xpertsystems/ins009-sample

INS-009 — Synthetic Underwriting Intelligence Dataset (Sample)

XpertSystems.ai Synthetic Data Platform · SKU: INS009-SAMPLE · Version 1.0.0

This is a free preview of the full INS-009 — Synthetic Underwriting Intelligence Dataset product. It contains roughly ~10% of the full dataset at identical schema, market cycle calibration, and UW workflow modeling, so you can evaluate fit before licensing the full product.

File	Rows (sample)	Rows (full)	Description
underwriting_records.csv	~5,000	~50,000	Per-submission records (161 columns)
underwriter_performance_summary.csv	~150	~150	Bound-only UW performance KPIs
segment_loss_ratio_table.csv	~62	~65	Loss ratio by LOB × UW tier
pricing_adequacy_distribution.csv	~25	~30	Pricing tier × UW tier adequacy distribution

Dataset Summary

INS-009 simulates the full commercial underwriting submission lifecycle — from broker submission through risk assessment, pricing, binding, and in-force performance — with realistic UW tier hierarchies and market cycle modeling.

Calibration sources (named, authoritative):

• NAIC Industry Aggregate Reports — combined ratio, loss ratio by LOB
• A.M. Best Combined Ratio annual reports
• Conning Strategic Study on UW Performance
• McKinsey U.S. P&C insurance analytics
• PwC Commercial Insurance UW Survey — bind / decline / NTU rates
• ISO loss costs — base loss ratio calibration

13 lines of business:

• Commercial property
• General liability
• Commercial auto
• Workers compensation
• Professional liability (E&O)
• Directors & Officers
• Cyber (first-party / third-party / combined)
• Marine cargo
• Inland marine
• Excess / umbrella
• Personal auto
• Homeowners
• Specialty (accident & health, aviation, agriculture, event cancellation, surety)

Market cycle modeling (configurable in full product):

• Hard market: rate increases, restricted capacity, higher decline rates, tighter terms
• Soft market: rate decreases, abundant capacity, lower decline rates, loose terms
• Transitional: mixed signals, varying by LOB

7 submission outcomes:

• Bound (~43%)
• Declined (~23%)
• Quoted not taken (~20%)
• Withdrawn (~6%)
• Incomplete/abandoned (~5%)
• Referred out (~2%)
• Remarket (~1%)

5 underwriter tier hierarchy:

• Junior UW (entry-level, lower binding authority)
• Mid-level UW (standard book)
• Senior UW (specialty / large accounts)
• Principal UW (major accounts, complex risks)
• Chief Underwriter (executive, portfolio steward)

UW skill gradient is empirically modeled: senior tiers produce better loss ratios and tighter pricing adequacy than junior tiers (realistic experience curve effect).

Submission/insured features (40+ columns):

• Submission ID, carrier ID, broker tier, distribution channel
• Insured: legal entity type, ownership structure, years in business, revenue/payroll/headcount, NAICS code, geographic spread
• Publicly traded flag, regulatory jurisdiction
• Submission completeness score
• Submitted ACORD flag, application type

Risk assessment features (30+ columns):

• Credit score (commercial), prior claims history
• Loss ratio history (prior, 5yr avg, segment benchmark)
• Experience modification factor (mod)
• Technical risk score
• Underwriter judgment score
• Risk quality score
• Final composite score
• CAT zone exposure, peril concentration
• Risk-improvement recommendations issued

Pricing & coverage (40+ columns):

• Quoted premium, written premium
• Pricing adequacy ratio (target = 1.0)
• Pricing tier (preferred / standard / non-standard / referral / bespoke / minimum)
• Rate adequacy filing flag
• Coverage limits (primary, retention, excess)
• TIV (total insured value)
• Commission rate, broker tier
• Expected loss cost, reinsurance cost
• ROEL (return on expected loss)
• Rate change vs expiring
• Cat load, expense load, profit & contingency

Binding & policy (20+ columns):

• Submission outcome (7 classes)
• Effective date, term length
• UW authority level used
• Manual referral count
• Decline reason taxonomy

In-force performance (20+ columns):

• Earned premium, current period incurred
• Loss ratio current
• Loss ratio segment benchmark
• Loss ratio vs benchmark
• IBNR estimate
• Adverse / favorable development flags

Regulatory & financial (10+ columns):

• IFRS 17 LRC / LIC / Risk Adjustment
• IFRS 17 loss component flag
• Solvency II SCR allocation
• Rate filing required, jurisdiction approval status

Calibrated Validation Results

Sample validation results across 10 underwriting-intelligence KPIs:

Metric	Observed	Target	Source	Verdict
n_lines_of_business	13	13	13 LOBs in product taxonomy	✓ PASS
n_underwriter_tiers	5	5	5 UW tier hierarchy	✓ PASS
bind_rate_pct	42.90	42.00	Commercial UW bind rate	✓ PASS
decline_rate_pct	22.64	22.00	Commercial UW decline rate	✓ PASS
quoted_not_taken_rate_pct	20.16	20.00	Commercial UW NTU rate	✓ PASS
pricing_adequacy_ratio_mean	0.961	1.000	Target pricing adequacy	✓ PASS
bound_loss_ratio_mean_pct	73.55	70.00	P&C industry loss ratio	✓ PASS
uw_tier_lr_gradient_pct	17.01	15.00	Junior-Senior LR gap (skill)	✓ PASS
composite_risk_score_mean	55.83	55.00	Composite score mid-range	✓ PASS
submission_completeness_mean	72.83	70.00	Completeness score (data quality)	✓ PASS

Note: The uw_tier_lr_gradient_pct metric measures the loss-ratio gap between junior and senior underwriters. A positive gap is correct: senior UWs produce better books due to selection and pricing skill. This is a key training signal for ML models predicting UW performance trajectory.

