Deep Dive · Notes by Xiaohu

OpenAI's CFO on rebuilding finance with AI: two goals, five lessons, and a scorecard you can steal

Two goals still in progress, a practical scorecard, and the details hidden in three internal tool screenshots.
The 60-Second Read
  • OpenAI's CFO details how her team rebuilt finance around AI, targeting near-instant close cycles and forecasting without manual refreshes—both still under construction.
  • The three internal tool screenshots are more revealing than the five lessons: every AI forecast adjustment comes with a trail of clickable evidence.
  • To measure if it's worth it, she offers four questions, none of which involve seat counts or token burn.
⚑This is OpenAI's own narrative about using its products (ChatGPT Work, Codex). No third-party verification is available, and financial figures and client names in images are anonymized. This article relays only verifiable parts; two external data points were checked against the original report.
The Starting Point

Finance work was mostly assembling materials for decisions

OpenAI CFO Sarah Friar describes how she spent two years building her finance team from scratch into an "AI-native" operation. It's a practical field guide: two goals, five lessons, a scorecard ready to copy, plus three screenshots of internal tools—the screenshots are far more concrete than the text.

First, the problem: finance runs periodic forecast reviews. The actual decisions boil down to a few items: how much can we make this quarter, where to add budget, where to pull back. But before anyone can make those calls, the team goes through this entire sequence:

Gather data
Reconcile
Investigate gaps
Build charts
Write docs
Make slides
In front of decision-makers, only now do they decide
The first six steps are preparation; the last is the reason for the meeting.

The other monthly ritual is the close—finalizing last month's books: how much was spent, where, and the variance against the approved budget. It sounds like simple arithmetic, except the numbers live in different places. They're typically scattered like this:

Money already paid

This sits in the financial system, the most solid part.

Committed but unpaid

Lives in the procurement system, invisible to the finance system.

Work done, invoice pending

Needs an estimate (an "accrual"), typically living in someone's Excel sheet. Like a late-night delivery on the last day of the month—you still count that meal in this month's food budget.

The "why we overspent" story

Often buried in a chat thread, known only to the people involved.

So at month-end, the team's job is essentially reconstructing last month from scratch: pulling numbers from four different systems to piece together what was spent and why.

The Direction

Two goals: zero-day close and continuous forecasting—both a work in progress

She set two goals, and they're sequential.

Zero-day close

See the company's financial position at any moment: numbers are correct, and every item can be drilled into to see its origin. Traditionally, a close takes days to weeks; "zero-day" aims to shrink that wait to zero.

Continuous forecasting

With the books always current, forecasts can follow suit: where the business is heading, what next quarter looks like, and which decisions change the numbers. No more manual monthly spreadsheet updates.

Why the first goal first? Because forecasts are built on book figures. With a constantly updated ledger, forecasts can be recalculated anytime. If the books only close at month-end, then even frequent forecasting uses stale data—the more often you calculate, the faster you compound the error.

Why spend two years chasing these? The payoff isn't speed itself. Faster closes and frequent forecasts are means to an end: seeing a problem while you can still do something about it. Discovering a line item collapsed after the quarter ends leaves no options; spotting it mid-quarter means you can reallocate budget, pull back on ads, or change pricing.

Both goals are still under construction. The article is about the direction and how work processes have already changed, not about finished results: it doesn't give a single number for how long a close takes now or how often forecasts refresh.

Lesson One

A one-day finance hackathon produced IR-GPT

The first step was giving everyone access and letting people experiment with their own tasks. But access alone isn't enough—most people try it twice and then drop it, because nobody has time to figure out "how could this task be done differently?"

So they ran a hackathon: one day, the entire finance team set aside regular work. Each person picked their most tedious repetitive task and built a tool for it on the spot. They brought in sales engineers to help—the tech-savvy folks who usually build solutions for clients. Anyone stuck had someone to ask. By the end of the day, someone could go the whole distance: pick a task they do monthly, build something usable, have a colleague try it, and iterate immediately.