protect the property from further damage until permanent repairs can be made, and that they should keep track of expenses and save receipts.
Insurer B does not have a claims triage system, and no one contacts their policyholders to remind them that their policies cover expenses necessary to protect their property from further damage. The policyholders of Insurer B who do not make temporary repairs and fail to take measures to protect the property from further damage may be disappointed in the outcome of the claims settlement process. While there is no obligation for either insurer to notify their customers of the policy’s coverage for provisional repairs, it is likely that Insurer A will be perceived as having provided a better customer experience and would rank higher in customer satisfaction.
Fraud detection is another application of data and analytics. Fraudulent claims increase costs for insurers and policyholders. Claims fraud prediction models can help by flagging claims that have the potential to be fraudulent so they can be properly investigated. To the extent the model succeeds in identifying fraudulent claims (true positives), this would be in line with the exhortation to “improve the function of the insurance mechanism,” as found in Canon 4 of the CPCU’s Code of Professional Conduct. On the other hand, if the model has a high proportion of false positives—honest claims that are flagged as potentially fraudulent, causing a delay or improper denial—then it could be contrary to the guideline in Canon 3 to “avoid any conduct or activity that would cause unjust harm to others.”
In some cases, insurers may have regulatory requirements to consider with respect to efforts to implement a fraud prevention and detection plan. For example, the New Jersey Department of Banking and Insurance has a section on anti-fraud compliance on its website, which reads, in part:
Insurers that transact private passenger automobile insurance business in this State on either a personal lines or commercial lines basis, or transact health insurance business in this State are required to submit to the Department for approval a Fraud Prevention and Detection Plan, as required by New Jersey Administrative Code (N.J.A.C. 11:16-6.
The N.J.A.C. 11:16-6 stipulates that an insurer’s fraud detection plan must generally include several features related to training programs, procedure manuals, investigation units, and referral of suspected fraud cases. Furthermore, these plans are subject to review by the department’s anti-fraud compliance unit.
Managing the Data
The preceding examples illustrate how predictive models can be very useful in predicting potential damages from catastrophes, identifying claims where bodily injury or property damage can escalate, and identifying potential fraudulent claims. The examples also indicate that there may be ethical or legal issues one must consider when using new sources of data or models. When new data sources or predictive models are being considered, it may be wise to have a framework for systematic consideration of issues related to law, regulation, ethics, and professional codes of conduct. The components of such a framework can include:
• Examining characteristics of the new source of data or model.
• Reviewing the applicable code of professional conduct.
• Consulting with a company’s compliance and legal departments about applicable laws and regulations.
• Consideration of ethical and legal implications from using a new source of data or predictive model.
• Deciding what data elements or model features can be used as-is, those that could be used with some modification, and those that it would be preferable not to use.
• Documenting, implementing, and monitoring.
Although the items above are presented as a list, they aren’t necessarily meant to be taken sequentially. Rather, they can provide a framework of related considerations. In reviewing one item, a company also would likely need to consider information from other items. Lists of questions to explore for each of these items should follow. A first step in the process could be to examine the new source of data or model. This can consist of looking for the types of things that might cause concern for stakeholders—those who will use or be affected by the new data source or model. For example, regulators may be concerned about the impact of a new anti-fraud plan based on a predictive model, such as seen in the previously cited N.J.A.C. 11:16-6.
Other stakeholders may include company employees, agents, brokers, customers, regulators, and consumer groups.
Letter of the Law
This leads to another key item in the framework: a company’s compliance and law experts. Many companies have compliance, filings, government relations, or law departments. Employees and attorneys in these departments continually monitor bills and regulations introduced in state legislatures, state regulatory agencies, and the federal government for implications related to property and casualty insurance. Those working on new data sources or predictive models would be well served to find out who the experts are on insurance law, regulation, and compliance within a company. Discuss with them the new data source or predictive model at a project’s start, and keep in touch with them throughout the project. They can advise about applicable laws and regulations. They may also have regular contact with regulatory agencies, and should have an understanding of potential issues of concern.
Once a team has been advised of applicable laws and regulations, confirm that everyone understands the implications. Are there any features of the new data source or predictive model that cannot be used? For example, if the model generates a high rate of false positives, then this could impact the handling of honest claims, which could lead to an increase in consumer complaints, which could increase the likelihood of a market conduct examination. Regarding market conduct examinations, the New Jersey Department of Banking and Insurance states on its website, “[C]onsumer complaints, problems in the marketplace, and other areas of concern identified by the commissioner and department staff are significant factors in determining which companies and lines of insurance will be examined.”
After considering the issues, plans for using the new data source or predictive model may change. A good next step is to once again check with the experts in a company’s compliance or legal department to explain any changes. Even if changes are few, it can be prudent to meet with them and discuss the issues. Finally, it’s important to appropriately document issues and decisions, along with the reasons for those decisions. As a new data source or predictive model rolls into production, questions may arise from regulators or other stakeholders. Having good documentation will make it easier to answer questions that may arise.