Halfway through a review, a deal broker has a question the screen does not answer in one look: which open losses are large, what the buyer has to do at closing, what to ask the seller for. The diligence copilot is the side panel on every engagement that answers those questions from the engagement itself, with numbered sources, and never takes an action.
How an answer is made
The eight tools
The copilot can only see what its tools return, and the tools only return the engagement in front of the asker, inside their scope: overview, program, claims, findings, gaps, Target 360, search over the data room, and the cost after closing.
| Ask the copilot | Lookups it calls | What the answer covers |
|---|---|---|
| “What are the large open losses?” | claims, findings | Claims over the large-loss threshold, paid and reserved, what happened, and what the record does not say |
| “What does the buyer need at closing?” | gaps, program | Tails and run-off, change-of-control clauses, subsidiaries to add, collateral to settle |
| “Which lines are on the seller's program?” | program | Each line's structure, carrier and premium, with the schedule page |
| “What will the program cost after closing?” | cost | Today against the standalone estimate and the one-time tails, with the assumptions |
| “What should we ask the seller for?” | gaps, findings, search | Missing documents and the open questions the agents raised |
| “Who leads the target?” | target | Leadership from InsightPulse, cited with its as-of date |
What it will not do
- Act. It suggests questions and links into the workspace; approving a finding, delivering a report or sharing with the deal stay buttons a person presses.
- Invent a figure. Answers are read to the precision each figure is written in, and a figure no tool returned is replaced with “not in the record”.
- Reach outside scope. A deal broker sees their own engagements; a practice leader or compliance sees the practice.
- Pretend. With no model configured it answers in facts-only mode from the same tools, and says so.
Saying what is missing is part of the answer
In the screen above, on an illustrative industrial target, the copilot reports one large open general liability loss and then says what the record does not hold: whether the reserve is adequate, whether the claim is in suit, and which program responds. Those are the questions to take to the seller, and the copilot offers them as follow-ups.
Questions we get asked
Which model does it use?
Claude on Bedrock in the practice’s own AWS account. Prompts are versioned and every answer is traced, like the agents.
Is the conversation kept?
Each answer is traced for audit with its tools and sources. The copilot keeps no memory beyond the conversation on screen.
Keep reading
The program after closing: what the buyer will really pay
The seller's master program ends at closing. The target's premium does not stay the same.
An industrial target on its parent's program
Four lines on the seller's master program, a TCE plume and a product suit.