Gary Bundock and Graham Spark (MD of MML Solutions, which provides the platform for SelectX’s RiskApps underwriting manual) offer some considered views on a controversial subject.
The rise of artificial intelligence (AI) has sparked a wave of excitement and anticipation, with many businesses eager to harness its potential to transform their operations. However, amidst the hype and enthusiasm, it is crucial to approach AI with a level head and avoid getting swept away by unrealistic promises and exaggerated claims.
One of the primary challenges of AI is the temptation to pursue its implementation solely out of fear of being left behind by competitors. This fear-driven approach often leads to the pursuit of solutions for problems that don’t exist, resulting in a waste of resources and a failure to address genuine business needs.
While AI undoubtedly holds immense potential, building a robust and effective AI model tailored to a specific business problem is a complex and time-consuming endeavour. It is not an overnight transformation; it requires careful planning, strategic implementation, and continuous refinement.
Organisations should exercise caution when turning to internal IT departments or agencies to provide AI solutions. These entities are unlikely to possess the necessary expertise or experience to deliver effective and reliable AI implementations at this point in time. The first wave of AI solutions are likely to be riddled with bugs, or hallucinations and come at a premium, similar to early data warehouse, or ‘big data’ projects that often required costly rewrites. Just maybe, with AI, something can be done with big data.
Many AI enthusiasts tout the ability to achieve remarkable results within a short timeframe, typically 12 to 18 months. While some projects may indeed meet this ambitious goal, many more will fall short, resulting in expensive failures and setbacks.
In the realm of underwriting, AI and machine learning (ML) will indeed offer significant potential to assist and inform underwriting, including automating more of the point-of-sale (POS) processes, and understanding claims experience to inform POS systems for example. However, it is essential to approach AI adoption with a measured and pragmatic approach, avoiding the pitfalls of hype and inflated expectations.
One should have realistic expectations and carefully define the problem before rushing into a solution (as with implementing any technology). For example, having the expectation that the ingestion of an underwriting manual into an AI model will produce a system capable of making complex underwriting decisions seems unlikely to work in exactly the same way that copying a rating table from a manual into a rules engine will be futile.
It has been said that AI models have difficulty with decision making. On the other hand, AI has been demonstrated to be helpful for sifting through large amounts of data, separating the unimportant from the significant, for example in the review of electronic health records or financial information. Here the machine does not suffer from lack concentration, meaning (in theory) that nothing gets missed and the human underwriter becomes more efficient. Using the capabilities of generative text to improve question wordings in POS systems should also have practical application.
However, AI can only bring benefits if the data it is using to learn, draw conclusions and make decisions is good enough. Within its considerable power is the ability to learn in a flawed way and to steer a business off in the wrong direction. Any AI-driven shift in underwriting policy, for instance, needs to be identified, reviewed and judged to be acceptable.
By carefully evaluating the true needs and challenges of the organisation, businesses can make informed decisions about AI implementation, ensuring that their investments yield tangible benefits and contribute to long-term success.
Like it or not, AI is poised to impact everyone and every business sooner than one might anticipate. Embracing AI is not a choice; it’s going to be a necessity. Back in the 1990s, the introduction of underwriting rules engines was greeted with scepticism and, from some underwriters, suspicion, resentment – and even trepidation. As Sam Cooke sang, ‘A Change is Gonna Come’, and underwriters need to be ready and prepared.
By adopting a pragmatic approach to AI implementation, organisations can avoid the pitfalls of hype and unrealistic expectations, ensuring that their investments yield tangible benefits and contribute to long-term business benefits. Starting small, learning from initial projects, and building a team of experts will pave the way for effective and impactful AI adoption.
This article first appeared in the January 2024 edition of Hot Notes, published by Hank George Inc.
