From Spreadsheet to AI Agents: How Actuaries Can Lead the Productivity Revolution

What if the most stressful hour of your Monday morning could become the most productive?

In this article  — based on our recent IAA webinar on AI Agents and Actuarial Enablement — we'll walk through how a digital team of AI agents helped complete a full analysis, model revision and report generation in under an hour. Whether you're new to AI or already experimenting with automation, you'll see what's possible and where to start.

It's 9am Monday. Coffee hasn't kicked in yet when your inbox pings: “Customer complaints are up. We need modelling scenarios, analysis, and a full report by 10am.” Your pricing expert is out sick, your analyst won't be in till 10, and you're staring at hundreds of pages of data and feedback. Sound familiar? It usually means a scramble — unless you have a digital team of AI agents at your side.

The AI moment we're in

AI capabilities are advancing remarkably quickly.

trackingai.org measures weekly model performance across 27 models, and we're seeing top-tier models like Claude 3.7 Sonnet, Gemini 2.5 Pro, and OpenAI-o3 regularly scoring above 110 on IQ tests. Six months ago, only one model (OpenAI-o1) broke 100. Model Evaluation and Threat Research plots how models perform with longer tasks successfully — current models can handle hour-long tasks, and this capacity appears to be doubling every seven months.

Claims about rapid advancement always need some scrutiny but the practical applications we're seeing in actuarial work are becoming harder to ignore. During our webinar, we suggested that actuaries need to move beyond spreadsheets toward orchestrated digital AI assistants. To stay relevant and trusted, our profession should lead rather than lag in AI adoption.

What makes AI agents different from ChatGPT?

AI agents are systems that operate with a degree of autonomy to achieve goals. AI agents operate on what's called a Perceive – Reason – Act loop. They don't just generate text; they can remember goals, evaluate their own results and sequence actions across multiple tools. They're not autonomous employees (thankfully) but they aren't just chatbots either. With proper oversight, they become powerful extensions of the actuarial team.

Important, agents can take on many different roles. For instance, the AI team we demonstrated during our webinar included:

• The Business Analyst: Summarises technical reports using document parsers and NLP tools.
• The Data Analyst: Processes unstructured customer feedback, classifying sentiment and producing analysis tables.
• The Modelling Expert: Updates assumption and premium rate tables, executes modelling code and displays model output.
• The Report Architect: Provides detailed report structures with clear instructions on section content.
• The Content Generator: Populates each section based on the instructions from the Report Architect.
• The Review Agent: Fact-checks, edits tone, and flags incomplete arguments for the Content Generator to act upon and redraft.

Monday mornings reimagined

During the webinar, we simulated a time-constrained actuarial task using this agent team. Here's roughly how it went:

Understanding the background: The Business Analyst tackled a dense 2023 pricing report. Within seconds, key themes were extracted and summarised on screen, drawing attention to prior concerns that may have now materialised.

Understanding the present: The Data Analyst processed 50 pieces of customer feedback (roughly 25-30 pages), extracting structured tags for product, sentiment and pricing methods. This quickly turned large volumes of unstructured complaints data into something we could actually analyse.

Modelling the impact: The Modelling Expert promptly displayed premium rates and adjusted them based on our natural language request. The profitability model was rerun and results were displayed on screen, without us even needing to press a button. Complex premium rate tables are no barrier.  The system can allow different agents to specialise in different table sets, which could be useful for complex modelling applications where hundreds of table inputs may exist.

Reporting the solution: The Report Architect generated a custom report structure tailored to our specific brief. The Content Generator diligently filled in each section and subsection. The Review Agent provided feedback on style and accuracy. The section-by-section feedback loop improved clarity and completeness. The entire report generation cycle completed in under 10 minutes through the streamlined collaboration of three AI agents —though we did have to proofread the generated report and perform some manual cleanup afterwards.

How this actually works

The agent cycle looks like this:

• Perceive: Interpret user instructions ("Summarise this report")
• Reason: Decide which tool or process to use
• Act: Execute using prebuilt Python tools or document parsers
• Perceive: Review outputs (assess the quality of the summary, model output accuracy)
• Repeat: Refine or pass output to the next agent

Each agent stays focused on their role and can collaborate across functions. The system shows promise, though human oversight remains essential.

Actuaries remain in the driver’s seat

As we emphasised in the webinar, AI agents aren't autonomous professionals. They can be misled, hallucinate results or misuse tools.

Without proper governance, they create more risk than value. That's where actuaries come in — ensuring professional judgment, managing AI limitations and translating data into strategic action.

So what does this mean for the future role of actuaries? Less spreadsheet work and more orchestration of digital tools.

Getting started - without breaking anything

Experimenting with AI agents is relatively straightforward, although best results requires engaging more deeply with current workflow tools. Our tips on the best way to get started:

• Try a pilot task: Build a simple agent like a Business Analyst that can summarise a document.
• Start with CrewAI: Role-based design that's beginner-friendly.
• Advance to AutoGen or LangChain: Adds human-in-the-loop workflows and more sophisticated execution.
• Learn Python: Especially for defining tools and debugging when things go wrong.
• Apply governance: Use frameworks like Model Context Protocol (MCP) or Agent-to-Agent (A2A) to manage complexity — though honestly, these are still evolving.

The bottom line

The question isn't whether AI will transform actuarial work — it's already happening. The question is who's going to lead that transformation.

With tools becoming more accessible and frameworks maturing, actuaries can and should take the lead.

The agent-enabled future isn't science fiction anymore. It might just be your Monday morning, reimagined.

This article is based on the IAA Data Analytics Webinar Series presentation by Jared Spowart and Zeming Yu on "AI Agents and Actuarial Enablement." For replay and slides, visit the IAA website .