Sarah Friar, OpenAI’s chief financial officer, published an account of how she has spent two years rebuilding the company’s finance department around AI, setting two long-term targets: a “zero-day close” that reconciles the books in real time, and forecasts that refresh continuously instead of on a monthly cycle. The post is OpenAI describing its own internal transformation, built on its own products, with no outside auditor or independent finance team confirming the results. That framing matters because Friar is also the executive best positioned to sell other CFOs on adopting the same tools.
A zero-day close is a genuinely ambitious target. Most finance organizations run monthly closes that take days or weeks precisely because actuals, purchase orders, and accruals live in separate systems that do not talk to each other cleanly. Friar describes connecting those systems into one continuously reconciled view, with AI drafting an initial explanation for each variance and flagging exceptions, while finance staff keep sign-off authority. She says the close itself will not disappear. What goes away, in her account, is the scramble to reconstruct what happened after the period has already closed.
The five moves Friar credits for getting there are less about any single tool than about how the work gets restructured. Broad access to AI came first, paired with a structured venue for testing it: a hackathon that paired finance staff with sales engineers produced IR-GPT, an internal chatbot grounded in approved investor relations material that now drafts responses to diligence questions in seconds rather than hours. From there, her team redesigned entire workflows around a single decision rather than automating isolated steps, treating the monthly close and the forecast review as one continuous pipeline instead of a sequence of handoffs. Finance staff who had never written code, including one teammate on the advertising team, used Codex to build their own dashboards, turning static spreadsheets into live tools that update as new data arrives. Friar cites internal OpenAI research finding that 40 percent of finance staff’s specialized AI use falls outside conventional finance tasks, with 22 percent counting as engineering work, a sign of how far the job description has already shifted. Speed was paired with explicit controls: every AI-drafted output still routes through a human who owns the final answer, and budget limits, routing rules for which model handles a task, and thresholds that trigger approval now govern AI usage the way any other variable cost would be managed. Finally, Friar argues CFOs should judge AI by a scorecard tied to outcomes (did it complete work that mattered, at what cost, well enough to use) rather than by seat counts or token spend.
The useful question for a reader outside OpenAI is which of these five pieces travel. The zero-day close itself depends on a resource most finance teams do not have: unlimited access to frontier models and to the engineers who built them, sitting one Slack channel away. Few CFOs can pull in the researchers behind their own AI vendor to help redesign a reconciliation pipeline. The access-plus-hackathon model, the accountability structure around AI-drafted output, and the outcome-based scorecard are portable regardless of company size. The zero-day close is not a near-term template. It is a preview of what becomes possible once AI infrastructure and engineering talent stop being the bottleneck, which for most finance organizations they still are.
CFOs evaluating this playbook should borrow the accountability and measurement pieces first, since those require no new infrastructure, and treat the zero-day close as a multi-year target that assumes engineering resources most finance departments will need to hire or contract for before attempting it.
According to OpenAI’s own company blog, in a post by CFO Sarah Friar titled “What building an AI-native finance function taught me.”
