Submission Intake Summarizer

Purpose

Parse incoming new business or renewal submission packages and extract key risk details into a structured, at-a-glance summary that enables rapid triage, appetite matching, and assignment to the appropriate underwriter.

When to Use

Use this skill when a new submission lands (via email, portal, or broker package) and you need a quick structured summary before deeper underwriting begins. This is the lightweight complement to the Underwriting Risk Profile Builder — use this for fast triage, and the Risk Profile for full analysis. Especially useful during high-volume submission periods when speed matters.

Required Input

Provide the following:

1. Submission documents — ACORD applications (125, 126, 130, 131, 140, etc.), supplemental questionnaires, broker cover letters, loss runs, SOV/COPE schedules, financials, prior policy forms/endorsements, or any raw submission materials
2. Line of business — Commercial property, GL, auto, workers' comp, professional liability, umbrella/excess, cyber, management liability, environmental, specialty, etc.
3. Submission channel — Broker email, wholesaler portal, agency management system, appetite-finder platform (First Connect, SortSpoke, Brisc), direct bind portal — drives required-field checks and receipt-acknowledgment template
4. Broker relationship context (optional) — Producing broker/agent name, prior-12-month submission volume, hit ratio, loss-ratio of bound business, contingency tier, any open service or compensation issues
5. Carrier appetite reference (optional) — If not already in config.yml, paste or reference the current appetite matrix (approved/declined class codes, state footprint, limit/attachment bands, minimum/maximum premium, required surplus-lines posture)
6. Priority context (optional) — Whether this is a target account, a wholesaler escalation, a large-broker production imperative, or a routine submission — affects the priority score, not the appetite call

Instructions

You are a commercial underwriting assistant. Your job is to quickly parse submission materials and produce a structured summary for underwriter review.

Before you start:

• Load config.yml from the repo root for company details, appetite matrix, surplus-lines posture, and preferences. The following config hooks are authoritative and consumed in the steps below:
  • config.yml.operations.submission_efficiency_rules (new in v4.0 — primary efficiency hook) — per-LoB × per-state × per-class × per-premium-band STP-eligibility threshold library and inference-vs-ask matrix for new and renewal submissions. Every standard submission field is classified required / inferable / confirm-after-infer / ask against the carrier's rule library. Inference-source chain (rule-prescribed in order): prior-policy record → broker-submission packet → AMS/BMS broker record → public-filing record (SEC EDGAR / Secretary of State / state corporation records) → BLS / Census NAICS bridge → OFAC / sanctions list → loss-run-database → ISO experience-mod feed → property-record API → CAT-modeling feed → third-party data-aggregator (Verisk, Moody's RMS, etc.). Each inference carries explicit provenance and confidence; fields above the carrier's confidence threshold are marked INFERRED — <source> — confidence <n>; fields below threshold are marked INFERENCE — UW CONFIRM. The Missing Items Punchlist is now rule-pruned — only asks fields the carrier's rules say must be broker-attested AND that fall below inference-confidence threshold. Every Punchlist entry cites the rule ID. The priority-score weighting (appetite fit / broker tier / premium size / data completeness / strategic flag) is now per-rule — the carrier's rule library names the per-LoB-per-state weighting and the rule ID stamps the priority score. NO EFFICIENCY RULE — UW DECISION flag for unmapped LoB / state / class / premium-band tuples. Inspired by the BriteCore Submission Intake & Readiness Copilot 80–90% manual-intake reduction (2026-05-20 landscape monitor) and Zurich + Cytora 80% triage cut empirical anchor.
  • config.yml.sales.carrier_appetite_map — per-line × per-state × per-class appetite matrix (shared with AI-Native Distribution Channel Spec v2.0 and Cross-Sell Opportunity Analyzer v3.0): approved classes / declined classes / surplus-lines-only classes / referral-required classes / limit-band per class / attachment-band per class / minimum and maximum premium per class / CAT-zone exposure caps / target-loss-ratio band / target-hit-ratio band / target-bind-ratio band / per-class quota status (open / capacity-constrained / closed) / per-class effective date. The appetite verdict step (Step 2 below) is now wired against this hook; the verdict cites the appetite-map entry ID rather than the analyst-authored rule.
  • config.yml.operations.broker_tier_table — per-broker × per-LoB tier-history table: prior-12-month submission volume, hit ratio, bound-business loss ratio, contingency tier, target-account flag, open service or compensation issue. The priority-score broker tier & hit ratio slot is pre-populated from this hook rather than asked. NO BROKER ENTRY — DEFAULT TIER 3 flag for unmapped brokers.
  • config.yml.operations.ofac_sanctions_pattern — sanctions-check pattern (named-insured + owner-list + DBA-list), rule cadence, hit-handling SLA. The OFAC sweep step (Step 2 below) is now rule-prescribed.
  • config.yml.operations.cat_accumulation_map — carrier's CAT accumulation map per peril × geography. The CAT-zone overlay step is now rule-prescribed.
  • config.yml.voice — communication tone
• Reference knowledge-base/terminology/ for correct industry terms including ACORD form numbers, NAICS/SIC bridges, and line-of-business abbreviations
• Reference knowledge-base/regulations/ for surplus-lines eligibility, OFAC/sanctions expectations, state producer-licensing rules, and AI-use disclosure rules where applicable
• Use the company's communication tone from config.yml → voice
• Never fabricate data the broker did not provide — if a required field is missing, list it in the broker-outreach draft, not in the summary body. Inference is permitted only where the efficiency rule classifies the field as inferable or confirm-after-infer and the inference-source chain delivers a confidence score above the carrier's threshold. Every inference carries explicit provenance and confidence in the summary body.

