AI for Actuaries: Three Compounding Benefits

Last week I had the opportunity to address a room full of actuaries from one of South Africa's leading banks. I was speaking at their annual internal actuarial conference on the broad topic of "AI for actuaries". At this point, the headline question I hear most often is: what will AI do, and what will the actuary do?

It is a near impossible question to answer, for a few reasons.

Firstly, the speed of change in the world of agentic AI is rapid to the point of continual obsolescence in understanding. A statement grounded in reality and experience today can be outdated within months or even weeks.

Secondly, each company's adoption path is going to look different, and depending on the success of that adoption, the lived experience of AI will vary vastly. Even at a surface level today, the choice of agentic harness produces dramatic differences in the quality of an agent's outputs.

Lastly, the question itself assumes a fixed division of labour: a world where tasks get neatly carved up into an AI pile and an actuary pile. That is not how this plays out. The actuarial role is not being divided, it is being redefined, and any answer that draws a static line between the two is describing a boundary that will have moved before the ink is dry.

So rather than attempt an answer, I settled on three compounding benefits that AI agents will provide, which we as a business are aiming for, and which can improve the lived experience of any actuary. For the sake of clarity, I focus below on the valuations actuary.

Gift 1 : More time

Agentic AI systems are already proving themselves in the area of process automation, but this automation is better described as flexible workflow orchestration. Let's break that down by comparison to traditional automation techniques.

Traditional automation was script based. It depended on highly predictable inputs, scenarios and outputs, and was generally rigid and brittle. Agentic systems, by contrast, can adapt to subtle changes in automation requests with high reliability. They plan their execution steps, execute, evaluate outcomes, adapt if needed, and retry, or even request additional guidance from the user.

These processes are quick to set up and iterate on when the agent harness has the necessary access and tools for the job at hand, and they are instructed in plain English as pseudo-code through skills and prompt files. The end result is that the valuations actuary can hand over the time consuming, mundane and low value work involved in going from raw data to results (toolstack execution) and spend more time on high value analysis and interpretation.

Gift 2 : Data fluency

Actuaries and data go hand in hand, but even the most skilful SQL guru can't compete with the sheer speed that LLMs, equipped with the right tools, bring to data exploration, summarisation and analysis. Agentic AI is going to give all actuaries high data fluency through more consistent semantic interpretation of data. It will significantly reduce the cost of questioning data, and enable the actuary to pursue many more investigation paths than ever before.

Gift 3 : Autonomy

This last one is the stretch goal. We are increasingly privileged to see the inside operations of more businesses each week, and while every business is different, they largely suffer from similar challenges. Key person dependency is one of them.

Key person dependency is primarily a problem of siloed knowledge. The pain is that completing an end-to-end valuations or investigations process depends on multiple inter-person handoffs, because the required process, system or product knowledge is not common or freely available. The goal of agentic AI systems is to capture, curate and grade knowledge progressively in a knowledge store that any actuary in the system can draw on, and importantly, that the agents themselves can draw on too.

This means all actuaries in the system can operate with increasing independence from each other over time. No more waiting on John, who is busy with something else, or Jane, who is on leave, for the critical information needed to perform even basic tasks.

The compounding effect

Any one of these benefits alone, whether time, data fluency or autonomy, would provide a significant uplift in an actuary's ability to deliver better insights and value for the business. The compounded impact of all three is game changing. As we sit on the cusp of this evolution in the actuarial operating model, I am increasingly grateful for the role we as a business get to play in bringing the actuarial profession, and many businesses, into that future.