Workpaper

Substantive testing, tied to source, every time.

CoAudit reads your client's source documents, runs your procedures, reasons through every assertion, and outputs a workpaper that's ready for partner and peer review. Detail testing, control testing, and contract review — one agent, one workpaper.

Trusted by RSM Ebner Stolz, Richey May, JLK Rosenberger. SOC 2 certified · GDPR-aligned.

Revenue testing workpaper
42 selections — 5 attributes
Running procedure
ID
Customer
Amount
Date
Appr.
Status
REV-104
Customer A
$12,450
Tied
REV-117
Customer B
$8,210
Review
REV-123
Customer C
$3,120
Open
Procedure applied
Reasoning attached
Exceptions flagged

Why a workpaper agent

Excel + ChatGPT doesn't produce an audit.

A prompt window can summarize an invoice. It can't run a firm-approved procedure on a sample of 400 with tolerances, tie each result back to the source page, and emit a workpaper your partner will sign. That's not a model problem — that's a context problem.

Editing Excel + prompting an LLM

No context, no traceability.

  • No audit procedure library — every prompt starts from zero
  • No access to PY workpapers or the current trial balance
  • No reasoning attached to cells — chat history rots
  • No structured tie-out — reviewer rebuilds from screenshots

Workpaper

Built for the audit, not the prompt.

  • Pre-loaded with your firm procedures and audit methodology
  • Reads the current TB, prior-year workpapers, and signed client docs
  • Reasoning trace + source citation on every assertion
  • Workpaper output — partner-ready, peer-review-ready

What it covers

One agent. Every substantive procedure.

Detail testing, control testing, and contract review all share the same shape — read the source, apply the procedure, cite the evidence. CoAudit handles all three from the same workpaper surface.

Detail testing

Selection-level tie-out across the population.

Revenue, expense, inventory, AR, AP — CoAudit reads source documents (invoices, POs, bank statements, contracts), runs every attribute test, applies tolerances, and cites the evidence for each conclusion.

  • Three-way match at scale
  • Cutoff and date attribute testing
  • Tolerance application with reasoning

Control testing

Operating effectiveness, sample by sample.

Test SOX and ITGC controls across populations. CoAudit reads approvals, segregation evidence, ticket trails, and access logs — and produces a workpaper that supports the operating-effectiveness conclusion.

  • Approval and authorization checks
  • Access and SoD verification
  • Sampling with deficiency tracking

Contract review

Memos and key-term extraction across the deal pool.

Lease, debt, revenue, and insurance contracts — CoAudit pulls the key terms, ties them to the trial balance, drafts the disclosure-level memo, and flags any clause that would change the accounting conclusion.

  • Key-term extraction with citations
  • TB-to-contract tie-out
  • Memo drafts with quoted basis

How it works

Match. Test. Cite. One workflow.

What used to be two separate punchcards — Match Agent and Test Agent — are now two parts of the same Workpaper workflow. Selections meet their evidence, procedures run, and the reasoning trace is in the workpaper when the auditor opens it.

01

Match

Surface the support

Likely evidence is linked to every selection — invoice to PO to bank statement, contract to revenue entry. The auditor sees the ranked candidates with a reasoning summary.

02

Test

Apply the procedure

Every assertion — existence, accuracy, completeness, cutoff, valuation — is run against the matched evidence. Tolerances apply. Exceptions stand out before review.

03

Cite & Review

Auditor judgment, surfaced

Reasoning trace and source citations attached to every cell. Reviewers see what was tested, why it passed, and where judgment is needed — without rebuilding the analysis.

Agent reasoning

Click any cell. See why.

Every assertion the agent makes carries the reasoning that supports it — what it read, what it compared, what it cited. Auditor judgment is preserved because the basis for the conclusion is in the workpaper, not in a chat log somewhere.

  • Read

    CoAudit opens invoice_1048.pdf, identifies header fields, line items, totals, and approval block.

  • Match

    Compares amount ($12,450) and customer (Customer A) against PO_8821.pdf and the GL entry. All three reconcile within tolerance.

  • Cite

    Tickmark generated. Reasoning trace and source page citations attached to the cell — reviewable in one click.

