Insurance
Which rule, on which facts, under whose authority.
That is what an examiner asks, and it is three questions, not one. Most claims systems can answer the middle one and reconstruct the other two afterwards. Archiveris produces all three as the decision is made.
The obligation
You are already required to explain.
None of the following asks whether a decision was good. Each one asks whether you can show how it was reached, on demand, long after the adjuster has moved on.
| Where it comes from | What it requires you to produce |
|---|---|
| Unfair claims settlement practices laws | Adopted in most states from the NAIC model. A prompt and reasonable investigation, and a written explanation of the basis for a denial or a compromise. |
| The ERISA claims procedure rule | For employer health and disability plans, 29 CFR 2560.503-1 requires an adverse determination notice to give the specific reasons and to reference the specific plan provisions it rests on. |
| Market conduct examinations | A state department of insurance takes a sample of files and asks you to account for the handling, after the fact, across a period you did not choose. |
| The NAIC AI model bulletin | Adopted by the NAIC in December 2023 and taken up by around half the states. A written program governing the AI systems that affect consumer outcomes, with documentation and testing to match. |
The last one is the new pressure, and it is not a prohibition. It says that if a model touches a consumer outcome you must be able to describe it, govern it and evidence it. The cheapest way to satisfy that is for the model not to be in the decision at all.
Why it is expensive
The obligation is discharged by archaeology.
The policy wording was written in a document. Somebody translated it into code. The code was changed eleven times and the document was not. So when the examiner arrives, nobody can say what the rule was on the date of loss without reading a commit history, and nobody can say which facts the adjuster actually had.
Appeals work the same way. A denial is rarely overturned because the rule was wrong. It is overturned because a fact was wrong, or stale, or came from a system nobody checked — and the file cannot show which.
Three questions, three answers
One for each part of the question.
This is the whole argument for composing three products rather than buying a rules engine. A decision engine answers the middle question well and leaves the other two to you.
Where it applies
Three places it earns its keep first.
Each is a published Business Operation your existing claims and policy systems call. None of them is a replacement for those systems.
Coverage, limits and deductible checked against the wording in force, with every approval or denial carrying the clause that produced it.
See the worked claim Underwriting triageReferral thresholds, exclusions and risk tiers applied the same way by every channel, with the reason for each referral recorded rather than inferred.
Kebiki Rate and eligibility rulesState-specific rules and filed rates held as versioned, tested assets, so what you executed matches what you filed and you can show it.
Risk and compliance Agents in the claims workflowLet an agent gather the file, draft the correspondence and call the decision. Do not let it decide coverage, and do not let it act beyond what the adjuster could.
Trustworthy AI
Bring the denial you would least like to explain.
Pick one determination a regulator or an appeal might come back to, and the systems it touched. We will show you what it looks like composed, governed and reproducible three years later.