Denki Is Betting That Internal Audit Is a Software Problem, Not a People Problem

The Macro: Auditing Is a Tax on Every Public Company

Internal audit is one of those functions that nobody outside of finance thinks about until something goes catastrophically wrong. Enron happened because auditing failed. Wirecard happened because auditing failed. Every major financial scandal of the last 25 years has an audit failure somewhere in the chain, and the response from regulators has been to make the rules more complex, not simpler.

SOX compliance alone costs large public companies between $1 million and $5 million per year. That is not a typo. The average Fortune 500 company has teams of internal auditors doing work that is, frankly, mind-numbing. They collect evidence. They document controls. They run walkthrough interviews. They produce workpapers that nobody wants to read but everyone needs to exist. It is compliance theater at industrial scale, and the tools most teams use are Excel, email, and whatever GRC platform their company chose in 2014.

The incumbents in this space are Auditboard and Workiva. Both are solid products by enterprise software standards. Both are also fundamentally built around the assumption that humans do the audit work and software helps them track it. That is the gap. Nobody has seriously tried to flip the ratio and make software do the actual audit work while humans provide oversight.

There are a few AI-adjacent plays in the space. Datarails does financial planning automation. Stampli automates accounts payable. But actual audit execution, the walk-throughs, the evidence collection, the testing, the workpaper generation, that is still almost entirely manual. The reason is not technical. The models are capable. The reason is trust. Audit is a high-stakes domain where errors have regulatory consequences, and nobody wants to be the first company to tell the SEC that an AI did their internal controls testing.

The Micro: Two Brothers From McKinsey and Imperial

Denki describes itself as a “99% software, 1% services” audit firm. That positioning is deliberate and important. They are not selling software to auditors. They are replacing auditors with software and keeping a thin services layer for the parts that still require human judgment. That is a fundamentally different business model than what Auditboard or Workiva offer.

Felipe Jin Li and David Jin Li are brothers. Felipe came from McKinsey and was doing PhD research on Explainable AI at University College London before leading product engineering at a GE-backed startup. David studied Computing at Imperial College, built financial data pipelines at MacroHive (which was acquired by BGC), and was ranked among the top 13 competitive programmers in the UK Informatics Olympiad. The combination is relevant. Felipe understands how consulting firms actually deliver audit work. David understands how to build systems that process financial data at scale.

They came through Y Combinator’s Fall 2025 batch and are based in San Francisco as a two-person team. The product integrates with existing platforms like Auditboard, Workiva, and enterprise ERPs to automate control processes. Evidence collection, walkthrough interviews, testing, workpaper generation. The traceability piece is critical. Every output can be traced back to its source data, which is table stakes for anything that touches regulatory compliance.

The competitive moat here is domain expertise married to technical depth. You cannot build an AI audit system without understanding what auditors actually do, why they do it, and which regulatory bodies will care about the output. The Deloittes and PwCs of the world have the domain knowledge but are not going to cannibalize their own consulting revenue. The pure AI companies have the technical chops but do not understand SOX compliance at a granular level. Denki is trying to sit in the middle.

I want to know more about their go-to-market. Enterprise sales in compliance software are notoriously slow. CFOs do not make impulsive purchasing decisions about audit tools. The services component might actually accelerate adoption by letting companies try Denki as a vendor rather than a software purchase, which sidesteps procurement entirely.

The Verdict

I think the 99/1 ratio is the right framing. Audit work is repetitive, document-heavy, and rules-based. That is exactly the kind of work that AI handles well. The question is whether companies will trust a two-person startup to handle work that has direct regulatory implications.

At 30 days, I want to see whether they have landed a pilot with any mid-cap or large-cap public company. The enterprise sales cycle is the biggest risk here. At 60 days, I want to know what the accuracy rate looks like on automated control testing versus human auditors. If the AI is catching things humans miss, that is the story that sells itself to every CFO in the country. At 90 days, the question is whether the services component is growing or shrinking as a percentage of revenue. If it is shrinking, the software is working. If it is growing, they have a consulting firm dressed up as a software company. The difference between those two outcomes is the difference between a $10 million business and a $1 billion one.