August 6, 2026

Audit workpaper review changes when output already traces to source

Amanda Waldmann

Increasing trust with AI for audit and advisory firms.

Key Insights

Source-traced output shows provenance on open, shifting review from reconstructing a number's origin to evaluating evidence and conclusions.
With provenance visible, review notes skip basic source questions and focus on sample representativeness and the judgment calls PCAOB inspectors scrutinize most.
Reviewers can rerun flagged rows instead of reworking a whole workpaper, while the review record still captures every check and sign-off.

A senior spends the first hour of review reconstructing provenance rather than judging conclusions. The reconstruction includes tying the AR confirmation back to the trial balance, matching the population to the aging, and walking each tickmark to its support. That reconstruction repeats on the next file, then the next, often across three or four engagements running in parallel through busy season. Audit workpaper review looks different when the output arrives already linked to the evidence behind it. This article covers what changes when a workflow step runs with a built-in Trace, how review notes shift once provenance is no longer the reviewer's first job, and what a review-ready workpaper needs to hold up under later inspection.

What a reviewer actually does at the desk today

Most seniors have run some version of this loop for years. A staff member pulls the client's PBC file, prepares the schedule, ties it to source, adds tickmarks, and signs off. The reviewer picks up the file, checks the tickmarks against the underlying evidence, and either clears it or writes a note back to the preparer: "reperform the aging cutoff test using the 12/31 detail, not the 12/28 export." The preparer chases the updated PBC, reworks the schedule, and returns it. The cycle repeats until the workpaper clears.

A review trail can be as basic as initialing each workpaper in a prescribed place, a practice that predates most current workpaper software. Sign-offs accumulate into an issues memo for the partner. Much of the effort in that cycle isn't judgment; it's reconstruction. Before a reviewer can decide whether a conclusion is right, they first have to work out the number's source and population, then check whether the evidence actually supports the tickmark.

The stakes on that reconstruction are higher than they look. PCAOB's AS 1215 sets that bar in its experienced auditor standard: someone outside the engagement should be able to follow the procedures and evidence without the preparer walking them through it. A partner preparing a summary memo for the audit committee shouldn't need to Slack the senior to explain where a number came from.

Inspectors keep finding trouble in that same area. PCAOB's 2024 inspection spotlight flagged repeated failures to support the data, reports, and populations feeding substantive procedures, including a case where Marcum never compared its inventory listing to the recorded balance. The same provenance gaps that drain review time show up later as inspection findings.

Where this fits in the audit operating model

That reconstruction cost is a big part of what's driving change in how engagement work gets done. Engagement work is shifting toward a model where practitioners and Field Agents work every engagement together: agents execute the procedural steps, practitioners review the results and apply judgment. Fieldguide is one engagement platform built around that model. Its Field Agents execute steps like evidence review and test procedures ahead of practitioner review, and each run produces a Trace: a record of the inputs used and the steps taken to reach the output, with citations to source documents where the surface supports them.

A Trace is not a summary of what an agent did. It's the same kind of evidence trail a reviewer would otherwise have to reconstruct by hand, like the tie-out of a lease schedule to its signed contracts. It arrives attached to the output instead of left for someone to rebuild later.

What changes when a Field Agent executes the step

When a Field Agent executes the applicable workflow step, the reviewer meets the evidence before the conclusion. Where citations are supported, clickable references take the reviewer straight to the highlighted section of the source document: the specific clause in a debt agreement, the row in the payroll register, or the paragraph in a board minute.

Earlier PDF-linked workbooks used the same logic on a smaller scale. One firm linked data from separate PDFs into a single workbook so that clicking a cell took the reviewer to the underlying source document. That linked workbook reduced testing and training time and improved quality. Agent output extends that same logic across the whole workpaper instead of the one link a team could build by hand.

For the engagement-team reviewer at the desk, this reorders the day. Fieldguide's Preparer Review Experience gives preparers a dedicated workspace to review and append to agent output before it moves up, and the workpaper captures the exchange and the evaluation that followed. Review time shifts to inspecting the cited source and evaluating whether the assumptions support the conclusion. That might mean checking whether the sample of expense transactions captures the higher-risk period around year-end, instead of just confirming that the numbers foot. This is desk-level review, distinct from the engagement quality review that evaluates significant judgments and approves issuance.

