The decision
When a claim arrives, does it fast-track, go to standard handling, or get referred?
Insurance operations
Triage every claim the same way, and be able to prove it
Route each claim to fast-track, standard handling, or referral the moment it arrives, using rules your team can read. Every decision keeps the trace that produced it, so a regulator, a reinsurer, or your own QA team can see exactly why a claim went where it went.
How it runs on AI Rule Engine
The claim arrives
Your claims system posts the claim to a single endpoint and waits for the answer, or hands it off to run in the background. Same rules either way.
Deterministic rules run first
Coverage in force, claim type, reserve amount, deductible, prior claim count, and jurisdiction are checked by rules you wrote and can read. No model is consulted to establish a fact the data already contains.
AI reads what only text can tell you
Loss descriptions, adjuster notes, and attached documents are classified and summarized. Those outputs become facts on the claim, and rules decide what they mean.
Rules cascade to a decision
Conclusions feed other rules until the routing decision settles. If a fact later changes, conclusions that depended on it withdraw themselves rather than going stale.
The decision is returned and recorded
Fast-track, standard queue, or referral, with the reason attached. Anything outside the rules is referred rather than guessed.
What the auditor sees
Most automation records that a decision happened. This records why it happened, on logic you can prove was in force at the time.
The inference trace
Every rule that fired, in what order, and the actual values each condition matched on. Not a log line saying the claim was fast-tracked - the reasoning that got there.
The exact ruleset version
Decisions are pinned to the published version that produced them. A claim decided in March is explained by March logic, not by what the rules say today.
Who approved the logic
Rule changes move through draft, review, and publish. The approver is recorded on the release, and a release can be blocked until its test cases pass.
A signed release record
Each publish produces an evidence record signed with RSA-SHA256, verifiable against a public key. It proves the logic in force on a given date has not been altered since.
An append-only audit trail
Who changed what, when, and from which address, in a container that is written to and never edited. Exportable, and streamable to your SIEM.
Where the human stays
Referral is a first-class outcome, not a failure. Any rule can pause the run and put the claim in front of a person, and the reviewer sees the trace that led there before they decide. Reviewers sign in, so the run resumes on their answer with their identity recorded on the decision alongside the rules that asked for it. A forwarded link does not let someone outside the environment answer on their behalf.
Where AI is used, and where it is not
The boundary is enforced by how the workflow is built, not by a policy document. AI turns text into facts. Rules you wrote decide what those facts mean.
AI is asked to
- Classifying a free-text loss description into a claim type
- Summarizing a long adjuster note or an attached report
- Extracting fields from a submitted document
- Flagging language that suggests injury, litigation, or a coverage question
AI is never asked to
- Deciding whether coverage was in force
- Applying a reserve threshold or a deductible
- Choosing the routing outcome
- Anything the claim data already states as fact
What changes
Consistent routing
The same claim gets the same answer on a Tuesday morning and a Friday night, because the rule is the process rather than a description of it.
Faster straight-through handling
Clean claims that meet every fast-track condition stop waiting in a queue behind claims that need a person.
A defensible file
QA, internal audit, and market conduct examiners read the decision off the record instead of asking an adjuster to remember.
Rule changes in hours
A threshold moves in the editor, gets tested against saved cases, and publishes behind an approval. No release train.
The objections you are about to raise
“We already have a claims system. Why add another?”
This is not a claims system and does not try to be. It answers one question your claims system asks it, over an API call, and hands back a decision plus the reasoning. The claim stays where it lives.
“Our routing rules are already in a stored procedure.”
Then they are correct and invisible. Nobody outside engineering can read them, changing one is a deployment, and the procedure keeps no record of why a specific claim went where it went. Moving that logic here keeps the determinism and adds the authoring, the tests, and the trail.
“We cannot let AI decide claims.”
Agreed, and it does not. AI reads text and returns facts. Rules you wrote make every decision. You can point at the rule that produced any outcome, and the boundary is enforced by the design rather than by a policy document.
“You are a small vendor. What happens if you disappear?”
A fair question and we will not pretend otherwise. On a dedicated deployment the platform runs inside your own Azure subscription, so the data, the execution history, and the rules stay in infrastructure you control and can export in full at any time.
Frequently asked questions
How fast is a triage decision?
The synchronous endpoint returns the decision inside your own request, with a default 30 second budget you can lower per call. It is a real run, so it produces a run record and a trace rather than a bare answer.
Can we reproduce a decision from six months ago?
Yes. The run is stored with the ruleset version that produced it, so the trace can be read back and the same inputs re-run against the same published logic.
Can business analysts maintain the rules, or does this need developers?
Rules are authored in a visual editor and as decision tables, so an analyst who understands the policy can maintain them. Verification catches rules that can never fire, conflicting writes, and gaps in coverage before anything is published.
Does our claim data leave our environment?
On a dedicated deployment, no. The platform runs in your own Azure subscription and, paired with Azure OpenAI, model calls stay inside your boundary too.
Related decisions
Bring us one decision you make by hand
Pick the decision that costs your team the most time and the most argument. In one working session we will model it, run it against your own examples, and show you the trace it produces. If it is not a fit, we will say so.