Schema Highlights

The 161-column schema is extensive. Key groupings:

Submission identification: submission_id, carrier_id, line_of_business, underwriter_id, underwriter_tier, broker_tier, distribution_channel, submission_date, effective_date.

Insured profile: legal_entity_type, ownership_structure, years_in_business, naics_code, naics_description, annual_revenue_usd, annual_payroll_usd, employee_headcount, publicly_traded_flag, multistate_operations_flag.

Risk scoring: credit_score_commercial, prior_loss_ratio_pct, loss_ratio_5yr_avg_pct, experience_mod_factor, technical_risk_score, underwriter_judgement_score, risk_quality_score, final_composite_score, submission_completeness_score, cat_zone, cat_concentration_pct.

Coverage: primary_limit_usd, retention_usd, total_insured_value_usd, deductible_usd, sublimit_count, optional_endorsement_count.

Pricing: quoted_premium_usd, written_premium_usd, pricing_adequacy_ratio, pricing_tier, base_rate_per_unit, schedule_rating_credit_pct, experience_credit_pct, commission_rate_pct, rate_change_vs_expiring_pct.

Outcome: submission_outcome (7 classes), decline_reason, referral_count, days_to_quote, days_to_bind.

Performance: earned_premium_usd, current_period_incurred_usd, loss_ratio_current_pct, loss_ratio_segment_benchmark_pct, adverse_development_flag, favorable_development_flag.

Regulatory: ifrs17_lrc_usd, ifrs17_lic_usd, ifrs17_risk_adjustment_usd, ifrs17_loss_component_flag, solvency_ii_scr_allocation_usd, rate_filing_required_flag.

Suggested Use Cases

• Submission triage — predict probability of binding from submission features
• UW workflow automation — predict which submissions need manual referral
• Quote-to-bind conversion prediction (NTU vs bound classification)
• Decline reason classification — multi-class decline taxonomy
• Pricing adequacy modeling — regression on pricing_adequacy_ratio
• UW tier performance ranking — predict junior vs senior UW outputs
• Risk score calibration — train composite_score predictors from features
• Loss ratio forecasting at bind time — predict future LR from submission
• Adverse development early warning for in-force policies
• Pricing tier classification (6-class: preferred → bespoke → minimum)
• Market cycle detection — train on hard/soft/transitional data
• NAICS-based risk scoring
• Cyber UW automation — first-party vs third-party vs combined modeling
• Workers comp class code rating
• Commission rate optimization by broker tier
• Reinsurance cost forecasting by LOB and TIV
• IFRS 17 LRC/LIC modeling at policy issuance
• Insurtech UW model training without proprietary submission data

Loading the Data

import pandas as pd

submissions = pd.read_csv("underwriting_records.csv")
uw_perf     = pd.read_csv("underwriter_performance_summary.csv")
seg_lr      = pd.read_csv("segment_loss_ratio_table.csv")
pricing_dist= pd.read_csv("pricing_adequacy_distribution.csv")

# Binary bind prediction
y_bind = (submissions["submission_outcome"] == "bound").astype(int)

# Multi-class submission outcome (7 classes)
y_outcome = submissions["submission_outcome"]

# Regression: pricing adequacy ratio (bound only)
bound = submissions[submissions["submission_outcome"] == "bound"]
y_adequacy = bound["pricing_adequacy_ratio"]

# Regression: bound loss ratio
y_lr = bound["loss_ratio_current_pct"]

# Multi-class UW tier prediction (5 tiers)
y_tier = submissions["underwriter_tier"]

# Multi-class pricing tier prediction (6 tiers)
y_pricing_tier = submissions["pricing_tier"]

# Multi-class LOB classification (13 LOBs)
y_lob = submissions["line_of_business"]

# Composite risk score regression
y_score = submissions["final_composite_score"]

License

This sample is released under CC-BY-NC-4.0 (free for non-commercial research and evaluation). The full production dataset is licensed commercially — contact XpertSystems.ai for licensing terms.

Full Product

The full INS-009 dataset includes ~50,000 underwriting submission records across 161 columns, with configurable market cycle (hard / soft / transitional), underwriter count, carrier count, and LOB filtering. Calibrated to NAIC Industry Aggregates, A.M. Best Combined Ratio, Conning UW Performance, McKinsey U.S. P&C analytics, and PwC Commercial UW Survey.

📧 pradeep@xpertsystems.ai 🌐 https://xpertsystems.ai

Citation

@dataset{xpertsystems_ins009_sample_2026,
  title  = {INS-009: Synthetic Underwriting Intelligence Dataset (Sample)},
  author = {XpertSystems.ai},
  year   = {2026},
  url    = {https://huggingface.co/datasets/xpertsystems/ins009-sample}
}

Generation Details

• Generator version : 1.0.0
• Random seed : 42
• Generated : 2026-05-16 20:59:33 UTC
• Market cycle : transitional
• Records : 5,000
• Underwriters : 150 / Carriers: 20
• Calibration basis : NAIC + A.M. Best + Conning + McKinsey + PwC
• Overall validation: 100.0 / 100 (grade A+)