← Back to Insurance

Read a loss run

Before any commercial submission with claims history.

ExpertSTEP BY STEPDecision-making

loss run · underwriting · submission · commercial lines · claims history · advanced

Pro tip

Open reserves move. A run pulled two months ago is a different document than the one pulled today.

How to use this prompt

  1. Pick your AI model. Choose the tab for Claude, ChatGPT, Gemini or Copilot. Each variant is tuned for that model.
  2. Copy the full prompt. Click Copy Full Prompt to copy the text to your clipboard.
  3. Paste into your AI tool. Open your chosen model and paste the prompt into a new chat.
  4. Replace the [placeholders]. Swap any bracketed fields for your company name, audience, product or tone.
  5. Run and refine. Review the output. If anything is off, ask the AI to tighten tone, length or format.

Prompt Variants by Model

Claude Sonnet 5
Act as a producer reading a loss run before it goes to market. Tell me what this loss run says and what story I tell an underwriter about it.

The loss run: [PASTE IT]
The business: [WHAT THEY DO]
What changed since the worst year: [SAFETY, STAFF, EQUIPMENT, PROCESS, OWNERSHIP, NOTHING]
Where I am submitting it: [MARKETS, IF KNOWN]

Produce, in order:
0. ASSUMPTIONS. Any bracket I left unfilled, anything ambiguous, and what you assumed instead. Write "none" if there were none
1. Frequency and severity, described in words, with the pattern if there is one
2. The years or claims that will worry an underwriter most, and why
3. Whether the pattern is getting better or worse, and how confident that read is
4. The honest story for the cover note. Never a spin, a story
5. What to fix before the next submission
6. WHAT THIS LOSS RUN DOES NOT SHOW. Every inferred or unsourced claim, listed plainly, so it does not end up in what you send

Rules:
- Anything you inferred rather than were told goes in "WHAT THIS LOSS RUN DOES NOT SHOW" instead, never in both places
- Distinguish an open reserve from a paid loss. If the run does not say, put it in the verify section
- Never describe a loss run as clean unless every year in it is shown
- If fewer than three years are shown, say so before anything else
- Hard limit: 600 words for the whole response. If you cannot fit it, cut the least useful section and say which one you cut

Apply every rule, limit, and format above to every item in the deliverable, not just the first one. Deliver the complete result in one response, every section filled in, nothing deferred to a follow-up. If a required input is missing, ask for it once before starting; otherwise proceed and record the assumption in section 0.

Notes: R42 build; insurance set. Tuned to vendor prompting guidance.

Frequently Asked Questions

What does the Read a loss run prompt do?

Before any commercial submission with claims history.

Which AI models is this prompt written for?

Claude, ChatGPT, Gemini and Copilot. Each one gets its own version up top, written the way that vendor says to prompt its own model. Pick the tab that matches yours.

Do I need a paid AI account to use this prompt?

No. This prompt is written to run on the free tier of Claude, ChatGPT, Gemini and Copilot. Paid tiers simply give you longer context windows and faster responses.

Can I customize this prompt for my business?

Yes. Any text inside square brackets is a placeholder you replace with your own business details, such as company name, audience, product or tone. You can also ask the AI to adjust format, length or style after the first output.

When was this prompt last checked?

The date is on each version up top. We check every prompt in full every month, and when a major new model lands we rewrite the prompt for it. One glance tells you how current it is before you paste.

More in Insurance

Browse all prompts →