Adopting AI in insurance: A strategic framework

From watercoolers to boardrooms, Artificial Intelligence (AI) is the topic on everyone's minds. Advocates of AI tout incredible productivity gains and cost savings that can be achieved. On the flip side, AI risks are also regularly reported. Headlines warn of ballooning server costs, intellectual property lawsuits, and embarrassing missteps caused by poorly planned, hastily deployed, or under-tested AI deployments.

The reality is that even with those risks, the adoption of AI is increasing. In the most recent CGI Voice of Our Clients insights, 40% of insurers cite AI as the top innovation priority over the next three years. Organizations that take a wait-and-see approach risk losing the first-mover advantage and seeing competitors reap the benefits promised by the AI revolution. How can insurance organizations balance the costs and benefits of AI adoption?

Costs of AI in insurance

Insurers must consider the direct "hard costs" and indirect "soft costs" involved when evaluating AI.

Hard costs

A hard cost is a new activity that has a quantifiable expense directly associated with it. These hard costs are tangible and can be accurately calculated, making them essential when budgeting for an AI initiative. Examples of hard costs include:

• Training and fine-tuning costs - For narrow AI, models must be trained on a company's specific data and processes, much like onboarding a new employee. This custom training is expensive but allows the model to provide maximum value that is tailored to the insurer's business. While pre-trained generative AI (GenAI) models are available, they often require fine-tuning or prompt engineering to adapt to insurance-specific tasks. This process, while potentially less intensive than training from scratch, still incurs significant costs in terms of expertise, computational resources, and data preparation. Additionally, keeping up with rapidly evolving GenAI models may require ongoing investment in retraining or fine-tuning as new versions are released.
• Staffing costs - AI experts are in high demand and short supply, commanding premium salaries across both narrow and GenAI domains. For narrow AI, specialists in machine learning (ML), data science, and domain-specific AI applications are crucial. For GenAI, experts in natural language processing, prompt engineering, and large language models (LLMs) are increasingly sought after. Both fields require professionals who understand AI ethics and governance. Turnover can be high as skilled staff pursue better opportunities, particularly in the rapidly evolving GenAI space. Experienced AI experts are invaluable for both types of AI: in narrow AI, they ensure models are accurately trained and avoid common pitfalls, while in GenAI, they can effectively fine-tune models, design robust prompts, and implement necessary safeguards. Their expertise leads to faster productivity and helps navigate the unique challenges of each AI type in the insurance context.
• Runtime and infrastructure costs - The computational requirements and associated costs vary significantly between narrow AI and GenAI models. Narrow AI models in insurance applications often have moderate runtime costs once deployed but may require substantial computational resources during the initial training phase. These models typically process specific, structured data inputs, making their resource needs more predictable.
  GenAI models, particularly LLMs, have a different cost profile. Initial training requires immense computing power and is usually done by the model providers. For insurers, the main costs come from fine-tuning, deployment, and inference (runtime use). Contrary to common belief, many modern LLMs are quite efficient at inference, especially when optimized. However, they may require more powerful hardware and specialized infrastructure to run effectively at scale.
  For both types of AI, cloud computing costs, data storage, and network bandwidth must be considered. The choice between on-premises deployment and cloud-based solutions can significantly impact the overall cost structure. Selecting the right model and infrastructure involves balancing these various costs against the model's capabilities and the specific needs of the insurance use case. It's crucial to conduct thorough cost-benefit analyses, considering both immediate expenses and long-term scalability.

Soft costs

 A soft cost is an indirect cost of deploying a model. While hard costs are direct and quantifiable, soft costs can be more significant and challenging to quantify. Soft costs tend to be unexpected and typically represent negative impacts or opportunity costs that may not be immediately apparent. Examples of soft costs include:

• Opportunity costs: Failing to adopt AI could give competitors a potential speed, efficiency, and profitability edge to undercut pricing or poach customers.
• Reputational costs: Poorly deployed AI can lead to harmful errors, such as wrongly denying healthcare coverage. Regulatory scrutiny, lawsuits, and brand damage can result in both financial costs and increased staffing needs.
• Competition costs: AI allows competitors to reduce costs through automation, customize pricing for more competitive rates, and provide better customer service — potentially leading to stolen market share.

