Skip Ribbon Commands
Skip to main content



 ‭(Hidden)‬ Catalog-Item Reuse

To Err Is Human: Using AI and Avoiding Unintentional Misinformation

Unintentional misinformation can seep into the underwriting process and cause significant consequences—here's how artificial intelligence (AI) helps reduce this rising risk.
Sponsored by
to err is human: using ai and avoiding unintentional misinformation

It is time to broaden the discussion about how artificial intelligence (AI) can reduce misinformation in insurance workflows and operations. Conventional views about misinformation focus on bad actors submitting false claim information to reap a financial reward. In those cases, generative AI can help to reduce claims fraud.

However, a less-discussed area of misinformation is actually more common: unintentional misinformation. This can occur during the underwriting and quoting processes, leading to dire consequences, such as wrong or inadequate insurance coverage or no coverage at all. Yet, with the proliferation of AI, there is a real opportunity to correct this type of misinformation and significantly improve outcomes for agents, insurers and commercial insurance policyholders.

Consider this scenario: An agent is guiding a contractor to acquire insurance coverage. The agent asks the contractor a series of questions and does online research about the business. The business gets classified as a concrete flatwork contractor specializing in sidewalks, patios and driveways based on responses provided by the client and the limited information the agent could collect. However, the contractor actually is performing concrete foundation work. As a result of misclassification, the contractor pays a much lower premium than necessary to cover risks, leaving them exposed.

There are two ways such a scenario can occur. First, the client may accidentally provide inaccurate data due to not fully understanding the questions or not knowing the answers. Or, the agent may face a lack of information, perhaps because they didn't ask enough follow-up questions or the answers provided by the client were incomplete.

Whether the eligibility information is incorrect or insufficient, the ramifications can reverberate throughout the insurance ecosystem. Insurers may inadvertently underprice policies, exposing themselves to unexpected losses in the event of a claim. They could overestimate risks based on erroneous information, which can lead to inflated premiums, alienating clients and undermining competitiveness. Or, coverage based on unintentional misinformation could be inadequate and leave the business vulnerable to financial strain in the face of unforeseen events.

The Rising Risk of Unintentional Misinformation

Unintentional misinformation can seep into the underwriting process through various channels. Human error, incomplete information, misinterpretation of questions and changes in circumstances can all contribute to the dissemination of inaccurate risk-quality data. But the biggest exposure comes from agents trying to juggle getting complete information from their customers while still providing a positive policyholder experience.

Commercial insurance is complicated, but clients expect the same on-demand experience they have become accustomed to when purchasing other products and services. Agents are trying to obtain a complete view of a business while asking the fewest questions possible and streamlining the customer's experience.

In an effort to avoid inundating clients, agents might limit their follow-up questions and turn to online searches to fill information gaps. But while large businesses often have a robust online presence, small or midsize businesses may be subject to the “small data" problem, meaning they lack a large enough online footprint to make it easy to conduct manual research about risk factors—and what information does exist could be spread across disparate sources.

Agents do not have the bandwidth to do extensive online searches, especially in small and midsize commercial insurance, which is typically more transactional. If the information online is incomplete or out-of-date, it could lead to the wrong classification for the business.

An Agent's Guide to AI Capabilities

AI solutions can reduce unintentional misinformation by quickly scanning many structured and unstructured data sources to get a more complete risk profile. AI can also limit the number of questions clients must answer, improving the service experience. While AI has been around for years, it has flourished only recently, making significant impacts in every industry.

For agents wanting to begin to take advantage of AI, it can seem daunting and overly technical, particularly when solution providers use confusing and overlapping terms. However, it is important to become familiar with these terms and understand how they can support efficiency in insurance operations.

Here are five of the most common terminologies and how these AI processes can improve quoting and underwriting. Many AI solutions feature either some or all of these capabilities:

1) Autonomous AI agents. These can assume distinct personas, such as risk assessment or document and content classification agents, and deliver accurate responses to risk questions. They can function as role-based assistants to agents and underwriters, learning instantaneously and adapting to their tasks.

2) Data validation and enhancement solutions. These sift through vast datasets to validate the accuracy of information provided by applicants. They cross-reference internal and external data sources, including historical claims data, financial records and industry-specific benchmarks to identify inconsistencies and flag potential discrepancies for further investigation.

3) Large language models (LLMs). LLMs can review unstructured data, such as text fields in application forms or policy documents, extract relevant information, identify key insights and discern nuances in language—thereby minimizing the risk of misinterpretation and ensuring comprehensive risk assessment.

