AI has become a prominent topic within the insurance sector, often surrounded by considerable hype. There is a prevailing notion that AI serves as a universal solution to every challenge, whether it be a problem, a need for improvement, or a desire for greater efficiency. However, this perception does not reflect reality. AI is not omnipotent; rather, it is a powerful tool when applied appropriately and in the right context.
Distinguishing types of AI tools
Within the industry, it is important to distinguish between general AI tools and those designed for specific purposes. General AI tools possess a broad, data-driven understanding of the world, yet may not be well suited to address particular problems. In contrast, specialised AI tools, which are trained for specific tasks, tend to outperform in their designated areas, even if they cannot answer every question. For example, translation tools such as DeepL are highly effective for language translation in international contracts, but are not designed to provide weather forecasts or address unrelated queries.
Key considerations for AI adoption
When considering the adoption of AI, three main factors should be evaluated:
- Tolerance for errors: AI systems are prone to errors and hallucinations. It is essential to assess whether errors can be tolerated in a given context and whether results can be tested or verified. AI will never deliver perfect accuracy, and this limitation must be acknowledged.
- Efficiency gains: While AI has the potential to enhance efficiency, this benefit may be offset if significant time is required for prompting, testing, or human oversight. In some cases, the anticipated efficiency gains may not materialise.
- Economic viability: The economic impact of AI must be carefully considered. Costs may shift from traditional areas such as underwriting to IT, and there may be ongoing expenses related to licensing, training data, and computing power. Achieving a comprehensive understanding of the cost implications can be challenging.
AI in underwriting
The use of AI in underwriting does not necessarily equate to automation. There exists a spectrum, with low-complexity problems involving homogeneous data being more amenable to automation. In contrast, highly complex or unique cases with heterogeneous data are less suitable for full automation, though AI can still play a valuable supporting role.
A notable strength of AI lies in its ability to process and structure unstructured data, such as risk assessment information. By standardising the presentation of data for underwriters, AI can improve efficiency and reduce potential biases. However, in complex cases, AI is best utilised as a support tool rather than as the sole decision-maker.
Applications in claims and fraud detection
AI also finds application in claims management, particularly in handling unstructured data from various sources. It can organise information consistently, highlight relevant contract language, and support fraud detection by identifying irregularities in data. While AI can point to potential fraud, it does not make definitive decisions, leaving final judgement to human experts.
The importance of data quality
A recurring theme in AI implementation is the critical importance of data quality. Effective AI models require homogeneous, high-quality data. Off-the-shelf AI products may not cater to the specific needs of an insurer’s portfolio, necessitating customisation and rigorous data management.
A balanced view of AI’s potential
AI should not be viewed as a miracle solution capable of resolving every issue. Instead, it serves as a tool that can smooth processes, enhance efficiency, and support decision-making. It cannot, however, eliminate all challenges or automate every aspect of the underwriting or claims process. The insurance industry benefits most from AI when it is deployed thoughtfully, with a clear understanding of its strengths, limitations, and the context in which it operates.
