Some questions don’t need a screen; they need an answer. Which of my claims has the weakest carrier position? Where is a notice deadline close? Which carrier is slowest to respond? The InsightAdvocate copilot answers them from the record and shows what it read.
How it answers
sequenceDiagram
participant U as Advocate
participant C as Copilot
participant T as Read-only lookups
participant G as Figure check
U->>C: Which carrier is slowest to respond?
C->>T: Claims in my scope, SLA clocks, scorecards
T-->>C: Results, each a source
C->>G: Every figure in the draft answer
G-->>C: Unsupported figures replaced
C-->>U: Answer with numbered sources and suggested next steps
- Read-only. Its tools look things up — the claim, its policies and positions, clocks, the loss schedule, the decision record — and only for claims the asker may open.
- Checked. Every figure in its answer is checked against what the tools returned; a figure that isn’t there is replaced, never shown.
- No actions. It can suggest a next step as a link into the workspace. It never sends, accepts or changes anything.
- Model. Claude Sonnet 5.5 on Amazon Bedrock by default. With the model off it answers from the facts alone.
| Ask the copilot | Lookups it calls | What the answer covers |
|---|---|---|
| “Which claim has the weakest carrier position?” | Claims in scope; coverage assessments | The claim and the position, with why it reads weak or not in the policy |
| “Where is a notice deadline close?” | Notice rows across the book | Each policy, its deadline and days left |
| “Which carrier is slowest to respond?” | SLA clocks; carrier scorecards | Days to acknowledge, clocks met on time, breaches |
| “What is the floor on this claim and why?” | The settlement engine’s table | Accepted, disputed, weights, retention |
On home, “Ask my book” offers the starter questions above. We don’t show a copilot answer on this site: our screenshots were taken with the model switched off, and we would rather show nothing than a staged answer.
