 |
|
 |  |  |  |
 MCP connects AI agents to tools and data, while agent-to-agent architectures let specialist AI systems collaborate, helping make actuarial technology more automated, scalable and accessible. Just as the insurance industry has begun to understand Generative AI and Agentic AI, another set of technologies has emerged that promises to make AI systems much more useful in practice. These are multi-agent systems or agent-to-agent communication, and Model Context Protocol (MCP). |
By Mark Brown, Global Proposition Lead, Life Financial Modeling, Insurance Consulting and Technology, WTW But what do these concepts mean? And what implications do they have for life insurers and actuarial modelling platforms? This article explores the next stage in the evolution of AI systems. What is MCP?
Large Language Models (LLMs) are very effective at generating text and code. However, they have traditionally operated in isolation. They can answer questions, but they cannot naturally access company systems, databases, modelling platforms or business processes.
The Model Context Protocol (MCP) addresses this challenge. MCP provides a standard way for AI systems to connect to external tools, applications and data sources. Rather than building custom integrations for each AI model and each application, organisations can expose capabilities through MCP connectors. You can think of MCP as similar to an API (programming interface) for AI to use. An AI agent can connect to many different systems through MCP, just like an app can link to other systems via their APIs. Examples could include:
Document repositories
Actuarial modelling platforms
Reporting systems
Development tools
Knowledge bases
Workflow applications
Instead of asking an AI model to explain how to run a process, users can ask it to execute the process directly using approved tools. What is Agent-to-Agent AI?
Agentic AI systems can already perform tasks independently. However, a single agent often becomes complicated when responsible for many different capabilities.
A more scalable approach is Agent-to-Agent collaboration. Under this model, multiple specialist agents work together. For example:
A Planning Agent defines objectives
A Data Agent gathers information
A Modelling Agent performs calculations
A Validation Agent checks results
A Reporting Agent prepares outputs
Rather than relying on one enormously complex system, each agent focuses on a specific responsibility. This mirrors how human teams operate today. Why does MCP matter for Agentic AI?
Agentic AI becomes far more powerful when agents can interact not only with each other but also with enterprise systems. Without MCP, Agents can reason, can write content and have limited access to operational systems.
With MCP, Agents can retrieve data, can update records, can trigger workflows, can access governed business applications and can collaborate across technology platforms. The combination enables AI systems to move from "advising" to "doing". What could this mean for life insurers?
Many insurance processes involve gathering information from multiple systems and coordinating work across many different specialists.E
Examples include:
Financial Reporting
Where agents could gather model outputs, retrieve assumptions, validate data quality, generate management commentary and prepare reporting packs.
Product Development
Multiple agents could review product specifications, generate model requirements, produce test cases, document results and prepare governance evidence.
Assumption Management
Agent teams could monitor experience studies, identify emerging trends, propose assumption updates, generate approval documentation, track implementation progress.
In each case, humans remain responsible for decisions and governance, while AI reduces manual effort. But isn’t that Business Process Excellence?
Both automation (BPE) and agentic AI emulate humans and can alleviate their workloads. Both are helpful to actuaries and in financial reporting. However, they approach the problem from opposite ends, and thus complement rather than compete.
BPE solutions, such as WTW’s Unify, replay processes and workflows, including their interactions with other toolsets (via APIs) and humans. They’re robust and reliable, provide progress reporting, governance and clear auditability. If something goes wrong, they fail noisily. Agentic solutions, on the other hand, solve the problem in a creative and dynamic way, again interacting with other toolsets (via MCP) and humans. They’re flexible, can cope with the unexpected to an extent; but if something goes wrong, they fail silently and you may never know. Cost is another important consideration when evaluating Agentic AI against automation. Traditional process automation derives much of its efficiency from being deterministic, with known execution paths and predictable resource consumption. By contrast, agentic AI introduces dynamic decision-making, variable execution paths, and thus higher and less predictable incurred costs. What does this mean for actuarial reporting?
Historically, actuarial platforms have focused on model definition and calculations. For many companies, this view has already expanded to cover the end-to-end reporting and management information cycles.
Agentic toolsets and application integrations can further reduce the low-value human workloads in both the modelling and reporting arena; including understanding requirements, maintaining documentation, validating changes, investigating issues, preparing reports and communicating the governance. Whether manual or automated, MCP-enabled agent ecosystems allow AI systems to support each stage of this process. What challenges remain?
Despite the excitement, several challenges remain. Security and governance become increasingly important when AI systems gain access to enterprise tools. Equally in their use, organisations must ensure appropriate permissions, audit trails, data protection considerations, human oversight and human controlled decision-making.
Users also need to be aware of costs. We’ve all heard anecdotes of AI spend costing more than the humans it replaced. You should consider: Is AI appropriate for the task?
For example, high-volume repeatable processes are often better automated, reserving AI use for judgement-intensive or highly variable activities.
Are you using the right AI models?
Some models can consider more options and in more depth than is needed, driving up their costs further than the value added. Many models also prompt for further use beyond the initial ask – the equivalent to the burger-shop asking “would you like fries with that?”
Is now the time to invest?
While MCP and Agent-to-Agent AI are still emerging technologies; they’re mature enough and clear enough in their direction that many organisations are experimenting with early proof of concepts. What’s clear from the early projects, though, is that two areas are key for success:
Operational Foundations
Having robust and reliable tools for the AI agents to interact with. For agents producing management information; this means building them around robust and automated reporting processes. For product design work, direct model access may suffice.
Governance
Where agents are able to contribute to workflows, either in modelling or reporting, you need a robust governance process that captures their recommendations, the review and approvals of those, and to be able to identify everything impacted by those agents.
The question is no longer whether AI will participate in actuarial processes. The question is how quickly organisations can establish the governance, tooling and operating models needed to do so safely and effectively.
|
|
|
|
| Pricing Actuary, Marine Cargo/Property | ||
| London - £130,000 to £150,000 Per Annum | ||
| Pricing Actuary – Cyber | ||
| London - £120,000 Per Annum | ||
| Reinsurance Pricing Actuary, Analytics | ||
| London - £130,000 to £180,000 Per Annum | ||
| Actuary - Financial Planning & Analysis | ||
| London/Hybrid - Negotiable | ||
| Reinsurance Pricing Actuary | ||
| London - £140,000 Per Annum | ||
| Head of Capital | ||
| London - £170,000 Per Annum | ||
| ART Pricing | ||
| London - £100,000 Per Annum | ||
| Pricing Transformation Actuary | ||
| London - £130,000 Per Annum | ||
| Pricing Actuary | ||
| London - £80,000 to £120,000 Per Annum | ||
| Pensions on Divorce Startup - Flexibl... | ||
| Remote - Negotiable | ||
| SVP, Head of Reserve Forecast Analytics | ||
| Bermuda - £200,000 Per Annum | ||
| START-UP, Lead Reinsurance Actuary | ||
| London - Negotiable | ||
| Senior Actuary | ||
| London - Negotiable | ||
| Reserving Manager | ||
| London - £130,000 Per Annum | ||
| Senior Reserving Consultant | ||
| London - £100,000 Per Annum | ||
| Head of Capital | ||
| London - £180,000 Per Annum | ||
| Head of Portfolio Optimisation | ||
| London - Negotiable | ||
| Pricing Lead/Manager | ||
| London - £130,000 Per Annum | ||
| Actuary | ||
| London/Hybrid - Negotiable | ||
| Capital Actuary | ||
| London - £110,000 Per Annum | ||
Be the first to contribute to our definitive actuarial reference forum. Built by actuaries for actuaries.
Copyright © 2026 Actuarial Post. Created by Zero-One Design Ltd
