← All agents
Triggered when verification campaign scans arrive
Reconciliation Agent
Physical verification produces thousands of scans per campaign. The Reconciliation Agent matches every scan against the register, classifies the result — matched, moved, missing, or unregistered — and routes only genuine exceptions to a human.
What it does
- •Auto-matches scanned assets to register records with a confidence score
- •Flags ghost assets (on the books, not on the floor) and unrecorded assets (on the floor, not on the books)
- •Drafts custodian confirmation requests for disputed exceptions
- •Drafts journal adjustments for approved write-offs — never posts them itself
A typical scenario
A plant runs its annual CARO verification campaign and 4,200 scans come back overnight. Instead of a controller manually diffing a spreadsheet, the agent matches each scan to the register, clears 96% automatically, and surfaces 38 genuine exceptions — including 3 ghost assets still on the books three years after scrapping — with a confidence score and evidence trail on each one.
Target metrics
- Auto-classification accuracy
- >95%
- Exceptions auto-resolved without human touch
- 40-60%
- Exception triage turnaround
- <60s p95 per asset
Frequently asked questions
- Can the Reconciliation Agent post directly to my ledger or file externally?
- No. It follows AssetIQ's draft-don't-commit model — it can propose journal drafts, exceptions, or transfer orders with a confidence score and evidence trail, but a human with approval authority must review and post anything that reaches your ledger, the GST portal, or an e-way bill system.
- What happens if it isn't confident about a result?
- It attaches a confidence score to every proposal. Below a per-agent, per-materiality threshold, the item routes to a human for mandatory review, even if policy would otherwise allow it to auto-commit as an internal draft.
- Is every action it takes logged?
- Yes. Every run records its trigger, inputs, model version, every tool call and result, its confidence score, the human who approved it (if any), and the resulting records — feeding directly into AssetIQ's immutable, hash-chained audit trail.