An independent assurance overlay above pricing, underwriting, fraud and claims models. We do not set rates and we do not sign the actuarial opinion. We evidence that each model was governed, that its inputs were permissible, that adverse decisions can be explained to the policyholder and the regulator, and that the whole decision can be reconstructed later.
Cabier does not act as actuary, appointed actuary or rate filer, and does not replace the insurer's own model risk function.
Resolved into testable control statements. Applicability is confirmed per engagement — jurisdiction, product and deployment scope all change what binds.
NAIC AI model bulletin
A written AIS programme with governance, risk management and third-party oversight that survives a market-conduct examination rather than a questionnaire.
State unfair-discrimination law
Evidence that inputs and proxies were tested, that disparate outcomes were measured, and that the testing method is documented.
Actuarial standards of practice
Model documentation, data quality and reliance disclosures sufficient for the actuary to rely on the model in an opinion.
Adverse-action and claims-handling duties
A reason for each declination, rating uplift or claim denial that can be given to the policyholder in plain terms and traced to the model that produced it.
Third-party and rating-vendor oversight
Control evidence for external scores, data enrichment and vendor models where the insurer carries the obligation but not the code.
Each module is a vertical projection of an asset already running on the platform. No parallel store, no second evidence chain.
Underwriting Decision Vault
Unified Evidence Vault
Inputs, external scores, model version, output, referring underwriter and final disposition captured per decision, with an audit identifier that survives the policy lifecycle.
Permissible-Input Control
Control Library — AI controls
Tested prevention of prohibited or proxy variables entering a rating or underwriting model, with the block recorded and routed for review.
Fairness Testing Workbench
AI Model Inventory — fairness evidence
Repeatable subgroup and proxy testing over pricing, underwriting and claims outcomes, with method, population and result held as evidence.
Adverse-Action Explainability Log
Unified Audit Trail
The reason given to the policyholder, tied to the model version and the decision record that produced it.
Vendor Model Oversight
TPRM Workbench
Control testing over external scores and third-party models the insurer relies on but does not build.
A scored position across inventory, permissible inputs, fairness testing, explainability, human referral, vendor oversight, monitoring and reconstruction.
A separate measure of whether the AIS programme would withstand an examination, assessed against the evidence an examiner asks for rather than the policy that describes it.
Where declinations, uplifts and denials are being issued without a traceable, explainable reason — by product and distribution channel.
A scoped attestation for the board risk committee or state examiner, naming models in scope, period and stated limitations.
Eight questions across the dimensions a market-conduct examination would actually test. Answers stay in your browser — nothing is submitted, stored or transmitted.
01 · Inventory
Is every model influencing pricing, underwriting, fraud or claims recorded in a single inventory with an accountable owner?
02 · Permissible inputs
Is there a tested control preventing prohibited variables and known proxies entering a rating or underwriting model?
03 · Fairness testing
Is subgroup and proxy testing run on a defined cadence, with method and population documented?
04 · Explainability
Can every adverse decision be explained to the policyholder in plain terms, traced to the model version that produced it?
05 · Human referral
Is there a defined referral threshold where a named underwriter or adjuster must intervene, and is the intervention recorded?
06 · Vendor oversight
Do you hold control evidence for external scores and third-party models you rely on?
07 · Monitoring
Is post-deployment performance and outcome drift monitored on your own book rather than the vendor's validation set?
08 · Reconstruction
Could you reconstruct, on request, how a specific declination, rating uplift or claim denial was produced?
Answer the questions above to see an indicative position.
Days 1–15
Scope and obligation mapping
Products and models in scope, state footprint fixed, obligation set resolved into testable control statements.
Days 16–45
Inventory and input control
Model inventory reconciled; permissible-input and proxy controls tested against live rating and underwriting paths.
Days 46–75
Fairness and explainability testing
Subgroup testing run on the live book; adverse-action reasons traced end to end; vendor evidence gaps named.
Days 76–90
Position and attestation
Maturity and market-conduct readiness positions issued, gap remediation sequenced, attestation scope agreed.
Scope, residency and retention are set per engagement. Commercial terms are quoted against scope.
Start a scoping conversation