How review notes change shape

Review notes get more specific once the source is already linked. The old note was a request to show the source and tie the population to it, like "confirm this AR aging reconciles to the GL and attach the reconciliation." That kind of back-and-forth exists only because the reviewer couldn't reconstruct the number from what was in front of them.

With the source already linked, the remaining notes address harder questions: whether the sample of 40 revenue transactions is representative across product lines and geographies, or whether management's estimate of allowance for credit losses holds up given a specific customer's recent aging deterioration. These are the notes PCAOB inspectors treat as the real test. RSM's revenue deficiencies in the 2024 cycle related primarily to testing the significant assumptions behind estimated costs to complete contracts. Review attention belongs on that kind of judgment question.

Resolving a note changes shape too. When a reviewer spots that three sampled invoices in a revenue test came from the same customer, they can flag it as feedback, specify that the resample should stratify by customer, and rerun just those rows against the updated criteria, instead of sending the whole workpaper back for a full rework cycle. Faster resolution is the visible benefit. The more durable one is that the review conversation stays on the evidence and conclusion the file has to support.

Keeping professional skepticism sharp when the trace is clear

A well-cited workpaper makes the review process faster, but it doesn't change what the review is for. File defensibility still rests on the reviewer's evidence checks and the record they leave behind, and reviewers should approach a traced output with the same skepticism they would bring to any other preparer's work. It's the same instinct that says the cash confirmation came back clean, but the bank statement still needs a look.

That instinct cuts both ways, and both versions show up at the same desk. One reviewer sees a Trace-backed conclusion, sees the citation already sitting there, and is tempted to sign off without opening the underlying document.

Auditors can extend less skepticism to work produced by an automated tool than to the same work produced by a person, a pattern significant enough that AICPA leadership has spoken to it: a machine drafting the first pass doesn't excuse subordinating judgment or skepticism to it. The opposite reaction shows up too, sometimes on the same team. A reviewer manually re-derives a number the Trace already shows, just because an agent produced it, treating agent output as inherently less reliable than a preparer's regardless of what the evidence actually shows.

The practical answer is the same either way: evidence gets tested consistently regardless of who or what produced it. That discipline doesn't disappear because the source is easier to reach. A visible Trace shortens the path to the evidence; it doesn't change the reviewer's obligation to walk it.

What a review-ready agent workpaper contains

A good Trace solves a downstream problem, too: the documentation that makes desk review fast is the same documentation an inspector will want later. PCAOB's AS 1215 asks for enough detail that someone unfamiliar with the engagement can understand the workpaper's purpose, source, and conclusions on their own. A review-ready agent workpaper is built to meet that test on arrival rather than during the assembly period.

Four items an inspector or reviewer will look for:

Source links to original evidence. Draft outputs should include source links or citations where the surface supports them, so provenance is verifiable without the preparer's explanation.
The reasoning and steps. The reviewer can trace the path from input to output without asking the preparer to walk them through it.
Exception flags. The workpaper should surface exceptions and elevated-risk gaps, like a confirmation returned with a discrepancy or a JE that posted outside normal hours, so reviewer attention goes where the testing and documentation risk actually lives.
The review record. The file has to show who reviewed the output and which checks they performed, because the evidence obtained and the conclusion reached need their own documentation.

The review record matters most when someone reopens the file months later, whether that's a peer reviewer or the same partner drafting the next year's risk assessment. Workpapers tell the story of how the engagement team planned and performed the work and reached its conclusion, and that story needs to hold up on its own, the same way claim defense hinges on documentation years after the fact. A preparer walking someone through it verbally isn't a substitute. The review record needs to capture the source checks and exception resolution that support the conclusion, recorded while the work is happening rather than pieced together during the assembly period.

Where Fieldguide fits

The shift from reconstructing provenance to judging conclusions only works if the agent that executed the workflow step built a Trace a reviewer can actually stand behind.

Fieldguide is an end-to-end AI-native platform purpose-built for audit and advisory, built around an operating model where Field Agents execute engagement work, and practitioners retain review and approval. It brings source-traced output and the review record into the workpaper context, so teams spend less time reconstructing provenance and more time evaluating the evidence in front of them. That's the same record the file needs for desk review and later inspection. To see how traceable agent output changes the review desk on your engagements, request a demo.