OpenAI's CFO Shares 5 Lessons From Building an AI-Native Finance Team

Hey everyone, it's me, Shiichan! Today I found a story that's a little different from my usual beat: how AI is changing finance and accounting work, not just engineering. Turns out the AI wave has reached the people who handle a company's numbers too.

OpenAI News openai.com

What was announced?

OpenAI's News published an essay by CFO Sarah Friar reflecting on how she built an AI-native finance function. Friar joined OpenAI two years ago, when the company was growing fast but the finance team was still small. From there, she set two big ambitions.

• A zero-day close (closing the books the same day, instead of over weeks)
• Automated, continuously updated forecasting

The article walks through five lessons learned while rebuilding finance from the ground up over those two years.

The story so far

Traditional finance teams tend to run on a "close, then look back" cycle: close the books at month-end, tally the numbers, and eventually produce a report. By the time the numbers are ready, the situation has often already moved on, which pushes decision-making behind the curve.

What Friar's team aimed for was a shift away from that pattern: instead of looking backward after closing, build a finance function that sees the business in real time and can support forward-looking decisions.

What changes

• Finance shifts from a backward-looking, reporting role to one that tracks the business in real time and moves proactively.
• Finance professionals without coding backgrounds have become builders, creating their own dashboards and tools.
• Speed isn't the only goal: approval flows and traceable data sources were built in alongside it, making it safer to hand work to AI.

Dive Deep

The article lays out five concrete lessons behind this transformation.

• Give everyone access, then create a reason to use it: AI tools work best when broad access is paired with structured experimentation tied to real problems. A finance hackathon produced tools like IR-GPT, a custom GPT for investor relations, plus similar tools for procurement and tax.
• Redesign the full workflow around the decision: rather than automating a single task, the team rebuilt the entire flow leading up to a decision. The zero-day close continuously reconciles approved spending plans, general-ledger actuals, purchase orders, accruals, and transaction details in one always-connected view. Forecasting evolved into interactive dashboards that combine statistical models, sales data, and scenario analysis, including adjusted scenarios against quarterly ARR projections and a visualization of diminishing returns for capital allocation.
• Finance professionals become builders: research cited in the piece found that 40% of finance professionals' specialized AI use involves work outside traditional finance, and 22% involves engineering-related tasks. Staff without coding backgrounds now build their own dashboards using ChatGPT Work and Codex.
• Pair speed with clear accountability and controls: running IR-GPT showed that acceleration needs guardrails, defined data-access scope, required approvals, traceable sources for outputs, and finance sign-off on any changes to baseline numbers. As the article puts it, "AI accelerates the work. People own the result."
• Measure value per unit of intelligence: instead of tracking adoption metrics, ask whether AI completed work that actually mattered, what it cost including review and rework, whether the output was good enough to use, and whether it helped the team move faster or decide better.

Wrap-up

• OpenAI CFO Sarah Friar looks back on two years spent rebuilding the finance function around AI.
• The zero-day close and continuous forecasting are the two pillars, with finance staff becoming builders as the key shift.
• Speed was paired with controls by design: approval flows and traceable data sources are what make it safe to hand work to AI.
• Measuring success by "did AI finish work that mattered" rather than adoption rates is a useful framing too.
• Worth a read for anyone in finance, accounting, or business operations thinking about how to bring AI into their own workflows, not just engineers.