Outcomes, not adjectives. Six carrier stories.

The reinsurer's treaty underwriters were drowning in unstructured broker submissions — slip PDFs, bordereaux spreadsheets, cat model outputs, and prior loss runs — averaging four days per submission before an underwriter could price. Renewals compressed the queue every October and April, and hiring more underwriters was neither fast nor economic.

• Document AI pipeline: layout-aware extraction across slips, bordereaux, and loss runs with confidence scoring
• Treaty Analyzer copilot that pre-fills the pricing worksheet and flags data gaps for the underwriter
• 60-person offshore triage pod trained on the carrier's product hierarchy, resolving exceptions the model couldn't
• Human-in-the-loop review with full audit trail for Lloyd's and Solvency II model risk requirements

The reinsurer now clears the same volume with the original underwriting bench — underwriters spend their day pricing risk, not chasing missing schedules. The pod scales up 3× during renewal peaks and back down in weeks four to six of the cycle.

The carrier's contact center absorbed 2,400 daily first notices of loss. Simple glass-only and minor collision claims consumed 40% of adjuster time, degrading NPS on the complex bodily-injury claims where empathy actually mattered.

• Computer-vision damage classification at the mobile intake step
• Rule + LLM triage: obvious claims routed to STP settlement, ambiguous ones queued to a specialist
• Vietnam and Manila 24/7 review pod for the ambiguous middle band
• Model governance framework aligned to the NAIC Model Bulletin and state DOI expectations

58% of FNOLs now close without an adjuster touch. Complex claim NPS climbed 22 points because senior adjusters got their time back. The model card and human-review evidence pack passed a multi-state DOI review with no adverse findings.

The composite carrier ran life admin on a COBOL mainframe patched across three decades of product launches. Every new rider needed six months of change requests, and the vendor had announced end-of-support. The board rejected the original 5-year big-bang replatform.

• Strangler-fig migration onto Guidewire InsuranceSuite Cloud, product family by product family
• Book-of-business data reconciliation platform with per-policy attestation for MAS, IRDAI and HKIA
• Parallel-run harness comparing legacy and target outputs down to the cent for six months
• 180-engineer joint pod (FPT + carrier) with shared standups, shared SLAs, shared on-call

The mainframe was decommissioned on schedule. New rider launches now take six weeks instead of six months. The unit economics of servicing an in-force policy dropped 42% — the business case paid back inside 30 months of go-live.

The MGA suspected recoverable third-party liability was slipping through into closed files — but their subrogation team could only review the top 5% of claims by reserve. Everything below the threshold got closed and forgotten.

• LLM extraction over five years of adjuster notes, police reports, and photographs
• Liability-signal scorer prioritizing files with third-party contact, statutory duty breach, or documented fault
• Human specialist queue for the highest-signal files, integrated to the existing recovery workflow
• EU data residency with Frankfurt hosting and a completed GDPR DPIA for reopening closed files

The MGA reopened 180,000 claims and recovered €22M in year one against a project cost under €2.5M. The model now runs continuously — every newly closed file gets a subrogation score before it goes to archive.

Cyber submissions arrived as 90-page questionnaires plus attestations, SOC 2 reports, and external scan attestations. Underwriters spent 70% of their time transcribing data before they could apply the pricing model. Response times pushed brokers to competitors.

• Submission Copilot pulling questionnaire, external scans, dark-web signals, and SEC filings into one underwriting workspace
• Cyber-specific extractor tuned on MFA, backup, EDR, and IR posture questions
• Underwriter-in-the-loop pricing — the copilot proposes, the underwriter binds
• SR 11-7 aligned model governance with quarterly evaluation reports for the Bermuda regulator

Underwriters price submissions in under two hours instead of a full day. New-business combined ratio improved 45% as pricing consistency tightened, and the carrier is now the preferred market for two of the top-five cyber brokers.

The mobility platform wanted to embed trip protection into every ride and rental checkout across LATAM, APAC, and EMEA. Every country had a different carrier partner, product form, and regulator. A bespoke integration per market would have taken years.

• Composable product factory: one API contract, country-specific product bundles configured, not coded
• Quote/bind/issue services deployed on regional Kubernetes clusters for data residency
• Partner onboarding kit that let a new country carrier go live in three weeks end-to-end
• Real-time telemetry and A/B framework tied to the checkout funnel

14 markets went live in Q3 with a 15th added post-launch. Attach rate stabilized at 8.4% — twice the platform's internal benchmark — and the product now contributes a material line to platform revenue with a fraction of the engineering headcount that similar programs required.