The agentic insurance glossary
Definitions for every term used in agentic insurance distribution. Terms are drawn from the the standard, FCA guidance, and the UK insurance data infrastructure. No invented definitions. Maintained by Marrow on behalf of the A2A working group.
The standard
The core standards that define how AI agents transact regulated insurance.
The open standards framework defining how AI agents discover, compare, and transact regulated UK insurance safely and auditably. Stewarded by Marrow on behalf of the A2A working group.
The regulated intermediation layer between an AI agent and an insurer's systems. Enforces consent and disclosure rules, validates quote data against product rules, strips PII in flight, and writes the immutable audit ledger.
A validation component of the AI compliance rail that checks every AI-assembled field against insurer product rules before a premium is returned, blocking fabricated, estimated, or out-of-appetite values.
An insurer's pricing and underwriting systems exposed to AI agents via the Model Context Protocol, governed at the boundary by the A2A compliance rail.
Distribution and transactions
How agentic insurance works in practice, from intent through to bound policy.
Insurance policies quoted, compared, and bound end-to-end by AI agents acting on behalf of customers, governed by a regulated compliance layer and canonical data standards.
The use of AI agents as a regulated insurance distribution channel, enabling policies to be quoted, compared, and bound through conversational interfaces including ChatGPT, WhatsApp, and embedded widgets.
The end-to-end A2A transaction flow from risk-data assembly through validated quoting, mandatory disclosures, step-up confirmation, and policy bind - all within a single AI-mediated conversation.
The process by which an AI agent assembles an insurance risk profile through natural-language dialogue, using database enrichment to minimise questions and assume-infer-confirm to ensure material facts are attested before bind.
The mandatory human-in-the-loop moment in the A2A protocol at which a customer provides fresh, conscious confirmation before any contract-changing or payment action proceeds.
The distribution model in which insurance is offered within a non-insurance product or service at the moment of highest relevance. Agentic distribution is one form of embedded insurance.
Agentic commerce standards
External standards the A2A protocol is built on top of or aligned with.
The open protocol introduced by Anthropic in 2024 for connecting LLMs to external tools, data sources, and services. The technical substrate for insurer MCP endpoints.
The commerce standard developed by OpenAI and Stripe, defining how AI agents manage intent mandates, cart mandates, and payment mandates for autonomous commerce including insurance transactions.
The payment authentication standard developed by Google and the FIDO Alliance, defining how AI agents prove and scope payment authority. Interoperates with ACP and the A2A protocol's Tier B authority model.
UK insurance infrastructure
The data sources, registries, and enrichment systems used in UK insurance transactions.
The UK insurance industry's shared standards body for data, messaging, and product codes. The A2A protocol uses Polaris code lists as the canonical reference for product and risk classification.
Independent financial product research and ratings firm. The A2A protocol uses Defaqto's product taxonomy for structured product comparison and the distinctive-rule logic in quote presentation.
The DVLA-backed motor data sharing service providing driving licence validation and history to insurers. Used by the A2A Motor standard as a primary enrichment source in conversational underwriting.
The UK insurance industry's shared claims history database. Used by the A2A protocol for enrichment in motor, home, and commercial lines to reduce mis-declaration and improve risk assessment.
Motor Insurance Anti-Fraud and Theft Register. Used by the A2A Motor standard as an enrichment source for fraud and total-loss risk signals.
The DVLA-linked database of all insured vehicles in the UK. Updated within 24 hours of policy bind. The A2A Motor standard includes MID notification in the fulfilment stage of quote-to-bind.
The persistent national identifier for every addressable location in Great Britain. The A2A Home standard uses UPRN as the canonical key for property enrichment including BCIS rebuild cost and flood risk.
The RICS-managed building cost database used by the UK insurance industry for property rebuild cost estimates. The A2A Home standard uses BCIS to populate rebuild cost estimates keyed by UPRN.
The UK government-backed reinsurance scheme enabling insurers to offer affordable flood cover to high-risk properties. The A2A Home standard includes a Flood Re eligibility check in the enrichment stage.
Regulation and conduct
FCA rules, conduct obligations, and consumer protection standards that apply to agentic insurance.
The FCA's principal-based conduct standard (PS22/9), in force from July 2023, requiring financial services firms to deliver good outcomes across four outcome areas. Applies to agentic insurance distribution channels.
The standardised two-page product disclosure document required by FCA regulations before sale. The A2A protocol mandates IPID delivery and in-conversation acknowledgement at Stage 6 of the quote-to-bind flow.
The FCA requirement that from January 2026, travel insurance distributors signpost customers to the MoneyHelper medical directory where standard cover cannot be provided for a declared medical condition above the £200 threshold.
Concepts Marrow is naming
Terms coined or formalised by the A2A working group to describe patterns that did not previously have standard names.
Marrow's 0-100 index measuring how a UK insurer performs when consumers ask ChatGPT, Claude, Gemini, or Perplexity for insurance recommendations. Blends visibility, sentiment, accuracy, and recommendation rate.
The A2A Core Pattern governing how enriched data is treated before bind: an enriched value is presented to the customer as a fact to confirm, never assumed to be warranted until the customer has explicitly attested it.
The A2A Core Pattern requiring that every mandatory disclosure is surfaced and explicitly acknowledged within the AI conversation before any contract-changing action proceeds. Recorded in the audit ledger.
The A2A Core Pattern requiring that when a value cannot be sourced from an authoritative source, the agent must explicitly surface the absence rather than estimate, interpolate, or fabricate a replacement.
The A2A SMB commercial lines metric that determines whether enough is known about a business's risk footprint to support a bind action. Derived from Companies House, SIC codes, and declared trading activities.
The A2A classification of agent actions by risk level: Class 0 (Read), Class 1 (Quote), Class 2 (Transact - contract change or payment), Class 3 (Claim). Escalating proof requirements apply at each class.
The default A2A conformance tier: the agent prepares each Class 2 action and the customer provides fresh, human-present step-up confirmation before it proceeds. No pre-authorisation required.
The opt-in A2A conformance tier in which a customer pre-authorises an agent to act autonomously within defined, capped limits. Requires stronger agent identity verification and heavier conformance obligations than Tier A.
This glossary is maintained by Marrow on behalf of the A2A working group. Definitions are drawn from The Standard, FCA guidance, and published UK insurance infrastructure documentation. Corrections and additions: corrections are welcome at hello@onmarrow.com. Last updated 18 June 2026.