AI audit software: journal entry testing, controls and what auditors still sign (September 2026)

Who uses AI audit software, and for which job

The category splits by buyer. External audit firms use tools such as DataSnipper, Fieldguide, AuditFile or AUDAVIS to plan engagements, test transactions and assemble workpapers. Internal audit and SOX teams use platforms such as Optro, formerly AuditBoard, or Diligent to manage audits, test controls and report to the board. Finance teams run ledger analytics such as MindBridge or zapliance to watch their own books continuously.

Some tools we list in the category come from neighboring finance software: close and reconciliation platforms such as BlackLine or FloQast, and accounts payable tools with control features. They matter for internal controls, but they are not audit software in the narrow sense.

The explorer below shows the category as of September 26, 2026. Filter by main buyer or headquarters region to see which share of tools publishes a price, offers a free trial, keeps data in the EU or holds security certifications such as SOC 2, ISO 27001 or ISO 42001 for AI management.

Sources: [1][2][3][4]

Journal entry testing: from a sample to the full ledger

Journal entry testing is the procedure these tools change most. Under PCAOB AS 2401, the auditor must design procedures to test the appropriateness of journal entries recorded in the general ledger and other adjustments. The standard notes that fraudulent entries are often made at the end of a reporting period and that computer-assisted audit techniques may be needed to find the entries to test.

The standard also names what should guide the choice of entries: the fraud risk assessment, the effectiveness of controls, the entity's financial reporting process and the characteristics of fraudulent entries. A risk score applies those factors to every line, but it does not replace the auditor's reasoning about them.

Traditional tests filter the ledger by rules: weekend postings, round amounts, unusual users. AI tools add a risk score to every line. MindBridge combines statistical models, business rules and machine learning, groups findings into high, medium and low risk and analyzes up to 1 billion transactions in a single analysis. Its own examples show the ranking: suspicious keywords and atypical changes in amounts count as high risk, rare vendor accounts and unusual volumes as medium, a Benford's law analysis as low.

European vendors take the same approach for smaller engagements. AUDAVIS from Munich lists 37 journal entry tests and 48 data-driven audit procedures and rates the risk of each posting line with an AI algorithm. zapliance says its SAP agents cut a manual review of 5 to 10 working days to 2 to 3 days, a vendor figure rather than an independent measurement.

A full-population test produces more flagged items than a sample, and the auditor still decides which ones to investigate and what they mean for the opinion.

Sources: [5][3][6][7][8]

What the standards require: PCAOB 2024 amendments and the revised ISA 240

Standard setters have started to write technology into the rules. On June 12, 2024, the PCAOB adopted amendments to AS 1105 on audit evidence and AS 2301 on responses to the risks of material misstatement, and the SEC approved them on August 20, 2024. They apply to audits of financial statements for fiscal years beginning on or after December 15, 2025.

The amendments do not tell auditors which tool to use. They specify that when technology-assisted analysis identifies items meeting certain criteria, the investigation must include whether those items, individually or together, indicate misstatements or control deficiencies. They also stress the auditor's duty to evaluate the reliability of the electronic information used as evidence, and that a procedure used for several purposes must achieve each of them.

Outside the US, the IAASB issued ISA 240 (Revised) on July 8, 2025. It applies to audits of financial statements for periods beginning on or after December 15, 2026, and requires a focused fraud lens when identifying and responding to risks, together with clearer communication with those charged with governance and more transparency in auditor's reports.

Sources: [9][10][11]

How audit teams use AI in 2026: drafting leads, testing lags

AI use in audit is common, but it sits mostly in writing tasks. In a Gartner webinar poll of 743 audit professionals, reported in August 2026, 93 percent said they used AI to some degree, while only 38 percent had an AI strategy.

The uses cluster around documents. 60 percent used AI to draft audit issues, ratings or reports and 41 percent to review drafts, while 35 percent applied it to risk assessment and audit planning. Only 30 percent used it for audit testing and 12 percent for quality assurance reviews, the work where full-population analytics would change results most.

Gartner's analyst James Bourke said that current use of generative AI concentrates less on strategic audit use cases and more on moderate productivity improvements. For buyers, that suggests starting with the testing gap rather than adding another drafting assistant.

Sources: [12][13]

Agents in the audit file: what they do, and who signs off

The newest tools are agents that carry out whole audit steps and hand the result to a person. Fieldguide names 4: Field Planner for evidence-backed planning, Field Auditor for testing and evidence verification, Field Reviewer for completeness checks before sign-off and Field Financials for drafting statements and disclosures.

DataSnipper, which works inside Excel, sells 3 plans: Start without agents, Accelerate with Excel agents and prebuilt agents, and Elevate with its Alwin agents, which do the heavy lifting while people review and sign off. AuditFile includes its Advantage AI audit agents from the Pro Plus plan. Thomson Reuters' CoCounsel Audit answers accounting questions with citations from FASB, GASB, AICPA and IFRS material and drafts workpapers and engagement letters.

On the internal audit side, Diligent's Audit Agent rolls forward prior audits, drafts PBC requests, executes control tests and flags exceptions for review. Optro's Autonomous Testing handles sampling, evidence gathering and document annotation, and the company describes its recommendations as machine generated and subject to human review.

Vendors attach large numbers to these agents. Fieldguide cites a 55 percent reduction in engagement hours and Diligent about $10,000 saved per audit; both are the vendors' own figures, and neither page explains the method. Ask for the calculation behind any such claim and test it on one engagement of your own.

Sources: [4][14][15][16][2][1]

Pricing: three vendors publish, most quote on request

Audit software is one of the least transparent categories in our directory. 84 percent of the tools in our audit category list their price as on request, and only 6 offer a free trial. In the neighboring tax category the share on request is 52 percent.

Three vendors publish per-user prices. AuditFile charges $179 per user per month on Professional and $249 on Pro Plus with an annual agreement, or $199 and $299 month to month, with a 14-day free trial. AUDAVIS lists 160, 220 and 350 euros per user per month for Basic, Professional and Enterprise, capped at 20 and 100 users for the first two. Priviq AI from Hamburg charges 160 euros per user per month with annual billing or 200 euros monthly for Pro, hosts in Germany and offers a 14-day pilot for up to 3 users.

DataSnipper, MindBridge, Fieldguide, Optro, Diligent and the CoCounsel products quote after a demo. Ask for the price per user and per engagement, and for which plan includes the agents, since at DataSnipper and AuditFile the agents sit in the higher plans.

Sources: [15][17][18][14]

Choosing AI audit software: start from the procedure, not the tool

Pick the procedure first. If journal entry testing and ledger risk are the pain, look at ledger analytics; if evidence collection eats the budget, look at request and matching tools; if control testing and SOX documentation dominate, look at internal audit platforms. The decision tree below sorts the main cases.

Then check what the standards will ask. Under the PCAOB amendments, you must evaluate the reliability of the data the tool analyzes and investigate what it flags, so ask each vendor how it documents data completeness, scoring logic and every override. The AI tools used by tax teams raise similar questions about sources and approval, with different rules behind them.

Finally, settle data location and model use. Durava states that it hosts only in Germany and does not use documents for AI training, zapliance runs on premises with customer-approved language models, and Optro offers an MCP server so that customers can use their own models on platform data.

Sources: [9][19][8][1]