Leveraging artificial intelligence (AI) in claims departments has become a pivotal strategy for enhancing efficiency and improving claims outcomes. However, the decision between building an in-house AI solution and buying one is filled with complexity. Below, we explore the top five mistakes that property and casualty (P&C) insurers make when building claims AI, shedding light on the challenges, and underscoring the advantages of purchasing a ready-made solution.

1. Lack of strategic focus: The search for a nonexistent silver bullet

The allure of AI is undeniable. Yet its implementation is not as straightforward as industry marketing would have you believe. Companies often fall into the trap of expecting AI to be a “set it and forget it” cure-all that can be implemented with immediate results. This often leads to:

• Diluted efforts across multiple tasks: Attempting to simultaneously apply AI across too many areas dilutes resources and focus, detracting from areas where AI could have the most significant impact.
• Misalignment with claims workflow: A nuanced understanding of where AI will fit into the claims process is vital. Without this, there's a risk of developing solutions that are misaligned with actual workflows, resulting in poor AI utilization and lower return on investment (ROI).
• Losing faith in AI: When AI initiatives fail to meet expectations because of a lack of strategic direction, stakeholders may lose trust in AI altogether. Incorrectly assuming that AI is not useful for their needs, they may prematurely forgo future adoption. This overlooks the potential benefits of AI when implemented strategically to align with business objectives.
 

2. Underestimate cost and time investment

The development of claims AI is not a simple research project. It involves a significant investment in resources and development that could last years before reaching a viable solution. These items are the major cost drivers when building:

• Data complexity: The success of AI heavily relies on the quantity and quality of data. Unstructured data, such as claims notes, which have the richest claims insights, requires extensive processing, cleaning, and annotating, which can be a daunting task, even while leveraging off-the-shelf large language models (LLMs).
• Talent scarcity and costs: In a highly competitive job market, insurers have found it challenging to hire and retain the necessary diverse team of subject matter experts, including data scientists, actuaries, and data engineers, all of whom play crucial roles in developing and maintaining claims AI. The unique nature of insurance heightens the demand for these skill sets, introducing a steep learning curve for professionals transitioning from other industries.
• Underestimated infrastructure costs: For insurers venturing into AI, the initial projections often fall short in accounting for the needed infrastructure. This can include procuring high-end computing resources, such as graphics processing units (GPUs), which are both essential and costly, for the effective training of modern AI models. Beyond hardware, there is a robust need for data security measures to protect sensitive data, adhering to regulatory compliance standards.
• Integration and maintenance hurdles: The integration of AI with legacy systems, deployment, and maintenance present significant challenges, often underestimated in initial planning phases.

3. Teams lacking domain expertise

Building claims AI is not a task for actuaries alone, as it often demands several specializations. Claims AI requires a collaborative effort from data scientists, engineers, and claims professionals, each bringing a unique skill set to the table. The team also requires extensive domain knowledge of insurance claims data, where the current environment may not match the historical data used to create models. Insurers often encounter challenges in equipping non-actuarial staff with this specialized knowledge, leading to a steep learning curve. This knowledge gap can result in modeling mistakes, resulting in AI that performs well on historical data but fails in production. Additionally, these expertise gaps can introduce significant delays in the project timeline, emphasizing the importance of comprehensive domain knowledge in claims AI.

4. Difficulty measuring outcomes

Measuring the success of claims AI is complex and requires more than simple comparisons of year-over-year results. Simple metrics such as average claim severity can obscure the true impact of AI due to annual fluctuation in the type of claims. This complexity is compounded by the dynamic nature of claims data and industry forces, which may shift, causing model drift—a scenario where the model's accuracy degrades over time as the underlying data patterns change. Effective monitoring and analysis are crucial to detect such drifts early and implement model adjustments promptly.

5. Post-deployment challenges

Launching an AI model into production marks the beginning of an ongoing journey to sustain its effectiveness in terms of reliability and scalability. After deployment, AI models enter a critical phase where ongoing observation is required to guarantee their adaptability and continued performance. This phase stresses the indispensable role of specialized machine learning operations (MLOps) professionals, who are experts in managing the life cycle of AI models. This proactive approach preempts potential issues, optimizing model performance, and ensuring that the AI continues to meet its intended objectives. In building claims AI, many insurers overlook the post-deployment period, resulting in diminishing model performance over time and missed opportunities for continuous improvement. This oversight can lead to a failure in realizing the full potential of AI.

Conclusion: The strategic advantage of buying

When considering the numerous challenges—from project risk due to strategic missteps and underestimated costs to the complexities of data management, integration, and continuous improvement—the advantages of buying an AI solution become even more compelling. By choosing to buy, insurers can leverage advanced AI capabilities within months and at a fraction of the cost required to build internally. Furthermore, partnering with a leading AI platform offers insurers access to ongoing innovations, industry benchmarks, and a level of expertise that would be costly and time-consuming to develop internally. In an industry where speed and efficiency are paramount, the decision to buy versus build is not just strategic—it's essential for establishing the foundation for future success and competitiveness in the P&C insurance industry.

We are here to help

At Milliman, we recognize the challenges insurers face in developing effective AI solutions for claims management. Our claims AI solution, Milliman Nodal, is a comprehensive solution that utilizes AI and machine learning to analyze both structured and unstructured data. With Nodal, insurers gain powerful insights into various aspects such as jumper claims, litigation likelihood, and excessive medical costs. Through intelligent, automated claims triage and cost benchmarking, Nodal helps improve internal processes, bridge skill gaps, and effectively manage the ever-growing medical and litigation expenses. We're dedicated to supporting insurers in leveraging AI to drive innovation and competitiveness within their organizations.