Psicle: The future of business-critical modelling
Adrian Ericsson, group managing director of Dynamo Analytics, discusses the impact of generative artificial intelligence on financial modelling, and issues to consider when integrating AI into business-critical processes
What does generative AI mean for the financial modelling landscape?
We all know the world of business-critical modelling is undergoing a rapid transformation, driven by advancements in technology, such as cloud computing and quantum hardware, and advancements in algorithms, such as machine learning and generative AI.
As businesses look for more speed, insight and accuracy in modelling, we are re-imagining traditional approaches to designing and implementing algorithms. At the heart of this transformation is how we are starting to wrestle with and use machine learning and generative AI’s potential to automate our tasks and judgements, generate insights, and build complex algorithmic processes. These technologies are not just promising to enhance efficiency; they are reshaping the very nature of how we anticipate we will interact with data and models, and ultimately make decisions.
What are the considerations for integrating AI into business-critical processes?
The integration of AI into business-critical calculations, particularly in regulated industries, introduces significant risks. The complexity of these algorithms can make them difficult to understand which raises concerns about transparency, accountability, and trust.
Executives cannot afford to rely on systems they do not fully understand or trust, especially when the stakes are high and regulators are vigilant. In this context, modelling principles of certainty, repeatability, integrity, and control are just as vital as speed and insight. While AI can offer powerful capabilities, we must be sure to balance it with robust governance and human oversight, so that outcomes are reliable and aligned with organisational requirements.
How can insurers balance AI’s risks and opportunities?
Processes undertaken by humans and machines should not necessarily remain separate, nor should one dominate. We see the most effective solutions arising from thoughtful integration, where both people and machines complement each other. By combining their strengths, organisations create modelling processes that are not only faster and more insightful, but also resilient and trustworthy.
To achieve the right balance, we must rethink how professionals and AI collaborate. Rather than focusing solely on automation or judgement, it is really important to consider the end-to-end algorithmic process and determine the optimal combination of human and machine involvement at each stage of the process.
We believe that each part of an end-to-end business-critical modelling chain (a data transformation, a chain ladder pattern selection, an allocation, a volatility parameter estimation, etc.) may require a different approach. For example, some parts will benefit from rapid automation with low human intervention and collaboration, while others may need to sacrifice speed in favour of gathering consensus on a few key parameters.
Tell us about cyborg patterns and why they matter?
We refer to these intentional and crafted combinations of human and machine roles as “cyborg patterns”, and they come in many flavours; from fully autonomous agentic AI structures, all the way to luddite humans typing numbers into a sheet.
In between these extremes are patterns where algorithms can provide value; information for human expert judgements; default suggestions which humans can override; structures for collaboration on single high-impact parameters; analysis of implications; and many more. Each of these patterns has their own personality; some are fast and imprecise; while others are slower and accurate; some are transparent while others are opaque; and some are dependent on human judgement; while others don’t need to be available.
Equally, the strategic requirements of business-critical processes must be considered, and careful thought needs to be given to the relative importance of factors such as speed, transparency and complexity.
How do you achieve an optimised AI-driven algorithmic estate?
An effective and optimal algorithmic estate is one where the business has thought about the strategic requirements of the business-critical process – speed, transparency, complexity, reliability, and others – and has crafted an array of intentional “cyborg patterns” for each part of the process in order to fulfil these requirements.
By embracing the strengths and characteristics of both humans and machines at a granular level through the algorithmic estate, your organisation can build resilient systems that are trusted as part of the decision making processes, and are more likely to stand the test of time. The outcome is an end-to-end process that is as fast as it can be, as transparent as it should be, and as robust as it must be.
The cyborg is not just a vision; it is a practical framework for further embedding algorithmic modelling as the heart of organisation in a variable and rapidly evolving world.
__________________
Psicle - transforming financial and actuarial processes across non-life, life insurance, banking and more.
