Most companies talk about using AI to transform their operations. OpenAI’s CFO Sarah Friar actually had to do it — at the world’s most scrutinized AI company, with the world watching every move. The lessons she’s sharing now aren’t theoretical. They come from building what she calls an AI-native finance function from the inside out, and some of them will surprise you.
Friar published a detailed breakdown on August 10, 2026, walking through five concrete lessons OpenAI’s finance team learned while deploying its own tools internally. This isn’t a press release dressed up as thought leadership. It’s specific, occasionally uncomfortable, and worth reading carefully if you work in finance, run a company, or just want to understand how AI actually lands inside a large organization.
Why Finance? Why Now?
Finance functions are, frankly, an ideal test bed for AI. They’re data-heavy, deadline-driven, and full of repetitive analytical work that eats up senior talent. Forecasting, variance analysis, close processes, audit prep — these are exactly the tasks where pattern recognition and speed matter most.
But they’re also high-stakes. A bad AI output in marketing costs you a bad campaign. A bad AI output in finance can distort a board presentation, trigger a restatement, or undermine trust with investors. So when Friar says OpenAI leaned hard into AI for its own finance team, that’s not a casual experiment. It’s a deliberate bet that the tools are good enough to trust with serious work.
OpenAI has been pushing enterprise adoption hard through ChatGPT Enterprise — and the company has documented real-world deployments across industries. We’ve covered how HSP Gruppe uses ChatGPT Enterprise for complex tax work, which offers a useful parallel: professional services firms using AI not to replace expertise, but to make experts dramatically faster. Friar’s lessons follow a similar logic, applied internally.
The Five Lessons, Broken Down
1. Automate the Forecast, Not Just the Report
Friar’s first lesson is about where AI creates the most leverage. Most finance teams start by automating reports — pulling data, formatting outputs, building dashboards. That’s useful, but it’s not transformative. The real unlock, she argues, is automating the forecasting process itself.
OpenAI’s team built systems that continuously ingest operational data and update financial models in near real-time. Instead of a quarterly forecast that goes stale the moment it’s published, they’re working with rolling forecasts that reflect what’s actually happening in the business. That’s a meaningful shift — it changes how finance teams interact with business partners, moving from backward-looking reporters to forward-looking advisors.
h3>2. AI Requires Better Data Discipline, Not Less
Here’s something Friar is refreshingly direct about: AI doesn’t fix messy data. It amplifies it. If your source data is inconsistent, poorly labeled, or siloed across systems, an AI model will confidently produce wrong answers at scale. Fast.
OpenAI’s team had to invest seriously in data infrastructure before the AI applications could deliver reliable results. This is a lesson most organizations learn the hard way. The temptation is to layer AI on top of existing systems and hope it figures things out. It won’t. Data hygiene isn’t a prerequisite that can be skipped — it’s foundational.
3. Controls Get Stronger, Not Weaker
One of the more counterintuitive findings: deploying AI in finance actually tightened their control environment rather than loosening it. Friar describes how automated workflows create clearer audit trails than manual processes. Every decision, every data pull, every model output is logged. There’s less “this is how we’ve always done it” and more documented, reproducible process.
That matters a lot for a company at OpenAI’s stage. As it prepares for greater scrutiny — financial, regulatory, and public — having AI-enforced controls is actually a competitive advantage. Auditors love paper trails. AI creates them automatically.
4. Measuring AI ROI Is Harder Than You Think
Friar doesn’t claim AI has paid for itself cleanly with a tidy percentage improvement. Instead, she’s honest about the challenge of measuring return on AI investment in a finance context. Some benefits are direct and quantifiable — hours saved on close processes, for instance. Others are diffuse: better decisions made because analysts had more time to think, or risks caught earlier because models flagged anomalies.
Her recommendation is to track both hard metrics and leading indicators of quality. Are forecasts more accurate? Are variance explanations faster? Are finance team members spending more time on strategic work and less on data wrangling? These aren’t perfectly clean ROI numbers, but they tell a real story about value creation.
5. The Human Layer Doesn’t Disappear
The fifth lesson is arguably the most important: AI-native doesn’t mean human-optional. Friar is explicit that the finance team’s judgment, relationships, and contextual knowledge remain irreplaceable. What changes is the mix of work. The team spends less time on mechanical tasks and more time on interpretation, communication, and decision support.
