Generic LLMs Are Not Enough for Insurance Work. Agencies Need Insurance Skills.

Colby Tunick -

Publicly available large language models are changing how knowledge work gets done.

Insurance teams are already experimenting with tools like Claude, ChatGPT, Gemini, and other AI assistants to summarize documents, draft emails, compare files, and speed up everyday work.

Generic LLMs are not enough for insurance to work out of the box.

That experimentation is good. But insurance agencies need to be clear-eyed about one thing:

Generic LLMs are not enough for insurance to work out of the box.

You would not ask an out-of-the-box LLM to review an X-ray and tell you whether there is a tumor without medical training, clinical context, and physician oversight. Insurance has the same problem. A model may be fluent, but fluency is not the same as insurance judgment.

For agencies, that distinction matters.

A generic model can read two quote packets and produce a confident answer. But will it know which differences are material? Will it recognize that two labels may be equivalent? Will it flag placeholder values before binding? Will it avoid guessing when a field is missing? Will it separate premium drivers from coverage differences? Will it know what should be sent back to the carrier, the insured, or the internal agency team?

Not consistently.

That is why ReFocus AI is beginning to open-source practical insurance skills for public LLMs.

The first release is available now in the ReFocus AI Insurance Tools GitHub repository.

What We Mean by “Insurance Skills”

An insurance skill is more than a prompt.

A prompt asks a model to perform a task once. A skill gives the model a repeatable workflow: what to review, what not to assume, how to structure the output, when to flag uncertainty, and where professional review is required.

That structure is what turns a generic AI assistant into something more useful for agency work.

It does not make the model a licensed producer. It does not replace professional judgment. It does not remove the need for review.

But it does help agency leaders bring more consistency, discipline, and insurance-specific context into the public LLMs their teams are already using.

Our First Open-Source Skill: Quote Comparison

ReFocus AI’s first open-source insurance skill is a Quote Comparison Skill.

The skill helps compare two insurance quote documents for the same insured and produce a structured review of:

  • Premium differences
  • Coverage differences
  • Deductibles
  • Rating inputs
  • Discounts and surcharges
  • Taxes and fees
  • Forms and endorsements
  • Missing or suspicious values
  • Bind-readiness cleanup items
  • Producer follow-up questions

The skill is line-of-business agnostic and designed to support quote comparison across personal lines, commercial lines, umbrella, workers' compensation, and other insurance workflows.

For setup and usage details, read the Quote Comparison Skill README.

Once the Knowledge Base article is published, link to it here: How to Use the ReFocus AI Quote Comparison Skill.

Why Quote Comparison First?

Quote comparison is one of the most common places where AI seems simple, but can quickly become risky.

An insurance skill can push the model to compare the right things.

Two quotes may appear similar. The premiums may be different. A producer may need to know why. A CSR may need to check whether anything is missing. An agency principal may want a more consistent review process before quotes are presented or bound.

A generic model can summarize the documents.

An insurance skill can push the model to compare the right things.

That difference matters.

The Quote Comparison Skill file tells the model to normalize formatting differences, review every visible field, flag placeholder or suspicious values, avoid unsupported assumptions, and separate coverage issues from premium drivers. It also asks for follow-up questions grouped by insured, carrier, and internal agency review.

That is the kind of insurance-specific structure public LLMs need before they become useful in real agency workflows.

Why We Are Open-Sourcing This

Historically, many insurance automation workflows have lived inside proprietary systems.

But insurance work does not happen in one system.

Agency teams move across AMS platforms, carrier portals, PDFs, spreadsheets, emails, proposals, and AI assistants. If agencies are already using public LLMs, the industry needs better ways to inject insurance-specific smartness into those tools.

That is the purpose of this open-source insurance skills library.

We want to make practical, reusable insurance workflows available to agencies so they can automate more work safely, consistently, and transparently.

The Quote Comparison Skill is the first release. More skills will follow.

What This Is Not

This is not a replacement for licensed insurance professionals.

It is not legal advice.

It is not a binding recommendation.

It does not access carrier portals, AMS data, policy databases, or outside systems. It works from the documents the user provides.

The goal is not to let AI make insurance decisions on its own. The goal is to help agencies use public LLMs with better structure and better guardrails.

Agency Leaders Should Set the Standard Now

Public LLMs are already entering agency workflows. The choice is not whether teams will try them. The choice is whether agency leaders will give those teams a better structure.

Insurance-specific skills are one practical way to do that.

They help define how the model should behave, what it should compare, what it should not assume, and how it should present uncertainty. They also help agencies establish a shared standard for AI-assisted work rather than relying on ad hoc prompts from individual users.

That is the future we think agencies need: not generic AI, but insurance-aware AI workflows with professional oversight.

Try the Skill

The Quote Comparison Skill is available now on GitHub.

Primary CTA: View the GitHub repository.

To learn how the skill works, read the Quote Comparison Skill README.

To review or copy the underlying instructions, open the Quote Comparison Skill file.

To share feedback, open an issue on GitHub.

We are excited to open this up to the insurance community and continue improving it with real-world input from agency leaders, producers, CSRs, and operators.

Disclaimer

This article and the Quote Comparison Skill are provided for informational and workflow-support purposes only. They do not constitute legal, compliance, or underwriting advice, or a binding insurance recommendation. Insurance professionals should independently review all outputs and consult appropriate legal, compliance, underwriting, carrier, or licensed insurance professionals before making coverage, binding, or client-facing decisions.