This piece is by Karli Kalpala, Head of Strategic Transformation at Digital Workforce.
The insurance industry is at an inflection point. Disruptive digital entrants, a changing regulatory landscape, and constantly evolving customer expectations are combining to upend sector norms. Attempts to increase revenues through premiums are limited by the ongoing cost-of-living crisis, particularly regarding non-essential products. As such, improving customer experience and operational excellence, and in doing so enhancing value, has become the priority for insurance digitisation initiatives.
To support this, insurers are turning to innovations such as artificial intelligence (AI). For example, 99% of insurers are either already investing or making plans to invest in Generative AI (GenAI).
There are certainly opportunities to streamline operations, improve decision-making, and enhance customer service. In claims alone, GenAI could potentially reduce payouts by between three and four percent, and drive a 20-30% reduction in loss-adjustment expenses.
The Limitations of Out-of-the-Box AI Models
Most insurers are starting with off-the-shelf AI models. These solutions can offer incremental benefits; what they do not deliver is the fundamental transformation of workflows or operational structures.
Off-the-shelf LLM models are marketed as quick, plug-and-play solutions. While effective for simple, repetitive tasks like answering internal queries or summarising meeting transcripts, these models could struggle with the nuanced, complex work of underwriting, claims processing, and fraud detection.
A key limitation is data quality. Off-the-shelf solutions are often trained on generic, broad datasets that lack industry-specific context. As a result, they may provide generalised or outdated responses when applied to the actual domain specific work of an insurance company. For example, solutions based on off-the-shelf LLM might misinterpret policy terms or provide outdated claims information, leading to customer dissatisfaction and regulatory risks if used operationally as such.
Additionally, these models are static and unable to adapt in real time to the unique needs of the insurance industry.
Improving Accuracy with RAG and Fine-Tuning
To address these shortcomings, insurers can improve AI performance through techniques like Retrieval Augmented Generation (RAG) or fine-tuning.
With RAG, companies enforce the AI models to build their responses to prompts by only referring to specific knowledge to provide context. It requires data scientists and engineers to build dedicated vector databases, review output quality and revise prompts accordingly. Insurers might use RAG to ensure the model gives claims adjusters access to up-to-date policy information.
Fine-tuning uses private data to teach the actual Large Language Models themselves on how to respond to prompts, increasing the likelihood of the model becoming tailored to specific use cases. However, it is resource-intensive, loses value over time if it is not retrained regularly, and can be prone to hallucinations.
Although both approaches enhance accuracy, they still position AI as an assistant to human decision-making. The AI provides recommendations, but human employees remain in control of the planning, decisions, and actions. These methods focus on helping humans extract useful data from the AI solution, rather than empowering AI to take independent action.
AI Assistants vs. Autonomous Agents: Reimagining the Insurance Workflows
The real transformative opportunity lies in moving beyond AI as an assistant and embracing autonomous AI agents. These solutions take full responsibility for tasks, reasoning based on natural language instructions and your enterprise data, making decisions, and executing actions within enterprise systems, requiring at most limited human intervention.
AI agents work by following a runbook, Natural Language Instructions—a document typically created by the business subject matter expert (SME) that defines the task’s objectives, purpose, and high-level workflow steps. They are instructed to stay strictly within the activities defined in the instructions, ensuring focused and precise execution.
So, agents differ from approaches such as RAG in their capability for in-context learning during inference time. Rather than retraining a model on large datasets and risking obsolescence, AI agents adapt in real-time by integrating insurer-specific data and context. This allows them to continuously improve and respond to new challenges without the need for regular retraining.
The result is an AI agent can plan its own worksteps, collaborate with humans to get guidance on next best action and most importantly they take action against your enterprise systems giving them autonomy to take ownership of the entire piece of work they are designed for. And this is a paradigm shift as it allows automation of knowledge work as we know it.
AI Agents in Insurance
Imagine a customer has an accident and needs to file a car insurance claim. Traditionally, this would mean reams of paperwork and multiple handoffs, dragging the process out and amplifying the stress. Now, picture a coordinated team of AI agents working in unison: one conducts the initial triage to confirm coverage and decide if straight-through processing is possible, another summarises every claim document to give adjusters a clear snapshot, while a third updates policy reserves after settlement to release working capital. With these agents collaborating autonomously and only escalating when anomalies arise, what once took weeks can now be resolved in hours.
Fraud detection is another area where agents could transform not only processes but full operating models. Many insurers want to implement a Straight-Through Process (STP) for simple claims, but struggle to manage the business risks related to it. Now AI agents can be deployed as an infinitely scalable resource to oversee the process and monitor for fraud even in STP set-ups. These agents can automatically flag suspicious claims, such as those where the same individuals appears as claimants or witnesses of each other, by analysing patterns and cross-referencing them with the companies full claims history. By automating the detection, AI agents not only reduce human involvement but allow the implementation of the STP without increasing financial risk.
In both instances, agents work concurrently, in contrast to human teams, which only focus on their specific tasks and often must wait until another team has completed their own.
This shift from manual work to full-task automation empowers insurers to reimagine how they approach their operations. AI agents not only automate repetitive tasks but also drive innovation, create new roles, and streamline workflows, freeing up human resources for higher-value work.
Overcoming the Obstacles to Agent-Driven, Enterprise-Wide Transformation
Insurers need to tackle several challenges to fully realise the potential of AI agents.
1. Regulatory ambiguity: There is still a significant amount of concern regarding the regulation of AI. How much is relevant depends on the type of cover an insurer provides; the EU AI Act, for instance, defines life and health insurance using AI as high-risk, but not other areas. Insurance providers will need to keep up to date with regulatory changes, but this should not restrict their investments.
Much of the relevant regulation is focused on the impact on the customer; if insurers ensure their AI deployments are customer-centric, then they may well be in a better position should the rules change.
2. Culture: AI may have huge potential applications in insurance, but the industry is not known for its acceptance of innovation. Adjusting company cultures to embrace the value of AI is critical to deployment success.
3. Scale: One agent is a pilot, but for AI to have a truly transformative effect insurers need to scale their agent deployments. That requires an enterprise-wide AI infrastructure that supports hundreds of AI Agents in autonomous operations. This infrastructure should ensure that AI agents can integrate seamlessly with existing systems, operate autonomously, and empower business experts to configure AI agents according to specific business objectives.
By building themselves or buying from existing vendors such an infrastructure, insurers can scale AI agents across the organisation, leading to a transformation that revolutionises fields such as underwriting, claims management, and fraud detection.
Embracing AI Agents for True Transformation
While fine-tuning and RAG improve pre-built models, they still leave insurers reliant on human decision-makers. True transformation comes when AI no longer merely assists teams but takes full responsibility for tasks. By adopting autonomous AI agents, insurers will be better placed to navigate continuous industry disruption and position themselves for future-proofed success.
Be the first to comment