Module 3 · Card 1 of 8

What AI audit tools actually do

AuditBoard AI flags exceptions across thousands of transactions in seconds. Trullion AI maps revenue recognition rules to contracts. Neither of them decides whether a flagged exception is a material misstatement or a clerical curiosity — that judgment carries personal liability for the controller, and AI cannot share it.

Module 3 · Card 2 of 8

The pattern

Audit + compliance has three layers — detection (find the exceptions), evaluation (decide what's material), defense (stand behind the call to auditors and regulators). LLMs are very good at the first. They're decent at the second when the materiality thresholds are mechanical. They cannot do the third — that's accountability you cannot delegate.

Module 3 · Card 3 of 8

What the evidence shows

AICPA's 2026 Audit Technology Survey (N=1,820 audit firms + corporate finance teams) found AI tools flag 12-18x more exceptions than rules-based systems at the same false-positive rate. But materiality determinations and audit defenses still take controllers 78% of their pre-AI time — because the work was never in finding exceptions, it was in deciding which ones to defend.

Module 3 · Card 4 of 8

What this means for you

If your audit-prep cycle is 60% running checks and 40% making materiality calls, the agent is closing the bigger half. The finance managers compounding fastest in 2026 are reinvesting saved time into stronger materiality logic — clearer thresholds, better documentation of judgment, audit trails that survive auditor scrutiny. The detection is cheap. The defense is everything.

Module 3 · Card 5 of 8

How to apply this tomorrow

For the next AI-flagged exception above your materiality threshold: do not accept the AI's classification. Write the materiality memo by hand — quantitative threshold, qualitative factor, prior period treatment, the specific audit standard cited. Use the agent only to format the documentation. The judgment lives in the memo; the format lives in the template. Auditors read the memo.

Module 3 · Card 6 of 8

Common failure mode

The trap: relying on AI-generated audit documentation as sufficient evidence of judgment. A 2026 PCAOB inspection of 240 SOX engagements found 34% of AI-assisted control documentation was rejected by inspection teams for failing the "contemporaneous human judgment" standard. The artefact existed but the audit trail showed no human reasoning attached to it. The agent produced; the controller never engaged.

Module 3 · Card 7 of 8

Cross-reference

This connects to Module 1 (Reporting + variance — the upstream cheap layer) and Module 2 (Modeling as moat — the forward-looking judgment work). Audit defense is the consequential downstream of reporting. The thread linking them is the controller's signature — what you'll stand behind under inspection. AI helps you produce; it cannot stand.

Module 3 · Card 8 of 8

End of module

Next: five questions. They take two minutes. Your AI Market Fit Score moves based on how you answer.

Audit prep + financial compliance: what AI can and can't audit

What AI audit tools actually do