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

Hi everyone, I'm Stacy Fajoli. I lead finance for applications at OpenAI. Um, over the last couple of years, I have seen firsthand how my team has changed the way they work with AI. And so I'm really excited to be here today to just share a couple of examples with you. So um, in addition to managing the financial performance of the business, one of my OKRs at OpenAI is to build the finance team of the future. And so what exactly does that mean? Um when my team and I think about how we make the decisions to manage toward that objective, we really try to adhere to three principles. So first is we want to be AI native by design. What does that mean? As many of my colleagues have already mentioned, what that means is when we deploy a tool or an agent, we don't don't just think about it as a bolt-on to our existing processes. We really try to fundamentally reimagine how the workflow can be done. We really think about how we propagate the best practices not only throughout our team but across our cross functional stakeholders and also increasingly we're rethinking how we hire and design our org chart around our agents. The second principle is um we have to be able to show headcount leverage. In a very recent assessment that PWC did on our finance team and our operations, they have found that our finance team size is about 20% the size of our comparable technology peers. And so we are living proof that you can in fact do more with less with the right technology and the right tooling. And the third principle would be deploy early and iterate fast. Again, you don't need me to tell you that our technology is changing super quickly and even we feel it on a day-to-day basis. And so, whenever a decision comes up about whether we should wait for a more stable or more final version of AI, our bias has always been to just deploy it and keep iterating. This way, the team is able to grow very organically with our technology. And I feel like in real time, we are creating the finance team of the future. Uh I'll give you a little bit of context as to how we're structured before I go into some of the examples. So our finance organization is pretty simple. It has three different pill three separate pillars. Um the first pillar is what we call strategic finance. This is the team that's responsible for allocating capital, raising capital and also planning our growth. Uh the second pillar is what we call finance operations. So for example, accounting sits in here. This is the team that's really the engine of finance. They are responsible for collecting revenue, paying our bills, filing our taxes, closing the books on a monthly basis. And the third pillar is what we call enterprise financial technology or enterprise fintech. Um, this is the team that take looks after our finance systems and our data platforms. Okay. And our customers often ask us where do you embed or put the team that's building your tooling or your AI agents for you? Um on the finance team we have chosen to embed engineers directly into finance by sitting in the financial technology pillar. So what we have found is that by having these engineers sitting side by side with the finance subject matter experts we can speed up our deployment and iteration as opposed to you know me sitting over here in a corner and waiting for my colleague in the IT department to build something to my requirement and this way our tools can evolve in real time alongside our work. Okay. So in terms of actual deployment, I broadly think about it in two ways. One is how do we deploy tools to increase individual productivity and how do we also deploy agents to help um the entire team become more efficient. Uh on the individual side, I'll talk about a few different examples. Um so we use tools such as chat GPT that we all know and love. We use our chat GPT for Excel plugin. uh and more recently we've started becoming power users of codeex to really um drive efficiencies on the team. Let's start with chat GPT. So the investor relations agent was something we built with chat GPT. Uh this is a personal favorite. I love this example because it was so simple to implement. I'm a non-technical person and I could do this. Uh but also it is extremely high in ROI. So um in my time at OpenAI, I've been super fortunate to be part of two historic equity raises. Last year we raised 40 billion uh from the private markets and just a few weeks ago we announced we closed 122 billion in financing. Super proud of the achievement but as you can imagine behind the scenes it was not always pretty. Uh the team was absolutely inundated with diligence requests from investors. And so what we did was uh we trained a investor relations agent on our internal data as well as on the tone of a public company grade investment relations professional. And what this enabled us to do is when we were getting just sort of absolutely inundated with uh diligence requests, instead of assembling the responses manually, we were able to use the agent to provide data grounded, factual, consistent, strategically framed answers to our investors in minutes. And what's also important is because we only have one CFO, everyone on the team with the help of the investor relations agent was able to provide CFO grade answers to our investors. So we were able to ultimately run both raises entirely in-house and um I don't need to tell this audience that means hundreds of millions of dollars uh in in savings in terms of advisory fees. The other really interesting thing about our investor relations agent is I think it has really helped the finance team kind of scale our influence. What this is ultimately doing is just helping us sell our equity story. Turns out there are a lot of other places in the company where we have to sell our equity story. Every time we try to close a senior