The question every claims executive asks about agents is not “how smart are they?” It is “what are they allowed to do without us?” InsightCW answers it with a dial per agent, per line, per action — and ships with the dial turned to “ask first” for everything.

Four levels

LevelNameWhat it means
L1Draft onlyThe agent drafts; nothing is put in front of anyone as a decision
L2Ask firstThe agent prepares the decision; a person answers it
L3Act and tellThe agent acts, tells the owner, and the action can be pulled back for a window
L4ActThe agent acts

An action's effective level is the lowest of three: the level set for that agent, line and action; the action's hard cap; and the tenant ceiling. The ceiling ships at L2, so out of the box no claims agent acts on its own. The kill switch drops an agent to L1 at once.

Six actions a person always takes

Some actions should carry a person's name whatever the evidence says about an agent's accuracy. These are capped at L2 regardless of the dial: filing an SIU referral, deciding an approval, opening a subrogation opportunity, a defense review, a medical review, and closing a claim.

Prepared decisions: Accept, Change, Reject, Ask

flowchart TB A["Agent prepares:
recommendation, reasons,
facts, draft"] --> B{"Person answers"} B -->|"Accept"| C["Carried out as
the person"] B -->|"Change"| D["Carried out with
the person's edits"] B -->|"Reject"| E["Nothing happens"] B -->|"Ask"| F["Answered from
the facts"] C --> G["Feedback recorded"] D --> G E --> G style B fill:#e8f8f4,stroke:#22c1a4,stroke-width:2px

Accept, Change and Reject carry the decision out as the person and record feedback on the agent. Ask is answered from the facts and records nothing. Carrying a decision out goes through exactly the same code as a person doing it by hand, so an agent can never do something a person couldn't.

Decisions prepared by the agents with Accept, Change, Reject and Ask, on demo data
Prepared for you, on a general liability claim (demo data): the intake agent's information request, the coverage agent's position read against CG 00 01, and the reserving agent's change from $69,500 to $142,500 — each with its reasons, the facts behind it, and Accept, Change, Reject or Ask.

Evidence earns a notch; poor answers lose one

An hourly job rolls up 30 days of answers per agent, line and action. Agreement is accepted divided by accepted plus changed plus rejected. At 90% or more over at least 50 answers, it recommends a promotion, which a manager accepts or not. Below 70% over at least 20 answers, it demotes at once, audits the change and tells the claims managers.

The autonomy page with the tenant ceiling and each agent's level per action, on demo data
Autonomy on demo data: the tenant ceiling at L2 (ask first), agreement and overrides over 30 days, and each agent's level per action and line.

Pull-back at L3

When a carrier raises the ceiling and an agent acts at L3, some actions open a pull-back window. A payment request can be voided; a letter is held and sent only when the window closes. The owner sees active windows on the claim and in their own list. If nobody pulls it back, the action stands.

Where most carriers should start

At the shipped ceiling. Run the agents at L2 for a quarter, read the agreement numbers by line, and raise the ceiling only for actions where the evidence is strong and pull-back exists. See the first ninety days.

Keep reading