Guide
Plain-language definitions of AI and insurance terms — and how brokerages, carriers, and MGAs put each one to work.
16 terms
Standardized industry forms for insurance data.
A common set of forms used to exchange applicant and policy information across the industry.
In practice
Extraction and auto-fill turn ACORD forms into structured data agents can act on.
An AI system that takes actions across tools to finish a task.
Beyond answering a single prompt, an agent plans steps, calls tools or APIs, checks its work, and continues until a goal is met.
In practice
An agent can pull a loss run, fill an ACORD form, request missing info, and prepare a submission — keeping a human in the loop for approval.
Permission to commit a carrier to coverage.
Delegated power to put coverage in force on a carrier's behalf, within defined classes, limits, and rules.
In practice
AI workflows pre-check eligibility against the binding agreement, leaving the final bind to an authorized human.
A numeric representation of meaning used for search.
A vector that captures the meaning of text so similar items sit close together — the backbone of semantic search and RAG.
In practice
Powers 'find similar claims' or 'show comparable risks' across a book of business.
An amendment that changes an existing policy.
A document that adds, removes, or modifies coverage on a policy mid-term.
In practice
AI drafts endorsement requests from a plain-language description and flags coverage impacts.
Adapting a base model to your domain with examples.
Further training a general model on domain-specific examples so it adopts your tone, formats, and terminology.
In practice
An MGA might fine-tune on its own underwriting decisions to better match house appetite and style.
When an AI states something false but plausible-sounding.
A confident but incorrect output. Mitigated with retrieval (RAG), citations, constrained tools, and human review on high-stakes actions.
In practice
In regulated workflows, hallucination risk is why binding or filing steps keep a human approval and a full audit trail.
An AI model trained on text to understand and generate language.
A model trained on large amounts of text that predicts likely next words, letting it summarize, draft, classify, and answer questions in natural language.
In practice
Brokerages use LLMs to draft client emails and summarize policies; carriers use them to triage submissions; MGAs use them to standardize messy intake data.
Claims paid divided by premiums earned.
A core profitability metric — lower ratios mean a more profitable book, all else equal.
In practice
Analytics surface loss-ratio drivers so carriers and MGAs can adjust appetite and pricing.
A report of a client's historical claims.
A carrier-produced history of claims for an insured, used to assess risk and price renewals.
In practice
Extraction tools read loss runs and summarize claim frequency and severity for underwriters.
A firm with delegated authority to underwrite for carriers.
An intermediary granted authority by carriers to underwrite, bind, and sometimes handle claims within agreed guidelines.
In practice
MGAs use AI to scale underwriting decisions consistently while staying inside carrier-defined appetite and limits.
The instruction and context you give an AI model.
The text (instructions, examples, and data) provided to a model to steer its output. Better prompts produce more reliable, on-format results.
In practice
Teams standardize prompts so every producer gets consistent, compliant output from the same underlying model.
Grounding an AI's answers in your own trusted documents.
RAG retrieves relevant passages from a knowledge base and feeds them to the model so answers cite your real data instead of relying on memory.
In practice
Carriers ground answers in underwriting guidelines; brokerages answer coverage questions from the actual policy wording rather than guesses.
A request for quote sent from a broker to a carrier.
The package of applicant and risk information used by an underwriter to evaluate and price a risk.
In practice
Agents assemble complete submissions automatically, reducing back-and-forth and speeding up quotes.
Coverage placed outside the admitted market.
Insurance for risks standard (admitted) carriers won't write, subject to separate filing and tax rules.
In practice
Compliance tooling tracks state-by-state filing and tax requirements as rules change.
The unit of text models read and bill by.
A chunk of text (roughly ¾ of a word). Model context limits and pricing are measured in tokens.
In practice
Understanding tokens helps teams budget cost and decide how much policy text to feed a model at once.