Other key considerations of AI in insurance

AI Regulations

In the United States, no federal AI laws exist yet for insurance, although some states have started enacting their own rules. The National Association of Insurance Commissioners (NAIC) has published AI principles that insurers should follow.

Outside the U.S., AI regulation seeks to balance concerns about innovation with concerns about individual privacy and accountability. The EU AI Act provides a rules framework that applies to all industries, including insurance, and seeks to apply risk-adjusted regulation to the AI technology space. The Digital Operational Resilience Act (DORA) introduces stringent requirements for firms to manage and mitigate risks related to their digital operations.

Ethical AI

Beyond just compliance, insurers must ensure their AI models are transparent, unbiased, and ethical—especially when used for critical functions like claims decisioning that impact people's lives. Governance processes are essential.

The benefits of AI in insurance

Despite the costs and potential risks, AI offers insurers significant advantages across virtually all business functions when implemented effectively. Companies that successfully incorporate AI into their business stand to gain processing efficiencies, competitive advantages, and improved customer service.  

Claims processing 

AI capabilities like text recognition and sentiment analysis can drastically improve claims workflows through automation. There are several commercial products on the market, as well as in-house capabilities that companies have developed over time.

Rating and underwriting 

In ratemaking, AI enables advanced pricing models that can granularly segment and quantify risk at the individual policyholder level, allowing more personalized and actuarially precise premiums. Ratabase360 allows insurers to incorporate the results of AI models into premium calculations. Ratabase Actuarial extends this capability, allowing insurers to use AI model results in ratemaking and experimentation. 

For underwriting, AI surfaces relevant data insights upfront, streamlining workflow by minimizing low-value manual reviews so underwriters can focus on high-stakes cases that require human expertise. Across both functions, AI enhances pricing sophistication, decision accuracy, and productivity.

Fraud prevention and detection

AI also unlocks significant value in the underwriting process through advanced fraud detection and intelligent data triaging capabilities. AI models can automatically surface high-risk cases requiring underwriter review by analyzing submissions for anomalous data points and aggregating pertinent information from internal and external sources. This allows underwriting teams to optimize their time by focusing only on the submissions and risks that demand professional expertise and judgment. In the claims process, AI can help identify potentially fraudulent claims on an individual basis by examining claims in the aggregate to identify patterns of activity that could indicate fraud.

Three paths to AI adoption

With the costs, benefits, and risks understood, insurers can pursue three main approaches for AI adoption:

1. Build your own solution

Building proprietary AI models involves developing exclusive solutions tailored to an organization's unique needs and requirements. It offers the highest degree of customization, potential competitive differentiation, and the ability to generate valuable intellectual property. However, this bespoke approach comes with substantial upfront costs, resource demands, and the need for significant in-house AI expertise across data science, engineering, and product development. It makes sense primarily when no viable off-the-shelf AI solution meets an insurer's needs.

2. Buy and integrate a model

The most prevalent path for AI adoption currently is buying pre-built platforms and toolkits from vendors; an example CGI offers is CGI PulseAI. Insurers purchase these packaged AI solutions, then adapt and integrate them with their systems and data for specific use cases. This approach enables maintaining data residency and full auditability to meet regulatory requirements. It provides curated AI capabilities while allowing control over security, privacy, and model governance. 

3. Rent a model (SaaS-type services) 

The newest innovation for AI adoption is consuming models via third-party SaaS services and APIs. One example of this type of service is Azure Cognitive Services. This approach allows insurers to access cutting-edge AI capabilities while offloading infrastructure and hosting costs to providers. Companies pay subscription or usage-based fees rather than major upfront expenditures. However, rigorous data security and privacy controls are critical to prevent proprietary information from being incorporated into generic models that benefit other clients. Negotiating clear terms around data rights, monitoring, and monetization requires technical sophistication when engaging AI SaaS vendors. 

The bottom line

The AI revolution is redefining entire industries, including insurance. Companies that can successfully integrate AI into all levels of their business in compliance with legal and regulatory constraints and while adhering to ethical AI principles can expect to benefit from increases in employee productivity, increased customer satisfaction and improved profitability. No matter which path an insurer pursues, implementing AI is a journey that requires strategy alignment on opportunities and risks. Those who successfully navigate this transition can gain a powerful competitive edge through AI's transformative capabilities.

CGI is a trusted partner in AI and can help you in your journey. Connect with us today to learn more about our AI solutions and services and our responsible approach to artificial intelligence.