4) Predictive analytics. This can analyze historical data and patterns and anticipate future risks with unprecedented accuracy. These models factor in a myriad of variables, from market trends and regulatory changes to socioeconomic indicators, enabling insurers to proactively adjust underwriting strategies and pricing models to reflect evolving risk landscapes.

5) Continuous learning and adaptation. These are defining features of AI. Through continuous feedback loops and real-time monitoring, AI underwriting solutions refine their algorithms, incorporating new data and insights to enhance predictive capabilities and mitigate the impact of unintentional misinformation.

An AI solution with these capabilities can improve the policyholder's outcome. Looking back to the earlier example of an agent guiding a contractor through the insurance purchasing process, the contractor needs only to provide their business name and address.

Using just this information, the AI solution can answer a large number of risk-assessment questions. The solution can also scan many disparate sources and properly identify that the contractor is doing concrete foundation work. As a result, the agent can provide the contractor with the correct premium for adequate protection if an incident occurs.

Incorporating AI into Agency Workflows

Agents can take two approaches to using AI: They can partner with insurers that are implementing these solutions into their underwriting processes, or they can use AI within their own agencies for gathering information and quoting coverage. While agents should try to implement at least one approach, ideally, they should strive to do both.

For agents who are unfamiliar with AI, partnering with insurers using the technology is a great first step, and many insurance organizations are upgrading their operations to include this technology. Because AI enables faster responses and augments manual processes, insurers can underwrite a business faster. This allows agents to reduce the amount of information needed from clients and provide them with coverage in a matter of days or even hours instead of weeks.

Agents working with insurers using AI can also offer their clients more tailored coverage and uncover other potential risks. For example, if an accounting firm is located next to a restaurant with fryers, could this be an additional risk to the firm? AI can quickly identify those types of exposures.

But agents should also take the leap and begin implementing AI solutions internally. These tools will allow them to move from being primarily data collectors to trusted advisers. Before an agent has a meeting with a business client, they can run that business through the AI solution to get an initial overview of the prospect's risk profile. For renewals, agents can learn if the client's business has changed even before they reach out with the renewal notification. Consultations with prospects or clients no longer revolve around asking questions, but rather are productive discussions about the business's unique risks and how best to plan for and offer protection.

Agencies that want to start using AI should review their processes and see which areas require a significant amount of manual data entry, rekeying or research. These areas are prime opportunities to incorporate AI. When researching solution providers, ask how easily the vendor can customize the solution to meet the agency's needs, such as tailoring risk questions to reflect the types of businesses the agency works with.

Also, determine if the provider welcomes feedback. A good agent-vendor relationship should not just be a one-way transaction but rather a partnership, sharing information so the provider can make adjustments to enhance the solution.

Overcoming the Hallucination Barrier

Unintentional misinformation is not just the result of human error. Artificial intelligence can sometimes generate inaccurate information or answer a question incorrectly. This is called hallucination.

Remember, AI does not generate answers in a vacuum. It is trained on vast quantities of data. Nevertheless, sometimes during the training, unintended biases are programmed in, or the AI tool wants to deliver an answer but lacks the necessary information to make the correct assessment. While AI solutions are more accurate than manual processes, agents still need to remember these are machine-based interactions, and there could be errors or fabricated facts.

Agencies can work with solution providers offering data transparency to prevent unintentional AI misinformation. “The answer is in the algorithm" is not an acceptable answer to receive when you ask a provider how the AI produced a response to a certain question. Agencies should partner with providers that can easily identify and provide the data sources and information used to answer a particular risk question. This allows agents who are uncertain of a particular item to easily view the sources themselves and come to their own conclusions.

When working with insurers who use AI underwriting tools, agents should also ask about the built-in transparency of their risk-assessment platforms. If an agent delivers a quote and the client has questions about how a particular price was determined, the insurer's underwriting team should be able to point to the exact information used to assess and quote the exposure.

Unintentional misinformation, though not malicious, can cause real damage to agents' relationships with insurers and policyholders. If an insurer undervalues a risk, they could have to pay out a large, unexpected claim. A client might pay more than they need to or not have enough coverage, leaving them vulnerable if an incident occurs.

By taking advantage of AI internally and by working with insurers using this technology in underwriting, agents can reduce unintentional misinformation and provide a better customer experience.

Chris Schrenk is chief underwriting officer at NeuralMetrics, which enables actionable risk-assessment intelligence for property & casualty insurers, brokers and agents.

Friday, June 28, 2024
Digital Edition