Question 1

What LLM does the investigation agent use?

Accepted Answer

The agent uses advanced AI models through a structured tool-calling loop, not free-form text generation. It receives accounting context, then iteratively calls 7 data tools to gather evidence before producing structured findings. Every tool call and result are logged for full auditability.

Question 2

How much does a single investigation cost?

Accepted Answer

A typical investigation costs $0.05-$0.15 in AI API fees. Simple timing differences resolve in 2-3 tool calls. Complex multi-entity accounts may require 8-12 tool calls. GREEN accounts skip investigation entirely and use template-based work papers at zero cost.

Question 3

Can the agent create reconciling items or change data?

Accepted Answer

The agent can create reconciling items as drafts, flagged as AI-generated for human review. It cannot approve reconciliations, post journal entries, or modify GL data. SOX boundaries are enforced at the code level.

Question 4

What are the 5 finding types?

Accepted Answer

TIMING: a variance that will resolve in a future period. ERROR: a booking mistake that needs correction. PERMANENT: a variance requiring a journal entry. IC: an intercompany difference needing counterpart resolution. INFO: contextual information requiring no action.

Question 5

How long does a typical investigation take?

Accepted Answer

About 33 seconds, querying an average of 47 GL transactions across 3 prior periods. Complex accounts may take up to 90 seconds. The agent is rate-limited to prevent runaway investigations.

7 data tools. 33 seconds. A completed investigation.

FAQ

See Cortex investigate your accounts