Process:

0. Efficiency-rule lookup (new in v4.0). Resolve the applicable submission_efficiency_rules entry per (LoB × state × class × premium-band) before fact extraction. Stamp the rule ID at the top of the Submission Summary (EFFICIENCY-RULE: <id>). Resolve the applicable carrier_appetite_map entry per (line × state × class) and stamp the appetite-map entry ID alongside. Resolve the applicable broker_tier_table entry per producing broker × LoB and pre-populate the broker block. If no efficiency rule matches the tuple, set NO EFFICIENCY RULE — UW DECISION and proceed with the v3.0 baseline behavior as a fallback (do not block the analysis).

1. Extract and organize the following from the submission package. For every field, apply the efficiency-rule classification (required / inferable / confirm-after-infer / ask) and walk the inference-source chain for any inferable or confirm-after-infer field. Mark each field with its source (provided / inferred / asked) and the inference-confidence score where applicable:

   • Named insured — Full legal entity name, DBA, and entity type (corp, LLC, partnership, etc.)
   • Industry classification — SIC/NAICS code, business description, and class code
   • Effective dates — Requested effective and expiration dates
   • Lines requested — Each line of coverage with requested limits, deductibles, and any specific endorsements
   • Expiring program — Current carrier(s), expiring premium, expiring terms, reason for marketing (if stated)
   • Locations / operations — Number of locations, states of operation, any high-hazard operations
   • Employee / revenue data — Employee count, payroll (for WC), annual revenue, revenue trend
   • Loss history snapshot — Number of years provided, total incurred, notable claims, loss ratio if calculable
   • Broker details — Producing broker/agent, wholesaler (if applicable), submission date

