Cortea AI for auditors focuses on review trails and human sign-off

Cortea AI for auditors is being pitched around reviewability rather than unattended automation. In a sponsored Sifted interview published September 10, CEO and cofounder Valentin Neumann said the company’s software produces drafted or finished audit work, then has an auditor review findings, verify them and document approval. Cortea did not disclose revenue, valuation, headcount or customer contract values in the material reviewed.

The distinction is material in audit work, where Neumann said errors can create financial and legal consequences. Cortea’s proposition is that an audit firm needs more than a fast answer from a general-purpose model: it needs a way to assess the result against the engagement’s evidence and carry that work through its own review process.

How does Cortea make AI audit work reviewable?

According to Neumann, Cortea sets detailed instructions for each stage of an audit and directs the model to use those instructions rather than rely on its pretrained knowledge. He said Cortea can show a regulator the step-specific instructions used and the resulting output. That is narrower than a claim that every aspect of a model’s internal logic can be inspected, and the available material does not independently establish regulatory acceptance of the approach.

Cortea describes its product as providing supporting context and links back to the documents and data behind an output. The company says it can document testing results in an inspection-ready audit trail and connect working papers, procedures and disclosure checklists. Those are product claims, not independently tested measures of audit accuracy.

In practical terms, the company says its agents can cross-check figures, disclosures and assertions across the audit file, identify inconsistencies or potential errors, and flag items for auditor judgment. It also says the platform can test a user-defined sample against audit criteria, prepare risk analysis and audit plans from risk indicators and prior engagement data, and check financial statements against IFRS or UK GAAP disclosure requirements.

Where does the auditor remain in control?

Cortea says its system is designed to fit existing audit methods rather than replace them. Its product materials describe support for completion procedures, engagement-quality reviews and final partner reviews while leaving systems of record in place. The intended workflow is that the software flags, drafts and analyzes, while the auditor reviews, approves and signs the conclusion.

That positioning addresses a core adoption problem identified by Neumann: audit firms may want AI assistance without rebuilding familiar processes around a chat interface. Cortea says it has therefore borrowed interface concepts from incumbent tools. The evidence does not show how broadly customers use the product or whether those design choices improve adoption.

The company’s trust case is best understood as a set of claimed controls: constrain the task with audit-specific instructions; tie outputs to evidence; route work through quality-review stages; and leave professional judgment with the auditor. Cortea’s website makes wider claims about quality, standards alignment and performance, but the available material does not provide independent validation of those assertions.

This story draws on original reporting from Sifted.