This challenges a common fear and an equally common overclaim. AI won’t eliminate finance teams. But finance teams that don’t adapt to work alongside AI will find themselves outpaced by those that do. That’s not a threat — it’s just math.
What This Actually Means for Enterprise Finance Teams
Let’s be direct about who this matters to. If you’re a CFO or finance leader at a mid-to-large company, Friar’s writeup is one of the more useful first-hand accounts you’ll find. Not because OpenAI’s situation mirrors yours — it almost certainly doesn’t — but because the failure modes she describes are universal.
The data problem is everywhere. The measurement challenge is everywhere. The temptation to automate outputs before fixing the underlying process is everywhere. Reading this as a cautionary tale is as valuable as reading it as a success story.
For technology vendors selling AI into finance functions, the implications are also pointed. The companies winning enterprise finance deals aren’t just selling capabilities — they’re selling implementation frameworks, data readiness assessments, and control documentation. That’s the gap between a demo that impresses and a deployment that sticks.
It’s also worth placing this in context of what we’re seeing across OpenAI’s enterprise push. Circles used OpenAI tools to boost telecom revenue by 22% — that kind of outcome doesn’t happen without serious workflow redesign. The pattern is consistent: the biggest wins come when organizations rethink the process, not just plug AI into the existing one.
The Broader Signal Here
OpenAI publishing this isn’t purely altruistic. It’s a sales document dressed as a thought leadership piece — and there’s nothing wrong with that, as long as you read it with that lens. The company is making a case that its tools are mature enough for high-stakes internal use, and that the lessons from that use are transferable to customers.
The timing matters too. OpenAI is navigating a complicated moment: intense competition from Anthropic, Google DeepMind, and increasingly capable open-source models, while also making the case for enterprise reliability. A CFO-level endorsement of internal AI adoption — complete with candid lessons learned — is a more credible signal than any benchmark score.
And Friar, who previously served as CFO at Salesforce and CEO at Nextdoor, brings credibility that matters here. She’s not a technologist making promises. She’s an operator reporting results.
- Automate forecasting processes, not just reporting outputs — that’s where real leverage lives
- Fix data infrastructure first — AI amplifies bad data as readily as good data
- AI creates stronger audit trails — the control environment improves with automation
- ROI measurement requires both hard metrics and quality indicators — don’t expect a clean percentage
- Human judgment remains central — the work changes, but the team doesn’t disappear
Frequently Asked Questions
What is an AI-native finance function?
An AI-native finance function is one where AI tools are embedded into core processes — forecasting, reporting, controls, and analysis — rather than used as occasional add-ons. It means designing workflows around AI capabilities from the start, not retrofitting AI into existing manual processes.
What tools did OpenAI’s finance team actually use?
Friar doesn’t provide a detailed tech stack breakdown, but OpenAI’s internal use of ChatGPT Enterprise and custom-built AI agents is well-documented. The finance team would likely be using a combination of proprietary models and purpose-built financial automation tools layered on top of standard ERP and FP&A infrastructure.
Is this approach realistic for smaller companies?
The principles Friar outlines — data discipline, process redesign before automation, rigorous measurement — apply at any scale. The implementation complexity scales with company size, but a smaller finance team can often move faster precisely because they have less legacy process to work around. The data infrastructure investment is the real barrier for most.
How does this compare to what other companies are doing in AI finance adoption?
OpenAI is ahead of most, but not alone. Companies like Workday, SAP, and Oracle are embedding AI deeply into their finance platforms, and enterprise customers are increasingly building custom agents on top of those systems. The differentiator OpenAI has is that it’s both the tool vendor and the case study — which creates a unique feedback loop between product development and real-world use.
The honest takeaway from Friar’s account is that building an AI-native finance function is genuinely hard work, not a flip of a switch. But the organizations doing it well are pulling ahead in ways that will be difficult to close later. As AI capabilities continue to advance — and as the real-world usage patterns for ChatGPT keep expanding into professional workflows — the gap between early adopters and laggards in enterprise finance is only going to widen. The question isn’t whether finance functions will transform around AI. It’s whether your team is driving that change or reacting to it.
