Audit is the part of corporate life almost no one thinks about, and almost everyone pays for. Public companies spend thousands of hours a year collecting evidence, testing controls and writing it all up. Most of it is done by hand, in Excel, much the way it was in 2002.
Andera wants to hand that work to AI. The San Francisco startup, founded in 2024, has raised a $37mn Series A led by Lightspeed Venture Partners to automate internal audit from end to end.
A reasoning problem, not a rules problem
Audit resisted automation for a reason. An auditor has to read messy, multi-format evidence, judge whether it satisfies a control, and document it well enough to survive a regulator. No rules engine could do that.
Large language models can try. Andera’s platform reads financial evidence across spreadsheets, PDFs, screenshots and journal entries, then renders an audit judgment for controls under Sarbanes-Oxley and other regimes. “The models are already sufficiently smart,” chief executive Aryo Patel said. “The hard problem is getting from billions of tokens to the 30,000 that matter.”
Patel, a Microsoft and Jane Street alumnus, and co-founder Tinah Hong, formerly of Stripe, met in middle school in Chicago. They have paired engineers with career auditors, including a former Deloitte accountant, to build the product.
The Big Four’s dilemma
The pitch lands at a useful moment. The average public company spends about $3mn a year on audit fees, Lightspeed says, and a Fortune 100 firm more than $20mn. Finance chiefs now want their back offices to do far more with less.
That puts the Big Four accounting firms in a bind. Their business is built on billing hours, so the better the AI they build, the more of their own revenue it eats. It is the same squeeze now hollowing out the consulting industry, and it leaves an opening for a startup with nothing to cannibalise.
Coming for the white-collar professions
Andera is tiny, with a handful of staff, but it says it already works with Fortune 100 customers. It is part of a wave of startups aiming AI not at chatbots but at the well-paid professions: corporate lawyers, patent attorneys, and now auditors.
The reassuring version, the one Lightspeed is selling, is that auditors are extended, not replaced, freed from busy-season grunt work to focus on real risk. The less reassuring version is the question every white-collar worker is now learning to ask: what happens when the AI gets good enough to do the judgment too?