While the news media often treats artificial intelligence (AI) as an emerging development, it has been around for more than 50 years and is more integrated into modern life than many people realize. It is used in popular map programs to predict the fastest routes and traffic congestion; by speech recognition programs for texting and word processing; by virtual assistants, spam filters, and email sorting systems; and in many more applications.
What is newer about AI is that the exponential increases in computing capacity achieved over the last five to 10 years have greatly expanded its potential and increased the power of the tools that bring it to life, including natural language processing and machine learning. AI seeks to adapt to its environment—mimicking, replicating, and even replacing human cognition. It can recognize speech, translate languages, perceive visual images, and make decisions. AI can accomplish tasks humans cannot. For example, over 70% of claims can have some flag for fraud. It is commercially impossible to manually check such a high hit rate, but AI can be trained to close the gap.
Such capabilities are why AI is a transformative business technology. For insurers, AI can unlock new business value via enhanced experiences, improved forecasts and predictions, better processes, and uniform consistency of claims.
Yet, even as insurers embrace AI, they need to be wary of being seduced by the AI hype cycle. When insurers get caught up in the media frenzy and worry about missing their window to adopt AI, they are more likely to miss out on AI’s measurable benefits because the fundamentals get lost. In AI’s case, those include conducting a traditional cost-benefit analysis, and addressing serious and complicated ethical and accountability questions.
To achieve meaningful business results by putting AI to work in claims management, insurers must be clear about the problem they want to solve, consider whether AI is the right solution, and recognize the challenges to mitigate when applying AI. Here, we’ll look at these opportunities and issues.
Where AI Works in the P&C Value Chain
AI has shown great promise across the P&C insurance value chain and within claims management. AI can shine in the following functions:
Claims segmentation. Algorithms can quickly calculate the cost and risk of a claim, determining whether it requires special handling based on the timing of its filing, the customer who filed it, and the type of claim. AI can evaluate the impact of these variables over the claim’s lifecycle, generating assignments and escalations as complexities emerge that require intervention.
Predict total loss/required reserves sooner, with greater accuracy. AI can provide fast, accurate predictions about how much a claim will cost. AI model and algorithm accuracy will improve as more data about the types and classes of claims become available. That information can help the claims adjuster comply with best practices, improve quality, and accelerate decision-making. This makes it possible to understand reserve requirements sooner, leading to a more accurate financial picture and better reserve management.
Damage assessments. AI can compare photos of a pristine piece of construction equipment against images of equipment that has been damaged, identify the damaged segment, and predict the likelihood of damage hiding behind the visible segment. It can also research the cost and availability of the required replacement parts from different vendors and forecast the repair timeline. Algorithms could also determine cost and availability of interim rental equipment covered by a policy so insurers can manage insureds’ expectations more quickly and transparently based upon their policies.
Propensity to litigate. Insurers can consider more detailed factors when making litigation vs. settlement decisions with insights derived from AI. AI can identify the pertinent laws in the relevant jurisdiction. It can also identify the law firms handling such cases and their success rate, as well as forecast the ultimate potential settlement.
Subrogation. AI models can determine the likelihood another entity was at fault and identify the relevant insurance company to improve loss offset.
The Caveats
While the benefits of these AI applications are real, so are the potential pitfalls. Insurers must recognize and address the following issues when considering what business problems and opportunities to address with AI.
Responsibility. Accountability for how AI applications behave is a combined business and technology issue. The insurer is responsible for the AI applications it develops. It is also responsible for the decisions made by any AI software it licenses or white-label applications it uses under its own brand. In other words, an insurance company must validate the decisions of all AI tools it uses.
Bias and transparency. All AI models are inherently biased. Bias is inescapable because models are based on past human experiences. Give an algorithm a historic data set and it will learn the biases in that data. It is the insurer who must recognize potential bias.
Regarding transparency, insurers must make certain that models deliver the same results when they are fed the same outputs. Cities and states are passing laws to regulate this. Is the logic of the AI tools transparent? Is the decision it makes or action it recommends and/or takes explainable?
Auditable. Algorithms and applications must be designed from their outset to be auditable. That is, it must be clear what model an application invoked when making a decision.
Incorruptible. Insurers must make sure algorithms and apps cannot be corrupted by users or malicious third parties.
Caveat emptor. AI applications and the technology supporting them are rapidly growing and advancing. That is a mixed blessing. Be aware of the limits of new applications. For example, AI can draw on a wealth of historic data to predict risk in oil and gas fields. That is not the case with clean energy, such as wind and solar, which are much younger industries with unique and rapidly changing technologies. Insurers must carefully vet AI solutions that claim great results in emerging industries and technologies.
Getting Optimal Results
It is vital to recognize that deploying AI is not an IT initiative and should be led by the business. Insurers need a multidisciplinary approach to implementing AI that includes business leaders, actuaries, adjusters, and compliance. To wit: actuaries are the data scientists who can help determine whether an insurer’s AI models deliver consistent results.
For any AI application, the team should do the following:
Conduct a cost-benefit analysis. Few insurance companies take the time to do this exercise, so they don’t understand the timing of their return on investment. Executing a cost-benefit analysis also forces companies to have greater clarity about the business problem they hope AI will solve, whether that’s reducing headcount, improving reserve forecasts, tightening indemnity, etc. Clear-sightedness will be invaluable when cutting through AI hype and selecting appropriate solutions.
Recognize the need for a data strategy and data governance. AI models can’t be evoked without data. Insurance companies have a great deal of data but often don’t recognize the mechanics they need to use their data. Even with all their homegrown data, insurers must be ready to buy high-quality, high-volume data from external sources to feed AI models. Models must consume all types of data (sensor streams, images, video, text) from a wide range of sources.
Ask IT staff what data governance is in place to manage that data. For example, metadata must be created—that is, labels and information about the collected and purchased data. Data also must be transformed into shapes and forms AI models can use.
Use AI where it counts. It’s difficult to develop inexpensive AI solutions. Don’t use it to automate a simple function that will be quicker and cheaper to do with software bots carrying out rote, repetitive keystrokes.
Don’t minimize potential risk exposure. The methods used to correct AI biases may carry their own risks. For example, a company might decide not to use credit scores as a weighting factor because of regulatory changes. In general, credit score calculations are relatively transparent. However, new weighting variables that don’t have long track records are likely to be less understood. Be certain to monitor alternative variables for potential biases.
AI maturity varies across insurers. This includes some companies creating their own apps, others licensing apps from third parties, and some benefiting from AI that they do not even realize they are using. Popular claims-related AI use cases—including routing, triage, fraud, and audit—all typically leverage AI.
The pressure to adopt AI will likely grow as the hype gets louder. Insurers need to tune out that noise. AI can transform the industry, but only when insurers implement AI applications suited to their specific business needs, complete with strong transparency, accountability, and governance.
SIDEBAR
The Potential Impact of Generative AI
Combining generative AI—think AI-powered tools like ChatGPT—with a conversational interface will again provide the potential to significantly transform the P&C insurance industry. It can provide users like underwriters, claims adjusters, and customer service reps the ability to obtain and process information quicker and more accurately.
Generative AI is being leveraged to support chatbots and virtual assistants, with the goal of enhancing the customer experience. By analyzing large data sets, it may further improve current AI efforts around claims processing, risk assessments, and fraud detection. Generative AI will have to take further steps to make it fully effective. For example, it faces challenges deciphering nuances in human communications and multi-tasking multiple requests at the same time, and it does not have the ability to create complex technical documents.
While the industry needs to be cautious in the adoption of AI-powered tools, including generative AI, this potential to further improve claims operations across the industry should be watched carefully.