Reasoning — REV-104, Amount
CoAudit · 1.4s · 3 citations

Conclusion

Amount $12,450 ties to source. Tolerance: 0%. Tickmark: F&G

Reasoning

Read invoice_1048.pdf p.1, line total $12,450.00. Matched to PO_8821.pdf p.2 (line-item total $12,450.00) and GL_Revenue_FY26.xlsx row 41 ($12,450). All three reconcile within zero-tolerance threshold for Customer A.

Cited

invoice_1048.pdf
p.1, header total
PO_8821.pdf
p.2, line-item total
GL_Revenue_FY26.xlsx
row 41

In the workpaper

Procedures, results, and evidence — in one file.

The output is a workpaper, not a chat transcript. Reviewers see what was tested, what passed, what needs judgment, and the support behind each conclusion — sortable, citable, signed off in one place.

Revenue testing — FY26 workpaper
42 selections · 5 testing attributes · ready for partner review
35 tied5 review2 open
ID
Customer
Amount
Amount
Date
Approval
Cutoff
Status
REV-104
Customer A
$12,450
Tied
REV-117
Customer B
$8,210
Review
REV-123
Customer C
$3,120
Open
REV-128
Customer D
$21,090
Tied
Evidence — REV-104
invoice_1048.pdf · PO_8821.pdf
Amount, date, and approval support the tie-out.
For auditor judgment — REV-117
Approval stamp not visible
PO references approval but image is missing. Reasoning trace attached.
Partner sign-off
Workpaper-ready
Procedures applied · evidence cited · ready for peer review.

Why Punchcard is different

Context engineering, not generic prompting.

Two systems can both call an LLM. Only one is engineered to know your firm's methodology, the client's trial balance, the prior-year workpapers, the open requests, and the structured procedure library — at the moment the assertion is made.

Client context loaded

Current TB, signed contracts, PY workpapers, open requests — pre-loaded for every prompt.

Firm methodology applied

Your audit procedure library, sampling rules, and tolerance thresholds wired into the agent.

Workpaper-shaped output

Not text. A workpaper with tickmarks, citations, assertions, and a basis for partner sign-off.

Workpaper vs. editing Excel directly

Apples-to-apples on what the auditor (and the reviewer) actually has when the workpaper is opened.

Question
Excel + ChatGPT
Workpaper
Where the reasoning lives
In a chat log nobody re-reads
Attached to every cell, reviewable in place
Source citation
Manually copy/paste filenames
Auto-cited to page and paragraph
Population coverage
Sample what fits in a prompt
Run procedures across the full population
Audit context
Generic LLM — no firm methodology
Pre-loaded with your procedures, PY workpapers, client TB
Review traceability
Reviewer reverse-engineers from cells
Workpaper opens with reasoning + evidence behind every conclusion
Partner sign-off
Side workbook + side prompts + side memos
One workpaper — partner-ready, peer-review-ready

Common questions

What audit teams ask before a demo.

Is this just AI in Excel?

No. Excel + a chat window has no audit context. Workpaper loads your firm methodology, the client trial balance, prior-year workpapers, and the open requests — then reasons through every assertion against the source documents and emits a structured workpaper. The reasoning lives on the cell, not in a chat history.

Does the auditor still sign off?

Always. The agent does the mechanical work — reading documents, applying procedures, tying selections to evidence. The auditor reviews the reasoning, exercises judgment on flagged items, and signs the workpaper. CoAudit is named that way for a reason — judgment stays with the human.

What testing does it cover?

Detail testing across substantive procedures (revenue, AR, AP, inventory, expense, payroll), control testing for SOX and ITGC populations, and contract review (lease, debt, revenue, insurance memos). All three share the same workpaper surface.

How is reasoning attached to the workpaper?

Every cell carries a reasoning trace — what the agent read, what it compared, what conclusion it drew, and the page citation. Reviewers click any cell to inspect the basis. Exporting the workpaper exports the full evidence chain.

Is it independent enough for SOX or SAS audits?

Yes. Independence is a workflow property: the agent runs the procedures, the auditor reviews the conclusions. We work with firms operating under PCAOB, AICPA, and IFRS standards. SOC 2 certified. GDPR-aligned.

Assurance, one punchcard at a time.

See Punchcard handle a real engagement. We'll walk through requests, testing, and review using your evidence — every judgment call still in human hands.

Trusted by RSM, Richey May, JLK Rosenberger, and more.