Insurance

FCA ICOBS, Solvency II — Insurance claims and underwriting

Quick Reference

minimal.py

python

result = agent.check(
    case_type="motor_claim",
    depth="standard_review",
    documents=[open("claim_form.pdf", "rb")],
    reference="CLM-2026-00812",
)

Case Types

Name	API ID	Description
Motor Claims	`motor_claim`	Vehicle accident, theft, and damage claims
Property Claims	`property_claim`	Buildings and contents insurance claims
Liability Claims	`liability_claim`	Professional indemnity and public liability claims

Modifiers

Name	API ID	Applies To	Description
High Value	`high_value`	`motor_claim`, `property_claim`	Claims above £50k threshold — adds enhanced documentation requirements
Cross-Border	`cross_border`	All case types	Claims involving multiple jurisdictions — adds cross-jurisdiction regulatory checks
Subrogation	`subrogation`	`motor_claim`, `liability_claim`	Third-party recovery — adds subrogation opportunity assessment

Depths

Name	API ID	Duration	Scoring	Price	Description
Triage	`triage`	~30s	`pass_fail`	£0.50	Document presence, basic eligibility, obvious red flags
Standard Review	`standard_review`	~2 min	`rag`	£3.00	Full claims handling compliance, reserve adequacy, fraud indicators
Deep Audit	`deep_audit`	~3 min	`rag`	£8.00	Standard review + cross-claim pattern analysis, network-level fraud detection

Scoring Modes

`pass_fail`

Used by triage. Each check returns pass or fail. The overall result is pass if all checks pass, or fail if any check fails.

`rag`

Used by standard_review and deep_audit. Each check returns green (compliant), amber (review recommended), or red (non-compliant or fraud indicator). Amber and red findings include descriptions and remedial actions. The overall triage is the worst-case result across all checks.

Worked Example

Motor claim at standard_review depth.

Request

request.py

python

from curvestone import Agent

agent = Agent()

result = agent.check(
    case_type="motor_claim",
    depth="standard_review",
    documents=[
        open("claim_form.pdf", "rb"),
        open("incident_report.pdf", "rb"),
    ],
    reference="CLM-2026-00812",
)

Response

response.json

json

1{
2  "id": "job_9pLm3xRtK7",
3  "type": "check",
4  "status": "completed",
5  "triage": "amber",
6  "reference": "CLM-2026-00812",
7  "processing_time": "118s",
8  "scoring": "rag",
9  "checks": [
10    { "name": "Claims Handling Compliance", "triage": "green" },
11    { "name": "Reserve Adequacy",          "triage": "amber" },
12    { "name": "Fraud Indicators",          "triage": "green" },
13    { "name": "Subrogation Opportunity",   "triage": "green" }
14  ],
15  "findings": [
16    {
17      "severity": "amber",
18      "check": "Reserve Adequacy",
19      "finding": "Initial reserve of £4,200 is below the model estimate of £6,800 for this claim type and severity. Historical claims with similar characteristics have settled at an average of £7,100.",
20      "remedial_action": "Review the reserve estimate against comparable claims data. Consider increasing the reserve to at least £6,800 to align with actuarial guidance for this claim category."
21    }
22  ],
23  "cost": "£3.00"
24}

Ask

Use POST /ask to ask natural-language questions about a completed insurance job, a document, or claims data.

Example questions

→Why was this claim flagged for potential fraud?
→What comparable claims data supports the current reserve estimate?
→Is the subrogation recovery opportunity viable based on the policy terms?
→Summarise the compliance issues found in this batch of claims.
→What are the key risk indicators in the motor claims portfolio?

Questions can reference a job_id for context-aware answers grounded in the original documents and findings. £0.05 per turn.

Monitor

Use POST /monitor to set up recurring watches relevant to the insurance cluster.

Monitor type	What it watches	Typical schedule
regulation_change	FCA ICOBS updates and Solvency II changes	weekly
portfolio_drift	Claims portfolio for emerging patterns or anomalies	daily
website_change	Broker and claims handler websites for compliance	weekly

Monitors can trigger downstream actions via on_finding — e.g. automatically re-run a check when lender criteria change. £15/month per monitor.