2. Perform quick triage checks (appetite-first, per the 2026 broker-facing pattern):

   • Appetite verdict — Score against the carrier_appetite_map entry resolved in Step 0 and return one of: IN-APPETITE (matches class / state / limit band), BORDERLINE (partial match, needs UW override or flex on one dimension), OUT-OF-APPETITE (clear miss — declined class, unlicensed state, limit outside band), or REFER-SPECIALTY (fits a sister book or surplus-lines channel). Every verdict cites the appetite-map entry ID ([lib:appetite/GL-class-7382-CA]) and the specific dimension that drove it
   • Completeness check — Produce a Missing Items punchlist that is rule-pruned against the efficiency-rule resolved in Step 0. The Punchlist only asks fields the rule classifies as required AND ask (i.e., must be broker-attested AND not inferable) AND that are missing from the broker packet. Inferred fields above the confidence threshold are surfaced in the summary body marked INFERRED — <source> — confidence <n> and do not appear in the Punchlist. Inferred fields below the confidence threshold are surfaced in an internal-only UW-Confirm List (handler confirms internally rather than forcing a second broker contact). Each Punchlist item is tagged must-have vs nice-to-have and the ACORD form or schedule where it is usually found
   • Red flags — Immediate concerns: high loss ratio (>75% 3-year average), litigation history, distressed account signals, prohibited class codes, OFAC/sanctions hit, surplus-lines eligibility gap, prior-carrier non-renewal, broker churn (>2 carriers in 3 years), CAT-zone concentration, recent bankruptcy or material ownership change
   • Surplus-lines / diligence check — If the state requires diligent search and the class is declined by the admitted carrier, flag the surplus-lines eligibility path and the broker's diligence-declaration obligation
   • CAT-zone overlay — For property submissions, overlay the loss locations onto the cat_accumulation_map resolved in Step 0 and flag PML/COPE concerns
   • OFAC / sanctions sweep — Run the named insured, owners, and any DBA names against the ofac_sanctions_pattern resolved in Step 0; flag any hit for compliance before UW touches the file
   • Referral flags — Large-account threshold, reinsurance facultative threshold, loss-sensitive-program threshold, new-venture threshold — each auto-routes to the right UW desk per the efficiency-rule entry

3. Compute a priority score (0–100) using the per-rule weighting resolved in Step 0 (the carrier's rule library names the per-LoB-per-state weighting; the v3.0 defaults — appetite fit 35, broker tier & hit ratio 20, premium size 20, data completeness 15, strategic/target-account flag 10 — apply as a fallback when no rule matches). The broker tier & hit ratio slot is pre-populated from broker_tier_table (no ask). The data completeness slot is computed against the rule-pruned Punchlist. The rule ID and the resolved weighting are both stamped on the priority-score line. Priority is separate from appetite — an in-appetite routine submission can be lower priority than a borderline target account

4. Produce the summary in a consistent, scannable format

5. Draft a broker outreach message tailored to the verdict:

   • IN-APPETITE with gaps — short "received and working it" acknowledgment plus the Missing Items punchlist and a realistic turnaround SLA
   • BORDERLINE — specific flex request ("can you confirm sprinklered throughout" / "are the WC class codes open to audit"), named decision-maker, and decision timeline
   • OUT-OF-APPETITE — polite decline with the specific reason (class / state / limit), a brief note if a sister book or surplus-lines channel could help, and a thank-you that protects the broker relationship
   • REFER-SPECIALTY — internal handoff block (destination desk, why, payload) plus a broker-facing "we are routing internally, you'll hear back from X by Y" note
   • Never quote premium, commit to coverage, or promise a bind decision in the outreach draft

Output requirements:

• Four paired deliverables:
  1. Submission Summary — efficiency-rule stamp (rule ID + appetite-map entry ID + broker tier entry) at top; structured fields (named insured, NAICS/SIC, effective dates, lines, locations, exposures, loss snapshot, broker block, appetite verdict with cited entry, priority score with rule-resolved weighting, red-flag list). Every field is sourced (provided / inferred / asked) with inference provenance and confidence
  2. Missing Items Punchlist (rule-pruned) — must-have vs nice-to-have, ACORD-form location, owner (broker / underwriter / wholesaler), and the realistic turnaround SLA. Only fields the rule classifies as required AND ask AND missing. Every entry cites the rule ID
  3. UW-Confirm List (new in v4.0, internal-only) — fields the rule classifies as confirm-after-infer where the inference confidence is below threshold; the UW confirms internally rather than the broker confirming externally
  4. Broker Outreach Draft — tailored to the verdict (in-appetite, borderline, out-of-appetite, refer-specialty), short, professional, never quoting premium or promising a bind
• Handoff block — the 3-line internal summary an UW can paste into the AMS/PAS comments: appetite verdict (with entry ID), priority score (with rule ID), top red flag, next action owner
• Professional formatting — scannable with bold field labels and concise values; no speculative numbers
• Correct industry terminology (ACORD form references, class codes, coverage abbreviations, SIC↔NAICS bridges)
• Cross-references — if the submission is in-appetite and complete, point to the Underwriting Risk Profile Builder v2.0 for deeper analysis; if a COI is attached, point to the COI Compliance Reviewer v2.0; if prior-carrier loss runs are attached, point to the Loss Run Analyzer v2.0; if the submission arrived via an AI-native channel, cross-reference the AI-Native Distribution Channel Spec v2.0 (which consumes the same carrier_appetite_map)
• AI-disclosure — if the broker outreach draft is sent under a state that requires AI-interaction disclosure (TRAIGA, AB 489, IN HB 1271, AL SB 63, CO SB 21-169 + CO SB 26-189 ADMT carve-out, WA SB 5395, VA HB 736, UT AI-PA, NY DFS Reg 187, or a disclosure-triggering use case), attach the configured disclosure line
• Ready to paste into underwriting workflow or email to the assigned underwriter
• Saved to outputs/submissions/<submission-id>/ if the user confirms, with summary.md, missing-items.md, uw-confirm-list.md, and broker-outreach.md

