underwriting-narrative

# Underwriting Narrative Skill An underwriting file must let a peer reconstruct *why* the risk was written, at those terms, at that price — years later, possibly in front of an auditor or a large loss. This skill writes that narrative: the risk story, the numbers behind it, and the reasoning connecting them to the terms offered. ## What This Skill Produces - A risk story (who the insured is, what they do, why they're buying now) - Exposure quantification (values at risk, limits deployed, estimated/probable maximum loss) - A loss-history read distinguishing frequency from severity signals - Mitigating and aggravating factors, weighed - Terms rationale — subjectivities, exclusions, deductibles, each with a reason - Appetite fit and a refer-or-bind recommendation ## Required Inputs Ask for what's missing; from a thin submission, proceed and mark gaps `[information required before bind]`: - **The risk** — insured, operations, geography, line of business - **Exposure figures** — sums insured/TIV, revenue, headcount, limits sought - **Loss history** — ideally 5 years, with dates, causes, incurred amounts, open/closed - **Proposed terms** — limit, deductible, premium, exclusions, subjectivities - **Appetite/guidelines context** — target classes, referral thresholds, if available ## Narrative Framework **Risk story.** Three questions: who is this insured (operations, scale, tenure), what exactly is the exposure (the loss scenarios this line responds to), and *why now* (new buyer, remarketing, mid-term change)? A remarketed risk needs its reason stated — price, service, or non-renewal by the incumbent are very different signals. **Exposure quantification.** State total values at risk, the limit deployed against them, and an estimated maximum loss with the assumption behind it (e.g. single-site fire, top-location concentration). Limits materially above realistic maximum loss, or below it, both need a sentence. **Loss-history read.** Separate the two signals: - **Frequency** (many small losses) → a process/controls problem; responds to deductibles and risk management conditions. - **Severity** (rare large losses) → a volatility/limits problem; responds to price, limit management, and exclusions. Compute a rough loss ratio against premium if figures allow. Narrate any single loss over ~20% of annual premium individually: cause, fix, recurrence risk. A clean record with low tenure is *absence of data*, not evidence of quality — say so. **Mitigating vs aggravating.** List both columns honestly. Mitigants must be verifiable (sprinklers *confirmed*, not "believed"); unverified mitigants become subjectivities. **Terms rationale.** Every non-standard term earns its line: each exclusion tied to an exposure you're declining to price; each subjectivity with a deadline and what happens if unmet; deductible tied to the frequency read. **Appetite fit and recommendation.** In / edge-of / outside appetite, against which guideline. Recommend **bind**, **bind subject to**, **refer** (naming the referral trigger hit), or **decline** — with the one-paragraph reason. ## Output Format ### Underwriting narrative: [insured / line / inception date] **1. Risk story** — who, what, why now. **2. Exposure** — table: values at risk | limit sought | EML basis | premium. **3. Loss history** — frequency vs severity read, loss ratio, large-loss narratives. **4. Factors** — mitigating | aggravating, two columns, weighed in a closing sentence. **5. Terms & subjectivities** — each with rationale and deadline. **6. Appetite fit** — guideline cited, in/edge/outside. **7. Recommendation** — bind / bind subject to / refer / decline, with reason. End with: *"This narrative is analytical support, not a binding decision. Authority, referral, and bind decisions follow your organisation's underwriting guidelines and applicable regulation."* ## Quality Checks - [ ] The "why now" of the submission is answered, especially for remarketed business - [ ] Loss read explicitly separates frequency from severity and states which one drives terms - [ ] Every exclusion and subjectivity has a stated rationale; subjectivities have deadlines - [ ] Unverified mitigants are converted to subjectivities, not counted as credits - [ ] Recommendation names the specific referral trigger if referring - [ ] Data gaps are marked `[information required before bind]`, not papered over ## Anti-Patterns - [ ] Do not write a description in place of a narrative — every fact must connect to a term, a price, or the recommendation - [ ] Do not treat a short clean loss record as proof of good risk — label it as limited data - [ ] Do not list a mitigant you cannot verify without making it a subjectivity - [ ] Do not bury an outside-appetite feature in the middle of the file — surface it in the recommendation - [ ] Do not invent loss figures or survey findings — mark unknowns `[to confirm]`