Use-case library

Each entry describes an insurance workflow, the business outcome it targets, and what has been reported about it. Reported results show that a workflow can be built; they do not predict what it will return for your company.

How to use this library

Start from the outcome you need to improve, then read an entry's prerequisites and controls before its reported results. The operating mode follows the suggested sequence on Costs & value, and each entry's suggested order is its position in that page's nine-item evaluation order:

• Assistive. AI drafts; a person decides.
• Decision support. The model scores or recommends; a named person holds authority.
• Bounded automation. Gated work on defined segments, behind Phase 3 controls.

The mode describes the proposed use. Set the controls from what the system actually does: a leakage report that a person reviews is different from a system that blocks payments.

Underwriting capacity and speed to quote

Submission intake and triage

Outcome: underwriting capacity · Mode: assistive · Suggested order: 1

• Workflow. Reads broker emails and attachments, extracts and validates key fields, checks appetite and clearance, and flags incomplete or out-of-appetite submissions for an underwriter.
• Inputs. Submission emails, applications, loss runs, schedules of values, and the current appetite guide.
• Reported. AIG reports more than 370,000 Lexington submissions processed in 2025, up 26% year over year; that growth does not isolate AI's contribution. Cytora reports a 70% cut in Arch's intake-to-decision time, from a cycle of two to three days. Cytora's 2023 Markel case study reports 113% higher gross written premium per full-time equivalent. All are carrier- or vendor-reported.
• Prerequisites and controls. A named underwriter keeps the decision. Extracted fields link to their source, and missing or conflicting information follows a referral route instead of being assumed. Measure preparation time after exceptions and corrections.
• Guidance. The pilot pattern on Insurance & specialty, Phase 2, and the AIG and Markel source summaries.

Document extraction and summarization

Outcome: handling time across functions · Mode: assistive · Suggested order: 2

• Workflow. Extracts fields from loss runs, schedules of values, financials, and claims files, and drafts a summary for a reviewer to check.
• Inputs. The document set and a defined field list or summary template.
• Reported. No general savings range is assumed on this site. Measure field accuracy and net handling time, including review and rework.
• Prerequisites and controls. Confidentiality, accuracy, and downstream use set the controls; a summarization task is not automatically low risk. Values the source does not support stay flagged.
• Guidance. The extract-then-check exercise on Hands-on use and the worked loss-run pilot review.

Bounded agentic quoting

Outcome: speed to quote · Mode: bounded automation · Suggested order: 9 (last)

• Workflow. Takes a low-complexity submission from email through clearance, enrichment, and a pricing indication to a quote recommendation that an underwriter approves.
• Inputs. The submission, rating tools, appetite rules, and permitted external data.
• Reported. Hiscox reported that a 2023 proof of concept cut the time to quote a terrorism risk from three days to about three minutes. CFC bills Lane Assist as a world-first agentic underwriting pilot in specialty, moving low-complexity cyber submissions from email to quote recommendation in seconds. Cytora's Markel case describes a strategic-partner quote service commitment of two hours, down from 24; that is a commitment, not an average. None of these is a production average.
• Prerequisites and controls. Phase 3 gates, including human-override logs and drift monitoring. Put speed where the broker sees it and keep depth behind it.
• Guidance. Phase 3, the Hiscox and CFC source passages, and the harness controls.

Claims handling and recovery

Claims triage, severity and litigation prediction

Outcome: claims handling and litigation cost · Mode: decision support · Suggested order: 3

• Workflow. Scores new claims at first notice of loss for severity and likelihood of litigation, and routes them to an appropriate handler.
• Inputs. First-notice data, claim notes and documents, and historical claim outcomes.
• Reported. CLARA's vendor data put average indemnity for attorney-involved casualty claims at about 4.9 times that for unrepresented claims. The association explains the interest in early triage; it does not establish savings caused by AI.
• Prerequisites and controls. A claims professional holds authority. Test on reviewed historical files, minimize claimant data in prompts and logs, and monitor for drift.
• Guidance. Phase 2, the CLARA source passage, and Governance.

Fraud, leakage and subrogation

Outcome: claims leakage and recoveries · Mode: decision support (fraud); bounded automation (leakage, subrogation) · Suggested order: 5 and 8

• Workflow. Flags possible fraud for investigation, and checks paid claims for duplicate payments, missed recovery opportunities, and policy exceptions.
• Inputs. Claims and payment data, policy terms, and investigation outcomes.
• Reported. A case study reports 5x more fraud detected at Tokio Marine (vendor-reported). The site cites no reported result for leakage or subrogation.
• Prerequisites and controls. A person reviews every fraud score. Test leakage checks on reviewed historical files, and measure confirmed recoveries and false positives before attributing savings.
• Guidance. Phase 3 for leakage and subrogation pilots.

Pricing, actuarial research and portfolio steering

Actuarial filing research and pricing workbench

Outcome: pricing and actuarial research time · Mode: assistive · Suggested order: 7

• Workflow. Researches rate filings and pricing comparisons, and supports model development by profiling data, generating rater code and tests, and maintaining documentation.
• Inputs. Public rate filings, internal rating data, and existing models and documentation.
• Reported. An Akur8 customer testimonial describes pricing-comparison answers in minutes rather than hours or days. hyperexponential reports 13 pricing tools delivered in 13 weeks at Allianz Commercial.
• Prerequisites and controls. Actuaries retain assumptions, validation, rate, and release authority. Document outcomes testing before any selection or pricing use.
• Guidance. The practical pricing workflow and Phase 2.

Portfolio steering

Outcome: portfolio mix and combined ratio · Mode: decision support (proposed) · Phase 3; not ranked

• Workflow. Monitors the portfolio continuously, surfaces accumulations, and informs appetite steering.
• Inputs. Policy and exposure data, claims experience, and appetite and capacity limits.
• Reported. BCG estimates potential gains of 1–3% in gross premium written and a 1 to 2.5 point reduction in the combined ratio. These are estimates, not observed results.
• Prerequisites and controls. Data readiness established in earlier phases. People set appetite and approve changes to it.
• Guidance. Phase 3.

Operations and knowledge access

Bordereaux processing

Outcome: delegated-authority oversight · Mode: decision support · Suggested order: 4

• Workflow. Ingests and validates bordereaux from delegated-authority partners, reconciles them, and flags exceptions.
• Inputs. Premium and claims bordereaux and the governing authority terms.
• Reported. Verodat reports 85–94% processing time savings (vendor-reported).
• Prerequisites and controls. Validate exception handling and reconciliation on your own bordereaux, with a named owner for exceptions.
• Guidance. Phase 2.

Knowledge access over guidelines

Outcome: finding current guidance · Mode: assistive · Suggested order: 6

• Workflow. Answers staff questions from current guidelines, appetite guides, and procedures, citing the source passage.
• Inputs. Approved internal documents with their access permissions.
• Reported. No reported result is cited. Test whether staff find current, cited guidance faster, and track unsupported answers, stale material, and permission failures.
• Prerequisites and controls. Internal use only, document permissions enforced, and an owner for keeping sources current.
• Guidance. Phase 1.

From library to pilot

Pick the entry closest to a bottleneck you can measure today, with someone qualified to check the output. Then:

1. Record volume, handling time, quality standard, and current cost on the pilot decision sheet.
2. Agree the standard the pilot gate will use to expand, revise, or stop.
3. If you are buying, use Vendor evaluation to request results on comparable files, a definition of each metric, and the cost of exceptions and review.
4. Confirm the controls in Governance before the workflow touches underwriting, pricing, claims, or customers.