Audit 2.0: Efficiency, AI, and Human Judgment

The audit profession has been waiting for a meaningful redesign for years.

Firms talk about efficiency constantly, but the day-to-day reality for many auditors is still a grind of chasing support, matching evidence, updating trackers, reconciling disconnected systems, and writing the same testing language again and again.

That is not a people problem. Auditors are not short on ambition, judgment, or work ethic.

It is a workflow problem.

This post expands on my Miles Masterclass discussion, Audit 2.0: Efficiency, AI, and Human Judgment. That conversation was organized as a CPE course for accounting professionals and covered how AI changes audit evidence, workflow design, client collaboration, review, and professional judgment.

The core point is simple: audit teams should not use AI to make weak work look faster. They should use AI to remove the repetitive coordination work that keeps auditors from spending enough time on risk, exceptions, and conclusions.

The limits of traditional audit efficiency

Most audit efficiency efforts have focused on doing the same work faster:

• More templates
• More offshore capacity
• More workflow checklists
• More status meetings
• More portals for clients to upload files

Those changes can help, but they do not fundamentally redesign the work. They still leave auditors responsible for coordinating fragmented evidence, interpreting messy support, and manually connecting documents to procedures.

The result is predictable. Staff spend too much time on mechanical execution. Seniors spend too much time reviewing whether the mechanical work was done correctly. Managers and partners get pulled into status management instead of judgment-heavy questions.

Audit 2.0 should not mean adding another layer of software on top of the same process. It should mean changing which parts of the process require human effort in the first place.

Where AI belongs in audit

AI is useful in audit when it removes repetitive execution work while keeping auditors in control of risk, materiality, and conclusions. That is why AI audit automation should increase professional judgment, not replace it.

That means AI should help with work like:

• Requesting and organizing client evidence
• Matching uploads to specific selections
• Reading documents and locating relevant support
• Tracing values across reports, contracts, invoices, and schedules
• Identifying missing or inconsistent evidence
• Drafting first-pass procedure language
• Summarizing exceptions for auditor review

These are not replacements for professional judgment. They are the work around professional judgment.

When AI handles that surrounding execution, auditors can spend more time asking the questions that matter: Is the evidence sufficient? Does the exception change the conclusion? Is the risk response appropriate? Does the explanation make sense in context?

Why human judgment becomes more important

The better AI gets at execution, the more important auditor judgment becomes.

That may sound counterintuitive, but it is exactly what happens when routine work becomes faster. The bottleneck moves from finding information to evaluating it. The auditor's role shifts from manually assembling the workpaper to supervising the agent, reviewing the evidence, and deciding what the result means.

That is a better use of auditor talent.

It is also a better learning path. Staff and seniors should not spend years proving they can copy values from PDFs into workpapers. They should learn how to evaluate evidence, investigate exceptions, understand risks, and defend conclusions.

AI can accelerate that development if the system is built for reviewability.

Traceability is non-negotiable

Audit AI cannot be a black box.

Every answer needs a source. Every summary needs a citation. Every exception needs a path back to the document, page, table, or field that produced it.

Without traceability, AI creates review risk. With traceability, AI becomes a work accelerator because auditors can inspect the evidence directly instead of restarting from scratch.

This is why Punchcard is built around source-backed audit work. The agent can perform the first pass, but the auditor remains responsible for the conclusion. The software should make that conclusion easier to reach and easier to support.

What Audit 2.0 should feel like

Audit 2.0 should feel less like managing a file chase and more like supervising a well-organized audit assistant.

Clients should know exactly what to upload and where it belongs. Auditors should see evidence mapped to the right request, the right sample, and the right procedure. Exceptions should surface early. Draft work should be ready for review. The audit trail should be clear.

The goal is not to make audits less rigorous. The goal is to remove the administrative drag that keeps auditors from applying rigor where it matters most.

That is the future we are building at Punchcard: AI-powered audit automation that increases efficiency by increasing the share of time auditors spend on professional judgment.

Excerpts From The Conversation

These are the ideas from the Miles Masterclass discussion that are most important for audit leaders thinking about AI.

Audit was a technology problem long before AI became the headline.

Audit teams have been forced to treat evidence as if it were still paper. Client support arrives as PDFs, exports, screenshots, schedules, contracts, and emails. The work is not just reading those files. The work is turning that evidence into structured audit context: what selection it supports, what assertion it addresses, what procedure it relates to, and what exception it may create.

That is why generic AI chat is not enough. Audit needs systems that understand the workflow around the evidence, not just the text inside the evidence.

AI cannot sign the audit opinion, but it can change the work required to get there.

The auditor still owns the conclusion. AI should not decide materiality, override professional skepticism, or replace the final judgment about whether evidence is sufficient and appropriate.

What AI can do is prepare more of the work for review. It can match support to selections, summarize relevant clauses, compare values across documents, identify missing evidence, and explain where an exception came from. That changes the auditor's job from manual assembly to directed review.

The best audit AI brings its own receipts.

A useful audit system cannot stop at an answer. It has to show the evidence behind the answer.

If an AI agent says a grant restriction exists, the auditor should be able to see the exact source. If it says an invoice amount agrees to a schedule, the auditor should be able to inspect the path between the two. If it flags an exception, the reviewer should understand why the exception exists without starting the work over.

Traceability is what separates audit automation from audit theater.

The firms that wait are already making a technology decision.

Not adopting AI is still a choice. It means continuing to ask staff to spend large parts of their week on file chasing, evidence matching, status updates, and repetitive documentation.

Firms do not need to replace their audit methodology overnight. They do need to start learning where AI belongs in the process, which controls are required, and how review should work when an agent prepares the first pass.

The firms that learn those patterns now will have a practical advantage when AI-assisted audit work becomes normal.

The Miles Masterclass Discussion

The Miles Masterclass version of this conversation breaks the topic into six parts:

• Auditing the auditors
• Audit was a tech problem all along
• AI cannot sign an audit
• The audit that brings its own receipts
• The audit firms that wait are already behind
• Audit's last human mile

Those chapters are useful because they frame AI in audit as a workflow redesign problem, not a chatbot problem. The conversation covers why audit evidence is still handled manually, how AI can turn unstructured support into reviewable work, and why the final conclusion still belongs to the auditor.

For readers who want the course context, the Miles page lists the discussion as a 2 CPE credit auditing course created and updated on April 30, 2026. You can view the course page here: Audit 2.0: Efficiency, AI, and Human Judgment.