executive whether it's in research, whether it's in engineering, whether it's in product and they get a huge equity compensation package, they want to understand what is this equity worth? And so we have been able to share the investor relations uh agent with our recruiting teams and with our executive other executives to help them tell that same story that our CFO would tell. Okay. The second example um Chachi PT for Excel. We cannot have a finance discussion without talking about ChachiPT for Excel. Um as my co colleague Stephanie had already mentioned it really provides incredible leverage because to me it thinks like a human analyst. When you give it a task, it will first take a step back and think about how it would structure the analysis and then it would go into the workbook. It will write the formulas directly into the workbook so that at the end when I get it back, it is traceable. It is auditable and I can toggle assumptions to do scenarios. Okay, so let's do an example. Okay. So, um, when when I was thinking about what example to put together, I I decided to do this one because, uh, as a former private equity associate, I can still remember the all-nighters that I pulled, uh, building leverage buyout or LBO models. So, let's imagine I'm like back being 25-year-old Stacy uh, and I am being asked to analyze the take private of a publicly traded company called Openesk. So the first step is I would go pull up an equity research report that has some financials. Now with chat GPT the workflow is just completely transformed. I can open up so this is my Excel. Now open up chbpt for Excel. I can simply upload the PDF report into um chat GPT for Excel. I and I I'll just tell it think like an investment professional, lay out the financial projections and build me an LBO model. And that's it. And so here you can see Chad GBT4 Excel starting to plan the analysis and lay out the assumptions. Um here it is thinking and structuring the the project and the tasks and then it starts to populate the spreadsheet. And what I find incredible is um not only does it do exactly as I asked, it makes the projections, it builds an LBL model, it has very sophisticated assumptions and inputs, you know, very uh very detailed assumptions around the cap the cap structure, the cost of capital for each type of uh for each trunch of equity and debt and it even has sources and uses. So all of the outputs that I would want to see and that I would be building as an investment associate and just like a good investor it even makes a recommendation at the end. So in this case it decided that you should not proceed with the investment because it does not meet the return thresholds and um of course this was just a speedrun of what ChachiBT was actually doing. If you were following the time stamp it took about 10 minutes for Chachi BT to do all of that. And so, uh, in my day-to-day job today, obviously, I'm not looking at LBOs's anymore, but all of the same underlying analytical capabilities in terms of structuring analyses, running scenarios, creating outputs, our day-to-day workflows for me and my team. Okay. And next, I wanted to talk about codeex for data analysis. Uh, I know I've used the word exciting many times, but truly I am so excited about Codeex because it has put the power of coding and software engineering into the hands of non-technical people like myself. Um, there's so many use cases across data analysis and automation. So, let's dive in. Okay, so this is um this is actually a marketing spend dashboard that my team actually uses. Hence, we had to blur out some of the numbers. But the situation is this. We spend a lot on marketing uh and we get tons of data from our marketing agencies. We get very very detailed data on our spend across geographies, across channels, across keywords. Um we have no shortage of data. What we do have a shortage of is the time to actually crunch through all of that data to extract insights. So what did we do? We just uploaded all of that data into codeex and we said build me an ROI dashboard. And so this is what um codeex ultimately came up with. And here it allowed us to very quickly toggle between the different channels and start to understand our channel our spend for each channel and when we start to see the ROI diminishing because the channel is getting saturated. And further we ask codeex to give us five the top five recommendations for where it would um reallocate the spend based on the data it's seeing. So because we now have the power of this codeex dashboard, we are rebalancing our marketing spend on a weekly basis, moving our dollars from the less efficient channels to the more efficient channels. And this has been absolutely a gamecher for us. So another example we built with codeex just to go in a totally different direction is to get sales insights. Um we recently launched a new product and we started to ask the field to sort of push it to our customers and uh my CRO and myself you know we were very interested in getting a leading in some leading indicators on how this is going in the field. Are our sales reps actually talking to our customers about it or are they just more comfortable selling the legacy products? And um you can probably imagine the the kind of traditional way of doing this is you go into your CRM, you create a new field called product X or whatever. Um you make your reps enter data every day about when whether they talk to a customer about it and then you have a human download that data. You have a human analyze that data. With Codeex, we've what we were able to do is basically go to where all of these insights are already living. So, Codeex is pulling um the interactions and the insights from our Gong transcripts and from customer emails, our emails between our customers and our reps and creating a dashboard that tells me exactly by chair, by segment, by