AI belongs in
actuarial practice.

Adopting AI feels risky. It feels unactuarial. Actuaries don't recoil from risk, though; they build frameworks to manage it. Adopting AI is actuarial.

Let's talk

Healthcare Actuarial + AI

“But how do I trust it?”

This is the question I hear from Chief Actuaries.

The reality: you don't have to.

AI doesn't require you to trust it any more than the analyst running your monthly IBNR.

“AI can hallucinate.”

Hallucination is just a wrong answer. Your analyst can do this, too.

The difference is you know to manage your analyst. You establish a review process, validate outputs, and document methodology. Trust never enters the equation.

AI is no different. You don't have to trust it.

You just have to manage it.

Actuarial discipline applied to AI implementation.

When actuaries implement a new model, they have a process. Scope is defined, methodology is documented, and outputs are back-tested and validated. The phases of AI implementation go by different names, but the process is the same.

.01

Strategy

Your company has an AI strategy. Your actuarial team needs its own.

Scope boundaries: which workflows are deterministic enough for AI, which require human judgment, and where the line is drawn
Professional liability mapping: how AI-assisted work meets SOA standards, ASOP guidance, and regulatory requirements for signed-off outputs
Audit readiness: documentation trails that prove method selection, assumption validation, and peer review

.02

Diagnostic

Not every process needs AI. The diagnostic finds the ones that do.

Process maps: inputs, tools, steps, outputs, time, and judgment points documented for every recurring workflow
Execution vs judgment ranking: AI candidates sorted by ROI, starting with deterministic tasks before analytical ones
Institutional knowledge capture: data lineage, business rules, and unwritten conventions documented before AI touches any process

.03

Build

Start with one process. Prove it works. Expand from there.

Single-workflow: data pipeline, calculation engine, and review interface focused on one high-ROI process
Historical validation: output tested against past periods to prove accuracy before live deployment
Parallel run: automated and manual work side by side for multiple cycles so actuaries verify results with their own eyes

Built for how actuaries actually work.

Tristan Campbell, FSA

An actuary building in AI.

When I started ProActuarial Insights in 2024, AI was a curiosity.

I was doing the same work I'd been doing for years: combining line-level medical claims analysis with financial storytelling to help healthcare organizations manage and contain claims cost. ChatGPT had entered the picture, but I hadn't gone beyond the basics. I hadn't used it for actuarial work, and I wasn't convinced I ever would.

That changed in 2025. Three weeks into a redesign of my company website using Wix, I had spent more time fighting the tool than building the site. I knew enough then to know I could have used AI from the start, so I scrapped what I had and started fresh. Three days later, the redesign was done and the site was published.

Three days became three months. I moved from editing website copy to building production-ready claims dashboards in Claude Code, running networks of AI agents, and standing up my own private server to run local models. The website build showed me what AI was capable of; the exploration that followed convinced me it belonged in actuarial practice.

Today I'm focused on helping actuaries implement AI the right way. That requires domain expertise. Your company's AI strategy is the starting point, but it doesn't specify how to keep AI separate from actuarial judgment, or where that judgment even lies. It doesn't define the liability unique to actuaries, and how to ensure professional standards are met. For that, you need more than technical expertise. You need an actuary, one who's already building in AI.

Excel replaced paper spreadsheets.
PCs replaced the mainframe.
Neither replaced actuaries.
AI won't either.