AI in Auditing: DataSnipper’s Case for Human Oversight

Audit’s AI story starts with too few workers

AI in auditing is arriving in a profession that DataSnipper CEO Vidya Peters says already struggles to staff its work. She told Fortune there are two to two-and-a-half open jobs for every available person. That figure is her estimate; the article also cites U.S. workforce data showing the number of accountants and auditors fell from about 1.96 million in 2019 to 1.65 million in 2022, then recovered to about 1.77 million in 2025. The workforce figures and Peters’s estimate describe a shortage problem that does not fit neatly into the usual story of AI replacing office workers.

The pipeline and demand figures point in different directions. Schools awarded 55,152 accounting degrees in the 2023-24 academic year, down 6.6% from the year before, while accounting-program enrollment rose 12.4% in spring 2025. The Bureau of Labor Statistics projects about 115,300 openings for accountants and auditors each year through 2035, mostly to replace people who retire or leave. Those projected openings are not a count of unfilled jobs. The education and job-opening figures help explain why Peters sees automation as a way to support scarce staff, rather than simply a plan to eliminate roles.

Alwin separates agent approval from output review

DataSnipper launched Alwin, its agentic automation platform, shortly before Peters’s interview. She describes two different kinds of oversight: building and approving an agent, which the company restricts to experienced auditors, and reviewing the agent’s work, which more auditors can do. Peters’s description of Alwin and its approval roles makes the division of responsibility central to the product pitch.

Her example is a manager checking whether a junior auditor opened an agent’s work, commented on it and signed off, rather than letting it run unchecked. That is an illustrative workflow, not a reported measured result. Peters argues that the auditor remains accountable for the work, regardless of the technology used. In practical terms, automating steps does not by itself settle who must judge whether the evidence supports a conclusion. Peters’s account of the review workflow puts that judgment with people, even when an agent handles parts of the process.

Traceability is part of the trust pitch

Peters says Alwin connects each output to a source document, which DataSnipper calls a Snip. Her example: a user can inspect a stated amount of 101 euros and 23 cents and see where it came from. That gives reviewers a way to follow a result back to evidence, although the article does not independently assess the platform’s accuracy. The description of Alwin’s source links shows how DataSnipper says it supports review, not whether its outputs are correct.

The trust numbers in DataSnipper’s own survey sharpen the point. The company reports that trust in AI among audit and finance professionals fell to 55% in 2026, from 78% in 2025 and 74% in 2023. At the same time, 73% of respondents called AI essential, while only 13% of organizations had integrated it into real workflows. These are company-reported survey results; the article excerpt does not give methodology or sample details. DataSnipper’s reported survey findings suggest that seeing a role for AI and putting it into everyday work are different steps.

The useful distinction in Peters’s argument is between letting an agent do work and letting it stand in for professional judgment. In audit, the output has to be reviewable by someone prepared to approve it. Traceability can help make that review possible, but the final responsibility remains human in the workflow she describes.