geography, all the way down to the account level, which reps are talking to our customers about these new products. And so, way before the quarter is over, we can start to get a sense of, you know, how are things looking? Do we need to redeploy resources? Do we need to, you know, have um a team that specializes in this product really kind of assist our existing reps in pushing the product? And so this again has been a really uh very helpful tool for us. Okay, last codeex example. I could keep going but um the other amazing thing about codeex like I mentioned is it can code which means it can code front-end interfaces just like a front-end engineer would means it can automate a lot of the presentations that we have to do for our executives and for our board. So here's an example of a um a compute margin slide that we show our CFO uh every month. So again the the slide is real. The do the numbers are made up so we don't have to get too excited about it. Uh but before codeex creating this slide was just this highly manual process. You can probably imagine we have to reconcile raw infrastructure telemetry data across different products across different types of GPUs. Uh and then the accounting team has to overlay the accounting rules and how we allocate certain clock costs and so on and so forth. And then you have a human kind of make m it all into pretty charts. Um and with codeex uh we basically with codeex and reusable skills we have basically compressed that workflow into a few hours of work on codeex. It's important I would say to emphasize that you know of course we're not just shipping this directly to the CFO. We still have folks on my team look at the data. Uh we create QA right we create evals to make sure we're kind of stress testing the numbers to make sure they pass our sniff test and they look accurate. Uh but in general it has just saved us probably days of um time every single week, every single month. Okay. So um those were just a few example of in a few examples of individual productivity tools. Uh I'll also talk a little bit about how we use codecs um at the organizational level. Um we have built agents that's embedded directly in many of our finance workflows. These automate away a lot of the very tedious and routine work. Um and the agents are able to basically orchestrate across lots of different systems to take on repetitive tasks. So uh I'll just give you a flavor of this. There are four examples I wanted to talk about. So the first one is procurement. We have a procurement agent. It's quite simple. Uh but procurement and travel questions used to be handled by humans. You know I'm traveling to London. How much money can I spend on a hotel every night? But now the procurement agent will can deflect about 60% of these questions automatically. Um and it's getting better and better over time. We have a credit check agent. So this is for our customers who spend above a certain threshold. Uh in the past we had credit risk analysts literally research every customer um and pull information on them and to create a composite sort of credit risk score. And today we just have a credit risk agent do it in minutes. And we also embed the scoring directly into the CRM system. So it's really easy for our reps to see at one glance uh whether we can keep moving forward with a certain customer. Okay. Contract review agent. This is also very cool. Um the contract review agent basically ingests um agreements in bulk and it structures the data and it looks out for non-standard terms. So, you know, for example, you know, we have lots of agreements with customers every day. In the past, the accounting team would have to read every single agreement to make sure that there are no non-standard terms that will uh cause them to change how they recognize the revenue under GAP, ASC 606, revenue recognition rules, etc. Um, our deal desk would also have to manually review the agreements to make sure we're not signing up for anything that's too non-standard. And now with our uh contract review agent, we can basically do it in bulk and it will just flag these insights for us. And so as a result, you know, as our volume of agreements grow, we don't have to scale our accounting team linearly in uh in relationship to that in order to continue closing our books on time. And more importantly for our customers, our deal is able to give them responses much more quickly. And uh finally, vendor risk agent. So again as you can imagine uh re risk reports used to take a lot of manual research and manual compiling to do. Uh we have a vendor risk agent that does that and again we embed the vendor risk score directly into our procurement software system so that whoever is um in the approval flow can see that risk score and e and either proceed or just escalate it. Okay. So I think that was it that I just wanted to share a few examples. That's a very quick look at how we're running finance and open AI. Um the last thing I just wanted to say was I truly believe you know success comes from an AI mindset. Before we do anything that's challenging, we are now conditioned to think okay what can chat GBT do or for me or how can I use chat GBT to make this task easier? And so you always approach every problem from that angle. And the other thing I would say is um this is a picture from the team's hackathon. What we did was just democratize access, right? Put chachi, put codeex into the hands of everyone. Um I think it's really interesting that you know none of those things I just talked to you about were my ideas. Zero. You have to put the tools into the hands of the folks who are closest to the problem, who are in the data, who are plumbing the systems every day. And I think you will be really surprised at the innovations that they come up with. So with that, uh, thank you so much and I'll turn it back over to Matt. Thank you.