What are insurance product filings?
Insurance product filings are publicly available documents submitted to state departments of insurance that describe new insurance products or revisions to existing insurance products for regulated types of insurance, like homeowners insurance. Insurance product filings are a rich source of information for anyone involved in the filing process, including insurance companies, managing general agents, compliance professionals, and actuaries. Currently, analysis of these documents is limited, as much of the text is unstructured and difficult to work with. Artificial intelligence (AI) and large language models (LLMs) can be used to unlock that data and allow for novel insights by focusing on specific parts of a filing, like objections that are raised by state regulators when filings are submitted for approval. This new technology can be tapped to analyze the contents of filings in ways that were previously not feasible and may be used to increase a company’s regulatory insurance intelligence, giving them a competitive edge in a highly regulated environment.
Why analyze insurance filings?
Insurance is highly regulated, especially personal lines products, and many of those products have to be filed with DOIs. The filing process generally allows for a period where the DOI can ask questions to the filer about the filing, which are called objections, and the filer provides responses to those objections. Objections usually increase the overall review time for a filing, which may ultimately impact when the product can be implemented.
For example, if the filing includes a proposal for a rate increase, and objections are made throughout the filing process pushing the proposed effective date beyond the date contemplated in the rates, it may result in the company not collecting the premiums it projected. It may also result in the company needing to revise its analysis, which may result in additional resources, delays, and administrative costs for the company. If the DOI indicates other changes are required to the filing, such as changing the proposed rates or policy contract language and consumer notices (called forms), then the company may incur additional administrative costs to update and re-test the proposed rates or forms that have already been programmed. It is also possible that if the DOI discovers too many issues to be addressed via objections, the DOI may disapprove the filing, forcing the filer to start over, incurring filing fees and other expenses.
Submitting rates, rules, forms, or documents that do not comply with state regulations creates additional workload for the DOI to draft an objection explaining what is not compliant for the filer to address before the review can continue. This not only lengthens the time for the filing to ultimately get approved but impacts the DOI’s ability to review other company filings, causing delays for other insurance companies. Filers can use AI and LLMs to develop filings that:
- Address state regulatory requirements
- Meet Actuarial Standards of Practice (ASOPs)
- Strengthen the filing company’s relationship with the DOI
Reviewing the laws, regulations, checklists, and other published information that is publicly available allows the filer to be informed about the requirements and prepare a filing that can be efficiently reviewed by the DOI. This reduces the time and cost for the regulatory review process, getting products to market faster and premiums collected sooner.
Objections from historical filings are also a rich source of information that highlights discussions between filers and DOIs about pertinent laws, regulations, and “desk rules.” Desk rules are DOI interpretations of regulations, positions, and opinions that are usually not published and only learned through experience.1 Reviewing relevant objections from historical filings that are presented in an organized manner can provide insights into how a DOI enforces and interprets state laws and regulations so that the filer can consider this information while preparing the filing, creating efficiencies for both the filer and the reviewer.
Analyzing the filings
Most regulatory information tools available today provide metadata2 about filings, such as submission date, approval date, insurance company name, line of business, and the amount of the approved rate. Many also provide the correspondence between the DOI and filer, as well as the rates, rules, forms, and other support that were uploaded to the DOI. For rigorously regulated states that require an extensive amount of filing support, filings may contain tens of thousands of pages, offering a treasure trove of information. While the filing metadata is useful on its own, the additional latent data within the filings is usually untapped because of the complexities and cost to extract, sift through, and use t. Searching for keywords takes time and requires the reader to go through a trial-and-error approach of different words or numbers to find information that must then be read to determine whether it is relevant and then summarized for the product manager, actuary, or individual developing the product and/or preparing the filing support. This “old-fashioned” process of manually sifting through filings is labor-intensive, time-consuming, inefficient, and costly. AI and LLMs can be used to efficiently unlock the voluminous data in filings and drive new insights.
The high-level process used to analyze each filing using AI and LLMs is summarized below:
- Parse and extract the data of interest.
- Send a prompt3 to the LLM and save the completion.4 Figure 1 shows an example of input text transformed into the output text using an LLM.
- Embed the completion into an embedding.5 Figure 2, below, shows an example of the result when the output from step 2 is used to create embeddings.
Figure 1: LLM usage example
Figure 2: Text embedding example
Since embeddings are designed to reflect meaning via vectors, machine learning techniques are employed to deduce useful patterns, which are then further reviewed by appropriate experts.
We downloaded filings from the National Association of Insurance Commissioners (NAIC) System for Electronic Rate & Form Filing (SERFF) website submitted to California, New York, and Texas DOIs from January 2020 through June 2023, for the residential property line of business.6 The content of the filings (filing data) was processed through AI and machine learning algorithms to find common themes or topics. The topics were then reviewed for reasonableness by compliance professionals (lawyers, paralegals, and former regulators), actuaries, and product managers.
Results of the analysis
The following tables illustrate the results for the most common objections from the California Department of Insurance (CDI), the New York Department of Financial Services (NYDFS), and the Texas Department of Insurance (TDI). The Topic column describes the common concern among many analyzed objections and the Explanation column provides additional commentary and context about that Topic.
