# OpenAI’s controller on how AI can transform accounting Brex’s CAO Erik Zhou is joined by Sowmya Ranganathan, OpenAI’s Controller to discuss the different ways AI can make the accounting process more efficient. **URL Source:** https://www.brex.com/resources/controllers-classified/episode-05 --- Episode 05 OpenAI’s controller on how AI can transform accounting ### Spotify ![spotify](https://brand.brex.com/transform/556c1ade-18d0-479e-9ebc-c18d2e847315/spotify) ### Apple ![apple](https://brand.brex.com/transform/09f8bfeb-a8bf-4474-b6af-6e94e0082b59/apple) Episode summary On this episode of Controllers Classified, Brex’s CAO Erik Zhou is joined by Sowmya Ranganathan, OpenAI’s Controller. The conversation begins with a discussion on Sowmya’s career and her diverse experiences as an auditor and as a finance leader at both public and private tech companies, including Square and Rippling prior to OpenAI. Specifically, she highlights some of her insights for accounting during periods of business hypergrowth (i.e. when OpenAI launched ChatGPT) and the increasing need for technical savviness even as a finance professional (ex: Sowmya learned SQL as a way to manage the processing and analysis of a large data set of transactions at Square). She concludes her career overview with this advice: accounting leaders need to consider where their processes would break down if they were to grow quickly and focus their time and effort on making sure those processes are as scalable and automated as possible (hint: it requires building relationships with engineering). Sowmya and Erik then deep dive into the need for a tech-first approach to accounting that embraces AI to find efficiencies in the financial close process. Sowmya outlines how she thinks AI may evolve the job of accountants and the future skill sets that will be required as a result. She also spends some time considering audit implications and considerations as teams embed emerging technology into their workflows and systems. This episode is a must-listen for finance and tech enthusiasts. Finally, the episode closes with Sowmya’s bookkeeping fiasco at a previous company! Show notes - Joining OpenAI at the perfect moment [01:30] - Navigating different accounting worlds [03:46] - Navigating hypergrowth [14:40] - Using AI to create accounting efficiency [22:35] - Incorporating AI into financial tools & systems [33:20] - Sowmya's bookkeeping fiasco [42:03] Guest bio ### Headshot ![Sowmya headshot](https://brand.brex.com/transform/300862d5-c880-4af0-9dd5-a6cc83773600/Sowmya-headshot) ### Bio Sowmya Ranganathan, Controller, Open AI As the Controller at OpenAI, Sowmya leads the accounting function of a world-class research organization that aims to ensure that artificial intelligence is aligned with humanity's values and can benefit all of society. She has over 14 years of experience in financial analysis, accounting, and auditing, with a proven track record of building scalable processes, ensuring compliance, and delivering accurate and timely financial reporting. She has a Bachelor of Mathematics with a Finance option and a Master of Accounting degree from the University of Waterloo, and she is a Certified Public Accountant (California) and a Chartered Professional Accountant (Ontario). “I'm excited because I hope that [AI] will be the thing that makes all the manual and routine work of accounting go away and really free up [our] time to engage with the rest of the business. Maybe close as we know it goes away.” ### Transcript [00:00:00] **Sowmya:** AI is going to advance, like, really quickly, right? I often catch myself in the same thought process that we are in right now, and then I have to remind myself, like, maybe it can't do A, B, and C today, but tomorrow, or in six months, it might be able to do this. So, like, I think we kind of have to keep evolving what our role looks like, uh, and what our team does, and maybe teams will look different, right? [00:00:23] **Sowmya:** You don't need the same roles we need today, but maybe you need different goals. [00:00:27] **Narrator:** Welcome to Controllers Classified. The podcast where we take a deep dive into the dynamic world of controllers, accountants, and finance leaders, and hear how their ever evolving roles are redefining accounting and the future of business. [00:00:44] **Narrator:** And now, here's your host, Eric Zhou. [00:00:49] **Erik:** Welcome to Controllers Classified. I'm your host, Eric Jo, Chief Accounting Officer at Brex, and I'm super honored to welcome Soumya Ranganathan. Um, she is the controller at what is arguably one of the hottest tech startups out there in the world today. Uh, OpenAI, and Soumya has deep accounting experience in the tech industry broadly, spending time both as an external auditor and as a controller. [00:01:16] **Erik:** Welcome, Soumya. Super happy to have you on the show today. [00:01:19] **Sowmya:** Thank you for having me, Eric. Super excited to have this conversation with you today. [00:01:25] **Erik:** So I think the first question is You got to tell us about OpenAI. And right there, the creators of chat GPT for anyone who's maybe living under a rock. Um, and I, I know you just also started there about six months ago, so I'd also love to understand what your focus areas are going