The potential of gen AI in insurance: Six traits of frontrunners

Commercial lines insurers are becoming increasingly enthusiastic about deploying generative AI (gen AI), in light of its potential to contribute between $2.6 trillion and $4.4 trillion to the global economy annually—an amount around the size of the United Kingdom’s entire GDP in 2023 (approximately $3.3 trillion)[1]. For insurers, gen AI could enhance areas like underwriting, predictive risk assessment, and personalization. As an industry, insurance stands to benefit significantly, given its wealth of unstructured data—such as PDFs, visuals, Word documents, and web pages—and the prevalence of manual tasks across various stages of underwriting and claims processing.

Insurance companies are acutely aware of these opportunities. In a recent McKinsey survey of more than 50 leaders from the largest European insurer groups, more than half of the respondents say gen AI could lead to productivity gains of 10 to 20 percent, premium growth of 1.5 to 3.0 percent, and improvement in technical results by 1.5 to 3.0 percentage points. Meanwhile, a third of insurers indicate they have initial gen AI use cases in production, 20 percent describe their AI maturity is advanced, and 60 percent say their traditional data is “evolving.” Relatedly, we find that the unique mentions of AI and gen AI in the major insurer groups’ annual reports have more than doubled from 2022 to 2023.

The question is no longer whether to adopt gen AI but how to do so effectively. Drawing upon our work in supporting gen AI transformations, we have identified the following six traits that organizations leading in gen AI adoption are performing well.

1. Identify a core business area for gen AI to enhance key operations

Leading organizations strategically apply gen AI to transform their core business processes. The question to ask is “What top-priority function can gen AI improve to enhance our work fundamentally?”

For example, McKinsey identified knowledge management as central to its operations, leading to the creation of its knowledge platform, Lilli, which reinvented the management of content. Lilli, which serves 45,000 users, has answered more than 4.5 million queries with insights from more than 200,000 documents and has reduced cost per query by 96 percent. The journey has offered several important lessons:

  • Define the next horizon across functional teams and cloud strategy. A cross-functional team and a robust cloud strategy are crucial for the successful launch and scalability of gen AI use cases.
  • Understand the trade-offs between cost and performance. Understanding costs versus performance is essential. Focus on selecting individual models and components that contribute to answering prompts, while ensuring scalability and flexibility to avoid vendor lock-in.
  • Understand the data challenge. Addressing data-related questions before building the tool is essential. For the development of Lilli, this meant tackling the data retrieval challenge, as content is stored primarily in PPT format, and ensuring efficient extraction from the existing data structure.
  • Invest twice as much in change management and adoption as in building the solution. Investing in change management and adoption is necessary. Users effectively leveraging gen AI can boost productivity by over 20 percent, and scaling this across 45,000 users requires codifying and refining training.

2. Deliver E2E domain-level transformation to drive value capture

To unlock significant investments and drive value capture, leading organizations focus on delivering end-to-end domain and subdomain transformation over individual use case builds. This involves identifying end-to-end functionalities within a domain that link multiple AI and gen AI use cases. The question to ask is “Which domain would most benefit from E2E domain-level transformation?”

For example, while applications like claims eligibility checks, enhanced fraud detection, settlement optimization, and improved customer experience through gen-AI-enabled chatbots are valuable on their own, a full-scale E2E transformation of the claims-processing domain could, by our estimate, yield up to 14 times the impact. This is achieved through synergies where different use cases interact and reinforce each other, creating a more cohesive and efficient system.

3. Focus on gen AI and AI with integration of other technologies

While tools like ChatGPT have generated significant hype around gen AI, our estimates indicate that the overall value from AI in insurance is divided into approximately 60 to 80 percent from traditional AI and 20 to 40 percent from gen AI. Leading organizations understand that gen AI should complement, not replace, traditional AI. The question to ask is “How do we build a systematic ‘mapping’ of our business that identifies improvement opportunities regardless of whether they are powered by gen AI, traditional AI, or other technologies?”

For example, call center analytics rules and interactive voice response systems can be enhanced with gen-AI-powered intent identification and action. Similarly, an AI-enabled claims severity engine can be paired with proactive gen AI claims prevention by integrating weather data and personalized outreach. This holistic approach ensures that organizations leverage the strengths of both AI and gen AI, alongside other technologies, to drive comprehensive improvements across their operations.

4. Develop a strategic view on build, buy, or wait

Leading organizations develop a strategic perspective on when and where to build in-house gen AI solutions versus purchasing off-the-shelf vendor solutions. This balance helps harness the benefits of external technologies while mitigating the risk of vendor lock-in. The question to ask is “In what areas of our business do we want to build competitive advantage and therefore develop in-house gen AI applications?”

In the context of dealing with large language models that incorporate proprietary data, it may be advantageous to develop a bespoke solution. Conversely, for gen AI applications that integrate into enterprise-grade platforms like customer relationship management, off-the-shelf solutions can be more practical, especially if they require minimal or no customization. A strategic approach ensures that organizations invest in building internal capabilities where it counts while leveraging external solutions to enhance efficiency and effectiveness.

5. Build reusable code for scalability and maintainability

For organizations choosing to build rather than buy gen AI solutions, it is crucial to develop truly reusable code components and workflows, similar to automated assembly lines in car manufacturing. These reusable components can be packaged into various modules and application archetypes tailored to specific gen AI applications. The question to ask is “What modular components do we need to start building, and how do we create a recipe for this at scale?”

6. Enable risk management to accelerate operations

To effectively mitigate risks, it is important to involve the internal risk team in the product development process from day one. The question to ask is “How do we use the development of gen AI applications to rethink and enable risk management?”

For example, a global bank that recently implemented a gen AI customer-facing chatbot conducted more than 50 meetings with its risk stakeholders from the outset. The team defined risk frameworks across key categories with fail-safes, including the ability to monitor performance in real time and pull the plug if issues arose. This collaborative approach embeds risk considerations into the development process from the beginning.

There is no silver bullet for scaling gen AI, but its immense potential is set to shape companies’ performance. Leading organizations are distinguishing themselves by implementing bold, end-to-end, domain-level, and enterprise-scale transformations. They strategically balance build-versus-buy decisions, develop reusable code, and foster strong cross-functional teams. As the possibilities of gen AI continue to evolve, organizations at the forefront of adoption exemplify these practices.

Carlo Giovine, Phil Hudelson, and Sid Kamath are partners in McKinsey’s London office, where Khaled Rifai is a partner in the Berlin office.

The authors would like to thank Victoria Longhi for her contribution.

Copyright © 2024 McKinsey & Company. All rights reserved.


[1] World Bank, 2024.

Connect with our Financial Services Practice