Punchcard
Agentic audit workspace that turns each request or test into a punchcard, with CoAudit drafting cited workpapers
Some outbound links may earn The AI Ledger a commission. Editor scores and verdicts are never influenced.
About Punchcard
Punchcard, formerly Agentive and built by Substantive AI, is an audit work-management platform in which every client request or audit task becomes a punchcard: a structured unit holding the instructions, the evidence, the agent's work and the reviewer's sign-off. Each punchcard moves through three stages: request, where the evidence needed is defined; test, where the agent extracts fields and runs procedures; and review, where auditors check citations and validate, override or sign off. Punchcard types cover substantive testing, control testing, client requests and financial statement review.
Its CoAudit agent works across Punchcard, Excel and Word, reading contracts, invoices, bank statements, payroll, debt agreements and offering memos, and gives reasoning and a citation to source evidence for each result; the site still labels it as in beta. It covers substantive and revenue and expense testing, unrecorded liability searches and SOX operating effectiveness testing. A workpaper agent handles detail testing with tolerances and tick marks, and a Financial Statement Review module ties the trial balance to the statements, checks disclosure coverage and runs consistency checks. Punchcard is SOC 2 certified, offers GDPR compliance, and names RSM, Richey May, RSM Ebner Stolz and Stambaugh Ness as users.
It is aimed at CPA firms and internal audit teams.
Key Features
- Punchcards holding request, evidence, agent work and sign-off together
- CoAudit agent with reasoning and source citations across Excel and Word
- Workpaper agent with tolerances and tick marks
- Financial Statement Review with trial balance tie-out and disclosure checks
- SOC 2 certified, with GDPR compliance available
Our in-depth review
Punchcard's idea is that audit work goes wrong in the hand-offs: the request that was never quite specified, the evidence saved in the wrong folder, the test nobody can trace back to source. Putting each piece of work into one structured card, with an agent doing the first pass and the reviewer signing off in the same place, attacks that directly.
What it does well. The structure is the strength. Instructions, evidence, agent output and sign-off live together, so a reviewer can see exactly what was asked and what was done. Every CoAudit result carries reasoning and a citation, which is what quality reviewers and inspectors will want to see. Coverage of substantive testing, SOX operating effectiveness and unrecorded liability searches is broad, and the Financial Statement Review module, with trial balance tie-out and disclosure checks, adds value at the end of the file. Working inside Excel and Word reduces change management, and named users such as RSM and Richey May show real adoption.
Where it falls short. Pricing is not published and every route in is a demo. CoAudit is still marked as beta on the site, so firms should be clear about what is production-ready before relying on it. Efficiency claims such as up to ten times faster testing are the company's own case studies. Engagement types are US-centred, including employee benefit plan audits and SOC 2 work, and nothing on the site addresses UK standards. No ledger or audit file integrations are named beyond Excel and Word.
Best for. US CPA firms and internal audit teams, roughly 11 to 50 auditors and up, that want structured, reviewable agent work across substantive and control testing. A trial should take one area, such as revenue testing, run it through punchcards and time preparation and review against last year. UK firms wanting cited AI work inside Excel should compare DataSnipper and Moby Analytics, and those after a full engagement platform should look at Fieldguide.
Verdict. A well-structured audit agent with proper citation trails and real adoption, but its flagship agent is still in beta and pricing needs a sales call.
At a glance
- Category
- Audit & fraud detection
- Best for
- 11–50 practice
- Editor's score
- 4.4 / 5
- Pricing
- Custom pricing on request
- Last verified
- 2026-10-10
A well-structured audit agent with proper citation trails and real adoption, but its flagship agent is still in beta and pricing needs a sales call.
Integrations
- Microsoft Excel
- Microsoft Word
- Browser-based
Get the weekly brief
New AI tools for accountants land every week. Get the 5-minute Friday brief, free.