Actuarial work has always had three layers.
GenAI fits into exactly one of them.
Layer 1: The Language Layer
This is where you communicate.
Draft emails.
Write reports.
Explain technical concepts.
Summarize findings.
Respond to questions.
GenAI lives here.
It's exceptional at taking technical content and expressing it clearly.
It drafts the first version of almost anything you need to write.
It translates between technical and non-technical language.
It handles the repetitive communication tasks that consume hours of your week.
Layer 2: The Model Layer
This is where you analyze.
Build GLMs.
Run chain ladder.
Apply credibility methods.
Calculate present values.
Fit survival curves.
GenAI doesn't live here.
It can't build models.
It can't run methods.
It can't perform calculations.
It can describe what happens in this layer, but it can't do the work.
Your pricing models, reserve analyses, and valuation calculations remain unchanged.
GenAI doesn't replace them.
It doesn't augment them.
It sits above them, in a completely different layer.
Layer 3: The Judgment Layer
This is where you decide.
Choose assumptions.
Interpret edge cases.
Balance competing objectives.
Accept or reject model output.
Recommend action.
GenAI definitely doesn't live here.
Judgment requires:
- Understanding context
- Weighing tradeoffs
- Accepting accountability
GenAI has none of these.
It can articulate different perspectives on a decision.
It cannot make the decision.
How the Layers Interact
Traditional workflow:
- (Judgment) Decide what analysis to run
- (Model) Execute the analysis
- (Language) Communicate the results
- (Judgment) Decide what action to take
Workflow with GenAI:
- (Judgment) Decide what analysis to run
- (Model) Execute the analysis
- (Language + GenAI) Draft the communication
- (Language) Verify and refine the draft
- (Judgment) Decide what action to take
GenAI inserted itself into step 3.
Everything else remained the same.
Why This Mental Model Matters
When you understand GenAI as purely a language layer tool, you stop:
- Asking it to do calculations
- Expecting it to make recommendations
- Treating its output as analysis
You start:
- Using it to draft routine communication
- Leveraging it to explain technical work
- Treating it as a writing assistant, not a colleague
Common Mistakes, Explained
Mistake: "GenAI can help with my pricing model."
Reality: GenAI can draft the memo explaining your pricing model's results. The model itself is unchanged.
Mistake: "I'll use GenAI to validate my reserve assumptions."
Reality: GenAI can draft language explaining why your assumptions are reasonable. It can't actually validate them.
Mistake: "GenAI gave me three scenarios to consider for this decision."
Reality: GenAI gave you language describing three scenarios. You still need judgment to evaluate them.
The Litmus Test
Before using GenAI for a task, ask:
"Is this fundamentally a communication task or an analysis task?"
Communication task: GenAI can help.
Analysis task: GenAI cannot help.
Most actuaries think 70% of their work is analysis.
In reality, 70% of their time goes to communication around analysis.
That's where GenAI creates value.
What This Changes
You still need to:
- Understand actuarial methods
- Know how to build models
- Apply professional judgment
- Verify all technical work
- Accept accountability for recommendations
You no longer need to:
- Spend 30 minutes drafting routine emails
- Manually format executive summaries
- Rewrite the same explanation for different audiences
- Struggle with phrasing for sensitive communications
The intellectual work remained.
The administrative overhead around that work decreased.
One Clear Takeaway
GenAI is not an actuarial tool.
It's a communication tool that happens to work well with actuarial content.
Your models still live below it.
Your judgment still sits above it.
GenAI fills the middle layer—the one where you translate technical work into human language.
Use it there.
Ignore it everywhere else.
And you'll capture the value without the risk.
The three layers:
- Language (where GenAI helps)
- Models (where your technical skills matter)
- Judgment (where your experience decides)
Keep them separate in your mind.
Use the right tool for each layer.
That's the entire mental model.
That's all you need to know.