Figure 3: Common objection topics in California
| # | Topic | Explanation |
|---|---|---|
| 1 | Request by the CDI to waive the deemer provision. | Virtually all filings in California will take longer than the “deemed approved” date, which is 60 days past the submission date. The DOI asks that the filer waive the “deemed approved” date to allow more time for review. |
| 2 | Requests for additional information about nonrenewing policies. | The CDI is gathering details about policies being nonrenewed to understand the impact on rate, consumers, and property insurance availability in the state. |
| 3 | If a filer fails to provide information by the deadline, the filing will be disapproved and closed. | It is important that filers respond to CDI objections or request an extension to the response deadline provided by the DOI. Filers should have an active line of communication with the CDI to resolve issues such as an extension request or to clarify an objection. |
| 4 | Requests for information about underwriting guidelines. | Underwriting guidelines are required to be filed in California for new programs and for proposed changes to the underwriting eligibility. The CDI reviews underwriting the guidelines for compliance with regulations and consistency with the rates, rules, and forms, and other filing support. |
| 5 | Requests to fill out standard exhibits correctly and to ensure they reconcile. | The CDI requires that each new program or rate filing include the CDI rate applications, CDI rate template, and for programs with historic experience data, CDI Standard Exhibits. The DOI provides instructions and checklists to assist filers with completing these and preparing a compliant filing. |
Figure 4: Common objection topics in New York
| # | Topic | Explanation |
|---|---|---|
| 1 | Requests that changes to policy form wording are accompanied by side-by-side comparisons and/or redlined versions. | The NYDFS requires a side-by-side comparison of proposed rates from the prior program, or competitor program, that was used as the basis of the support. The NYDFS also requires marked-up versions of the proposed forms, from the versions that were previously approved by the NYDFS. This enables the NYDFS to efficiently concentrate its review on only the proposed changes that deviate from what has already been approved by the NYDFS. |
| 2 | Requests a dislocation analysis (e.g., rate change impact at the policyholder level) and other statistics. | The NYDFS requires support summarizing the impact of the proposed rate changes on individual policyholders that demonstrates that no individual policyholder will receive a rate increase of more than 25%. The 25% limitation on individual policyholders is a NYDFS position. |
| 3 | Requests additional information about how rate changes affect certain territories. | The DOI is concerned with the effects of rate changes and how they are expected to impact certain regions with a specific demographic or exposure (e.g., coastal). |
| 4 | Filers should comply with the filing instructions provided by the NYDFS. | Filers should review all filing instructions to ensure the submission is complete and to avoid procedural problems and rejection of their filing. |
Figure 5: Common objection topics in Texas
| # | Topic | Explanation |
|---|---|---|
| 1 | Requests to provide forms that meet TDI requirements. | The TDI has specific requirements about form names and numbering schema. The TDI will review the forms language and numbering schema for compliance with state laws and regulations. |
| 2 | Requests to provide accurate filing numbers when revising or adopting other filings. | When filers refer to other filings, the TDI requires Texas-specific filing numbers be provided. As a general rule, unless the state uses SERFF filing numbers as its state filing numbering system, it is best practice to use the state filing number rather than SERFF when referencing filings. |
| 3 | Requests by the TDI to include actuarial support, especially for things like fees and surcharges. | The TDI requires that actuarial support for proposed rates include factors and fees. The TDI cannot review whether rates are compliant with state laws and regulations without support that contains the appropriate level of detail. |
| 4 | Requests about underwriting criteria and how that affects new or existing business. | Texas prohibits insurers from denying, cancelling, or nonrenewing policies based on credit information alone without considering other risk characteristics. |
Conclusion
The data in Figures 3, 4, and 5 can be used to prepare a more complete and compliant filing by considering and addressing the above highlighted issues before the filing is submitted, reducing the number of objections. While some of the topics and explanations may seem obvious (e.g., Figure 3, Topic 5; Figure 4 Topic 4), such as following instructions, completing checklists accurately, and adhering to guidelines, many historical filings reviewed included objections to address these avoidable missteps. Furthermore, filers that include proposed rate changes must include support detailing how the proposed rates were developed (e.g., Figure 5, Topic 3), that also meets ASOP 41: Actuarial Communications, generally accomplished through an Actuarial or Filing Memorandum. The Actuarial or Filing Memorandum and supporting exhibits that detail the rate development should be sufficiently detailed with sources of data and formulas so that another actuary and the DOI can follow the analysis, validate that it meets ASOPs, is actuarially sound, and is compliant with state laws and regulations.
Many of the topics require specific knowledge about the state, its laws, and how the state DOI operates. For example, to avoid objections from the CDI on topics requiring more information related to underwriting eligibility (e.g., Figure 3, Topic 2, and Figure 3 Topic 4), it is necessary to review and understand the California Insurance Code (CIC), California Code of Regulations (CCR), CDI bulletins, and positions communicated through objections related to underwriting guidelines. Contemplating these requirements before submitting the filing and providing this information with the initial filing may avoid potentially time-consuming and costly efforts to respond to and address CDI objections. The CDI has been expanding the information required to be submitted with a rate filing, such as adding a Questionnaire for Homeowners or Residential Property, as well as a Mitigation in Rating Plans and Wildfire Risk Models Questionnaire to gather information related to how the company underwrites, rates, and notifies consumers about wildfire risk and mitigation opportunities. The additional information that must be submitted has increased the complexity of filings in California in addition to expanding the volume of material that the CDI reviews for compliance with CIC, CCR, and current CDI positions.
1 Some DOIs provide bulletins, instructions, or checklists that document the DOI interpretation, positions, and opinions. These may become outdated and change over time as the staff at the DOI changes.
2 Metadata is information that describes data sources, authors, owners, licenses, and relationships to other data about a dataset, but it does not include the data itself.
3 A prompt is a message that may include data and provides instructions to a LLM to perform a task.
4 A completion is the result of a successful prompt to a LLM.
5 An embedding is the output of a model that converts text to a numerical representation (a vector, or list of numbers) for AI, machine learning, or statistical models to analyze it more efficiently.
6 Filings included the following NAIC types of insurance: homeowners, homeowners combinations, other homeowners, mobile homeowners, personal property (fire and allied lines), dwelling fire/personal liability, condominium, owner-occupied, and tenant. There were over 5,000 filings that met these criteria for state, period of time, and line of business.