OpenAI on OpenAI: Stacie Faggioli, Business Finance Officer Applications, OpenAI

### TL;DR OpenAI finance lead Stacie Faggioli shared how their team fundamentally rebuilt financial workflows using three principles: embracing AI natively, demonstrating human leverage, and deploying early and iterating. By leveraging ChatGPT, Codex, and embedded agents, they significantly improved efficiency and reduced reliance on external consultants. ## Introduction Hi, I'm Stacie Faggioli, and I oversee the finance team for applications at OpenAI. Over the past few years, I've seen firsthand how our team has changed the way we work with AI. Today I want to share a few examples of how we are building the finance team of the future. ## Three Principles for Building the Finance Team of the Future My team follows three core principles when making decisions: ### 1. Embrace AI Natively by Design When we deploy a tool or agent, we don't treat it as an add‑on to existing processes. Instead, we fundamentally reimagine how a workflow could be done. We think about how to spread best practices across the team and cross‑functionally, and we hire and design our organizational structure around agents. ### 2. Demonstrate Human Leverage A PwC assessment of our finance team and operations found that our finance team is about 20% the size of similar tech peers. We are living proof that with the right technology and tools, you can actually do more with less. ### 3. Deploy Early, Iterate Fast Technology changes very quickly. When we face the choice of "waiting for a more stable version," our bias is always to deploy directly and iterate continuously. This allows the team to grow organically alongside the technology, and we feel we are building the finance team of the future in real time. ## Finance Team Structure Our finance organization is divided into three pillars: - **Strategic Finance**: allocates capital, raises capital, and plans growth. - **Finance Operations**: accounting and the like – the engine of finance: collecting revenue, paying bills, filing tax returns, closing the books each month. - **Corporate FinTech**: looks after the finance systems and data platform. We embed engineers directly into finance, under the FinTech pillar, so engineers work side‑by‑side with finance subject‑matter experts. This accelerates deployment and iteration rather than waiting for IT to build from requirements. ## Deployment Directions: Personal Productivity & Team Agents I think about deployment in two ways: tools that boost personal productivity, and agents that help the whole team become more efficient. ## Personal Productivity Tool Examples ### ChatGPT: Investor Relations Agent This is my personal favorite example – simple to implement (non‑technical people can do it), but with extremely high ROI. During my time at OpenAI, I participated in two historic equity capital raises: raising $40B from private markets last year, and the $122B round announced a few weeks ago. Our team was overwhelmed by investor diligence requests. We trained an investor relations agent based on internal data and the tone of public‑company investor relations professionals. It could provide investors with data‑driven, factually accurate, consistent, and strategically framed answers in minutes. With the agent, everyone on the team could deliver CFO‑level answers to investors. We completed both rounds entirely in‑house, saving hundreds of millions of dollars in advisory fees. Even more interestingly, the agent helped sell the company's equity story. Whenever we recruit senior executives who receive large equity grants, they want to know what those grants are worth. We shared the investor relations agent with the recruiting team and other executives to help them tell the story a CFO would tell. ### ChatGPT for Excel: LBO Model As a former private equity analyst, I still remember pulling all‑nighters to build LBO models. The workflow is completely different now. I open Excel, use ChatGPT for Excel, upload a PDF equity research report, and tell it: "Think like an investment professional, lay out the financial projections, and build me an LBO model." ChatGPT for Excel starts planning the analysis, lists assumptions, and then fills the spreadsheet. It builds a model with complex assumptions – capital structure, cost of capital for each equity and debt layer, sources and uses of funds. Finally it gives a recommendation: this investment doesn't meet the return threshold and should not proceed. The whole process takes about ten minutes. In today's work, all the same underlying analytical capabilities (building analysis, running scenarios, creating outputs) are part of my and my team's daily workflow. ### Codex: Data Analysis & Automation I'm really excited about Codex because it puts the power of coding and software engineering into the hands of non‑technical people. #### Marketing Spend Dashboard We spend a lot on marketing and get huge amounts of detailed data from marketing agencies (geography, channel, keyword). We don't lack data; we lack time to process it and extract insights. We simply upload the data to Codex and say, "Build me an ROI dashboard." The result lets us quickly pivot across channels, see how much we spend on each, and where ROI declines due to saturation. We also ask Codex for the top five recommendations for reallocating spend based on the data. With this dashboard, we rebalance marketing spend weekly, moving money from under‑performing channels to high‑performing ones. #### Sales Insights After launching a new product, we wanted to know whether sales reps were actually talking to customers about it. The traditional approach would be to create a new field in the CRM, have reps enter data daily, and then someone downloads it for analysis. With Codex, we go directly to where the insights already exist: Codex extracts interactions from Gong conversation transcripts and customer emails, then creates a dashboard that shows exactly which reps are talking about the new product – by region, by segment, and down to the account level. Well before the quarter ends, we know whether we need to redeploy resources or create a dedicated team. #### Presentation Automation Codex can code – it writes front‑end interfaces like a front‑end engineer – so it automates many of the presentations we build for executives and the board. For example, the slide showing margin calculations for the CFO each month. Before, we had to manually reconcile raw infrastructure telemetry across different products and GPU types, apply accounting rules, and then create charts manually. Now, with Codex and reusable skills, the entire workflow is compressed to a few hours. We still have people review the data, set up quality checks and evaluations, but we save days each week and month. ## Organizational‑Level Agents We have built agents that are directly embedded in finance workflows, automating tedious, routine tasks and coordinating multiple systems to handle repetitive work. ### Procurement Agent Procurement and travel questions used to be handled manually. For example, "I'm going to London on a business trip – how much can I spend per night on a hotel?" Now the procurement agent handles about 60% of these questions automatically, and gets better over time. ### Credit Check Agent For customers who exceed a spending threshold, credit risk analysts used to research each customer one‑by‑one and extract information to create a composite credit risk score. Now the credit risk agent does it in minutes and embeds the score into the CRM, so sales reps can easily evaluate whether to work with that customer. ### Contract Review Agent We receive agreements in bulk, structure the data, and look for non‑standard clauses. Previously, the accounting team had to read every agreement to ensure no clauses would affect GAAP or ASC 606 revenue recognition rules; the trading desk also had to manually review. Now the contract review agent processes them in bulk and only flags exceptions. As the volume of agreements grows, we don't need to scale the accounting team proportionally. ## Conclusion By embracing AI natively, demonstrating human leverage, and deploying early and iterating, we have fundamentally changed how finance works with AI tools and agents. We accomplish more with a smaller team and continue to build the finance team of the future in real time. --- Source: OpenAI on OpenAI: Stacie Faggioli, Business Finance Officer Applications, OpenAI (https://www.youtube.com/watch?v=1NtS2KdnDok)