Industry Voice: Managing AI responsibly in insurance
Artificial intelligence has game-changing potential for insurers, but those adopting it require a cohesive strategy in order to get the most out of it, argues Selim Cavanagh, Director of Insurance at Mind Foundry
As the adoption of artificial intelligence (AI) and machine learning (ML) models continues to proliferate across a wide range of industries, the emerging technology has already carved out a place of importance within the ever-evolving insurance market.
As data generation has surged and competition has intensified, insurers have begun to recognise AI’s game-changing potential as a way to analyse expansive volumes of data with unprecedented speed and accuracy. The advancement and adoption of these models seems unlikely to slow down as AI in the insurance market is projected to reach a market size of US$45 billion by 2031, according to Allied Market Research.
To make sure that the insurance industry reaps AI’s technological and economic benefits, insurers need to consider new solutions that will address the problems of today and the future.
Not a catch-all solution
ML was first introduced to the insurance market in the 2010s, which was a promising and exciting time. Newly discovered algorithms and models were produced ad hoc by in-house teams that helped insurers beat competitors and gain unprecedented market insights. At that time, this was a perfect solution. It was a chance to unlock
new intellectual property using skilled in-house expertise.
The benefits were felt across the entire business value chain, with senior leaders laying their hopes on AI models to improve insurance operations, enhance customer service and optimise risk management.
However, these same senior leaders faced new and unforeseen challenges. While there were many different causes, the overarching catalyst remained the same throughout the industry. Insurers were adopting AI without a cohesive strategy. This challenge remains pervasive in the industry to this day and managing these types of complex models often leads to hidden costs, such as increased operational expenses. Ad hoc models require specialist maintenance, consuming time and resources, all of which decrease the opportunity cost and relative gains.
Models that are created on an ad hoc basis also tend to degrade over time as the ever-changing market dynamic and customer preferences render them outdated. For some insurers, the sheer volume of AI models within one business also causes a problematic lack of transparency, making it difficult to identify inefficiencies and therefore creating a fragmented landscape of solutions.
Ad hoc models require specialist maintenance, consuming time and resources, which decreases the relative gains
Five solutions to consider
The key to overcoming these types of challenges is to look beyond a patchwork of ad hoc AI models and transition to a comprehensive and cohesive approach to AI deployment and management. The escalating demand for unique, valuable models calls for a strategic shift in thinking in order to ensure AI’s seamless integration into the insurance industry.
Deploying AI models across organisations, while also ensuring their continuous relevance and performance through ML governance, will set insurers apart. By adopting a bespoke AI model management solution, you can address all of the following aspects of AI adoption, but each will require its own strategic considerations:
• Integration and governance: Establish a centralised AI model governance framework to ensure adherence to regulatory requirements. This framework will help insurers oversee the seamless integration of compliant AI models into their existing business framework
• Scalability and flexibility: Adopting scalable AI that can increase the generation of data and cycle speed. This will ensure that insurers can meet market demands and rise to new opportunities, empowering teams to make data-driven and accurate decisions at every stage of the insurance value chain
• Continuous meta-learning: Implementing meta-learning strategies will enable AI models to continuously adapt and improve through self-learning. This means that models stay relevant and effective even with external market shocks such as inflation increases or regulatory changes. This improves the model’s accuracy and reduces the risk of model decay
• Collaboration and communication: Collaboration between all business stakeholders, from data science teams to senior leaders, will enable the seamless integration of AI into existing business models. Effective communication allows insights and knowledge to flow efficiently, ensuring models stay aligned with the business’s overarching goals
• Ethical considerations: Making sure that all who use the AI models do so in an ethical way is important. Insurers must be transparent with the data they use and the implementation of their AI, to ensure responsible and fair practices that don’t risk poor customer outcomes.
AI is only as good as the data it is trained on
Keeping AI models relevant and up-to-date is absolutely imperative for insurers. Research has shown that 91 per cent of models decay after the first year of being used as old data becomes obsolete. A recent report found that the cost of training the largest models, like large language models (LLMs), can range from $3 million to $12 million. The cost of training a model on a large dataset can be even higher, reaching up to $30 million – meaning insurers incur significant costs as models require retraining or rebuilding.
As new patterns emerge from new sets of data, models have the capacity to retrain and improve their performance automatically
The key to keeping models relevant and useful is the solution referred to as continuous meta-learning. As new patterns emerge from new sets of data, models have the capacity to retrain and improve their performance automatically, without having to incur expensive costs. This unlocks the ability for insurers to scale AI through their organisations, as well as converting significant capital expenditure costs into very low operating costs.
To integrate AI into a business in a way that derives real financial and societal value, insurers need to deploy an extensive portfolio of continuous meta-learning models that are customised to perform a variety of different functions across the organisation.
The first step of AI adoption is an exciting transition. But the next frontier will be
scaling this adoption throughout the whole insurance industry, unlocking a new stream of efficiencies, innovation and financial returns.