Home / Carrier brief
Written for your carrier's reviewer.
If a capacity partner has asked your MGA what your systems do with AI, who approves what, and where the data lives, forward this page. It answers those questions in one place, in plain language, with nothing behind a form.
Everything on this page is also stated on our public security and AI pages, nothing here is written only for reviewers.
A person approves anything that touches rating or binding.
This is the whole posture in one sentence. Our assists read, extract, and suggest. They do not rate, bind, or file on their own, and every suggestion is recorded with the reasoning behind it, so the decision path can be reconstructed after the fact.
What those assists take on is the mundane, redundant work that repeats on every submission: reading the file, chasing what is missing, proposing Classifications and data, assembling the quote file. That work runs as an AI Operational Workforce with feedback loops that correct it as it goes, while the authority to rate or bind stays with a named underwriter.
Intake
The AI reads the submission and extracts what matters.
Triage
Classifications and data proposed, ranked, reasoned, and logged.
Human approves
A named underwriter decides. Nothing proceeds without them.
Record
The full trail (suggestion, decision, outcome) kept for audit.
The approval step is structural, not a setting. It is not a toggle an administrator can switch off to go faster.
What each assist may do, and where it stops.
We name our systems so you can ask about them specifically. During the 30-day production trial you operate them yourself, on your own submissions.
You do not have to take our description of any of this: we stand up a working system and you operate it in production for 30 days before a monthly rate begins. iO reached v1.0 in production in August 2026; ask the MGA that forwarded you this page for a walkthrough on their own file.
Built for the look-through you already perform.
The NAIC model bulletin on insurers' use of artificial intelligence was adopted in December 2023 and has since been adopted by 24 states and the District of Columbia, with further states issuing their own AI-specific insurance guidance (NAIC adoption map, status as of 7 July 2026). Its expectation is that an insurer governs its use of AI, and that expectation does not stop at your front door.
Documented AI use
Which systems are in use, what each is permitted to do, and which model provider sits underneath, inventoried and disclosed per engagement.
Human-in-the-loop controls
A named person approves anything touching rating or binding. The control is in the workflow, not in a policy document describing the workflow.
Reviewable logs
Suggestion, source field, approver, and outcome are recorded, so a file can be reconstructed months later without reconstructing anyone's memory.
Deeper reading on what to require of any AI touching underwriting (including ours) is in our AI underwriting guardrails guide.
The MGA's data belongs to the MGA.
Their data stays theirs, exportable, and is not used to train models for anyone else. If the relationship ends, the data leaves with them. The full statement (including hosting, backup, and access control) is on the security page, written in plain text rather than as a badge.
We hold a SOC 1 Type 1 report, an independent auditor’s opinion on the design of our controls at a point in time. It is not a Type 2, which tests whether those controls operated effectively over a period, and we do not describe it as one. Beyond that report we do not display compliance badges we have not earned. Our security page states exactly where we stand
Three things any reviewer can request.
Written security practices
The controls, hosting, backup and access model as a document your file can hold, not a marketing page.
Architecture walkthrough
A live session with the people who built it, not an account manager reading from a deck.
The AI governance package
Model providers in use, approval points, logging scope, and the policy that ships with the engagement.
Reviewing the reporting itself rather than the AI? The formats, schedules and controls behind bordereaux delivery are on the carrier reporting page, and the governance posture in full is on governed AI.
Send the hard questions first.
If something here does not satisfy your review, we would rather hear it now than after an MGA has committed. Ask us directly, the people who built the systems answer.
No form stands between a reviewer and any answer on this page.