AI in Auditing, What Is in Use Today

Where it genuinely helps

Document reading. Contracts, leases and agreements arrive as PDFs. Extracting terms, dates and amounts from a hundred lease agreements used to be a junior's week. It is now fast and the auditor checks the output rather than producing it.

Anomaly ranking. Full population testing produces thousands of exceptions, most of them innocent. Ranking them by how unusual they are against your own history puts the few that matter at the top of the list.

Matching across systems. Reconciling a purchase ledger to a bank feed to a delivery record, where names and references never quite agree.

Drafting. First drafts of routine sections, checked and rewritten by the person signing.

Each of these has one thing in common. A human verifies the output before it means anything.

Where it does not help yet

Judgment. Whether a receivable is recoverable depends on facts nobody wrote down. Whether the customer answers the phone. What the owner knows about their business.

Evidence. An audit conclusion needs evidence a third party can inspect. A model's output is not evidence of anything, and cannot be attached to a file as support.

Responsibility. The opinion is signed by a person who is accountable for it. That does not delegate.

The failure mode to know about

These systems produce confident, fluent, wrong answers. Confidently wrong is worse than obviously wrong, because it survives review.

An audit conclusion reached by a model and not independently verified is not a conclusion. It is a guess with good grammar. Any firm using these tools properly treats output as a starting point, never as support.

What to ask a firm that claims to use AI

Two questions, and the answers are revealing.

What specifically does it do in my audit? A real answer names a task. Reading my lease agreements. Ranking exceptions. A marketing answer says "AI-powered".

Who checks the output before it reaches your file? If the answer is unclear, the tool is a risk rather than a benefit.

What it means for your fee

Where automation removes hours, that should show in the price or in the depth of work. It usually shows in depth first.

If a firm has invested in these tools and the fee is unchanged and the work looks identical, ask what the investment bought you.

Where we fit

We use software-driven fieldwork and work from your full ledger rather than samples. Everything that reaches an audit conclusion is verified by the person signing it.