Versioning

v4.0 (2026-06-01): Efficiency pass — wired five new operations-config hooks. (1) config.yml.operations.submission_efficiency_rules (primary efficiency hook) — per-LoB × per-state × per-class × per-premium-band STP-eligibility threshold library and inference-vs-ask matrix. Every standard submission field is classified required / inferable / confirm-after-infer / ask against the carrier's rule library. Rule-prescribed inference-source chain in order (prior-policy record → broker packet → AMS/BMS record → public-filing record → BLS/Census NAICS bridge → OFAC list → loss-run-database → ISO experience-mod feed → property-record API → CAT-modeling feed → third-party aggregator) with explicit provenance and confidence. Missing Items Punchlist is now rule-pruned. Internal-only UW-Confirm List for sub-threshold inferences. Per-rule priority-score weighting replaces the v3.0 fixed defaults. (2) config.yml.sales.carrier_appetite_map — shared with AI-Native Distribution Channel Spec v2.0 and Cross-Sell Opportunity Analyzer v3.0; appetite verdict cites the entry ID rather than the analyst-authored rule. (3) config.yml.operations.broker_tier_table — pre-populates the priority-score broker-tier slot (no ask). (4) config.yml.operations.ofac_sanctions_pattern — rule-prescribes the OFAC sweep. (5) config.yml.operations.cat_accumulation_map — rule-prescribes the CAT-zone overlay. New Step 0 (efficiency-rule lookup) before fact extraction. New EFFICIENCY-RULE: <id> stamp on every Submission Summary. New INFERRED — <source> — confidence <n> provenance system on every inferred field. NO EFFICIENCY RULE — UW DECISION flag for unmapped tuples. Inspired by the BriteCore Submission Intake & Readiness Copilot 80–90% manual-intake reduction (2026-05-20 landscape monitor) and Zurich + Cytora 80% triage cut empirical anchor.

Dimension moves: Efficiency 8 → 9 (the five new operations-config hooks + inference-source chain + rule-pruned Punchlist + UW-Confirm List + per-rule priority-score weighting). All other dimensions held. Strict superset of v3.0 — every v3.0 capability is preserved (three paired deliverables now four, appetite verdict, completeness check, red-flag list, surplus-lines / diligence check, CAT-zone overlay, OFAC sweep, referral flags, priority score, handoff block, cross-references, AI-disclosure, four-channel-pattern coverage). Cross-references updated to v2.0 / v3.0 / v4.0 / v5.0 versions of the dependent skills.

v3.0 (2026-04-24): Added appetite-first triage pattern with IN-APPETITE / BORDERLINE / OUT-OF-APPETITE / REFER-SPECIALTY verdicts; CAT-zone overlay; OFAC / sanctions sweep; surplus-lines / diligence check; referral-flag auto-routing; verdict-tailored broker outreach drafts; handoff block; priority score (0–100) with configurable weighting.

v2.0: Added structured-field extraction across nine canonical fields; Missing Items Punchlist with ACORD-form location; broker outreach draft.

v1.0: Initial release — submission summary.

Example Output

[This section will be populated by the eval system with a reference example. For now, run the skill with sample input to see output quality.]