What your examiner
asks for.
Six requests an exam or a coverholder audit actually makes. Three of them return from one query. Two are documents you write, and they are better for being written against a real population. One is not ours.
Every exam asks the same three questions.
Who changed it. Who reviewed it. Who approved it.
“A qualified person wrote it” stops being true the day an agent writes the code. Haltere writes the answer while the work happens.
Four regimes. One record.
Every step the AI took,
and where it stopped.
One rating change that shipped, and one that never got near a reviewer.
Fixture data in both; fields the spec marks proposed (the change number, the exhibit pack) appear as they will. A value outside the filed band fails at the gate, and the refusal is a row of its own. Download what your examiner would receive (PDF)
Carriers, and the vendors
who serve fifty at a time.
Standing up a new program or product.
One backlog item, built governed, with the evidence pack that shows the board AI progress and the examiner the receipts.
Your AI hub decides. Your customers answer for it.
Haltere sits under your delivery, so every customer build ships with its receipts and the third-party question is answered before it is asked.
Your first governed build.
Four to eight weeks.
The governed first build.
Intake, the rater, the bordereaux, a portal, a report.
A weekly review, on your calendar.
Live in your environment, with the pack mapped to the Model Audit Rule, the DOI exam, and your carrier’s audit.
Afterward, governed operations: we run the line, or your team runs it and we train your operator. You own the software, the map, and the receipts. Walk at renewal with everything.
What the record returns,
and what it doesn’t.
The lines an exam or a coverholder audit actually asks for, and which of them stop being a project.
Change management: evidence for the key control.
Receipts for the period: who asked, each check with the policy revision it ran against, the approver and the reason, frozen to the commit that shipped.
Every change made under the superseded rule.
Each check names the policy revision it ran against, so which changes ran under which rule is a lookup. Re-running today’s rule against yesterday’s changes is not something the line does, and we will say that on the call too.
Who and what had authority to make it.
Agents are principals: the row carries the grant and the named person it acted on behalf of.
Your AI governance policy and the model-risk narrative.
The record tells you what actually ran, when, and under whose authority. The policy and the narrative are yours, and they are better for being written against a real population.
The vendor due-diligence questionnaire.
We answer it, in writing, with the unbuilt items named. You sign it.
Underwriting file review, and rate adequacy.
Those stay a human read. We don’t pretend otherwise.
Three of these the record returns. Two are documents you write. One is not ours. Nothing here is legal, actuarial, or regulatory advice, No regulator or examiner has reviewed this exhibit set. These exhibits are what the record returns; the review they are built for is your examiner’s. The NAIC AI supplement, in plain words
Asked on every first call.
We’re not doing AI. Compliance won’t let us.
Good. This is not an AI purchase; it is the backlog item you already have to build, delivered so the change record writes itself. Put your compliance lead on the first call.
Who approves changes?
A named person on your team, every time. Agents propose; nothing merges without policy checks and that approval, and the receipt records both.
What does the examiner get?
The change record with human names on it, the filed rate every rating change was checked against, and the decision record for what the software did. Every exhibit is drawn from receipts the system already emits.
Where does it run?
Inside your boundary: self-hosted or your private cloud, air-gap on request. Policyholder data never leaves. Your identity provider plugs in from day one.
What does it cost?
A fixed fee for the first build, quoted in the first call and agreed before anything starts. Afterward, governed operations: we run the line, or your team runs it and we train your operator. You own the software, the map, and the receipts either way.
Start with the workflow you already have to build.
Thirty minutes, with the engineers who build the line. You name the workflow and the examiner you answer to; we tell you what a governed build of it looks like, what it costs, and when it can start. Bring your compliance lead.
Inquiries are answered within one business day, by someone who can answer them.