to be in the next six months going into the new year. [00:01:45] **Sowmya:** I joined OpenAI about six months ago, interesting timing because obviously like I was also just like most other people floored when I saw what ChachiPT could do, I immediately started thinking about what does this mean for accounting and I started kind of Doing a bunch of like kind of side projects to see, you know, how we could use ChatGPT in our work. [00:02:08] **Sowmya:** And then of course the controller role opened up here and, you know, I, I started the first month and closed when we had just launched ChatGPT So as you can imagine, it was just. Period of like crazy hyper growth, like champagne problems, definitely, but like still problems for clothes when like everything breaks in the same month. [00:02:27] **Sowmya:** But say, you know, six months, fast forward, like we've obviously built more structure. You know, my focus is essentially like, as a controller here, I oversee revenue accounting, opex accounting, payroll, finance data. Those are broadly like the big buckets. And of course, like our accounting for compute and infrastructure, which is like also big. [00:02:47] **Sowmya:** Component in the P& L in terms of focus areas for the next six months. It's honestly like get through a clean audit for this year, but also like build a good foundation for the next phase of growth. As much as we love the champagne problem world, like I would really like to not be in a problem world. [00:03:06] **Erik:** 100%. [00:03:07] **Erik:** So one of the things we'll get into all the stuff you've been doing with AI for your team. I think that's super interesting. But just career wise. You were a public accountant, you ended up working at Square, which I believe you even joined right before it went public, or just as it was going IPO, and now you're back in kind of like, or you're doing private company accounting. [00:03:29] **Erik:** What did you feel was the biggest difference, kind of transitioning from auditor, now you're at the public company, and then going from the public company, now you're private, like, I, I feel like those are all different worlds, I'd love to kind of, Hear from you what those experiences were like. [00:03:46] **Sowmya:** I feel like i've gotten a lot of lucky breaks throughout the course of my career And like I definitely found myself in the right place at the right time many times but particularly So I actually loved audit when I started and I thought I was gonna be an auditor forever And I thought that was my jam and like a few years in I actually moved from toronto to san francisco and And, uh, I, I got much more into the world of tech companies and actually like that was my first software client after I moved here until then I saw it at banks and investment companies. [00:04:19] **Sowmya:** And once I saw this industry, I, I was definitely like an early user of software. Square actually, as a merchant, I had a little like hobby business, uh, in college and I, I used Square and their reader to accept payments as a merchant and [00:04:33] **Erik:** the smaller reader that you attached the phone smaller, [00:04:36] **Sowmya:** and you have no kidding. [00:04:36] **Sowmya:** Okay, you have one time. This is like, you know what, like 2009, 10 when it first came out and it was game changing because like as a small business vendor, you know, people could now like. Pay you a lot more because you're not limited to how much cash they're carrying on their wallet. And so I love the product and I actually like, I wanted to work at Square much more because of the business and the company and what it was doing in the world and, and accounting was just like, well, I'm an accountant, so I guess I'll apply for a job in their accounting team. [00:05:06] **Sowmya:** That being said, it ended up being a really good time to join the company. It was like a year before they went public. There was a lot of pre IPO work to be done. I call it like a little bit of a bootcamp. It was my first. Like, real industry job. I, I did a small stint at a startup before that, but you know, that was a much simpler business model. [00:05:23] **Sowmya:** Square was definitely like going back to bootcamp and I was, I had to learn a lot of things really fast. [00:05:28] **Erik:** When you're going through the bootcamp of that IPO readiness, what was the biggest surprise? [00:05:34] **Sowmya:** Honestly, I think coming from Audit to Square, I think the biggest surprise for me was how much more like data and SQL savvy and tech savvy I have to be as an accountant. [00:05:47] **Sowmya:** And I think I, I kind of sort of knew it, but I hadn't internalized it until I took that job. And, you know, we were obviously dealing with like a very large team. data set of transactions. We often had to sit down with auditors and explain like, what does this data table represent? And my skillset had to basically take a huge step function growth in a very short time period. [00:06:11] **Sowmya:** That being said, I actually had a little bit of a I had a quarter life crisis when I was at Square and thought I would quit accounting altogether and like do coding bootcamp. Uh, I'd done some programming in college. Like, I think I like this stuff a lot more than accounting. But anyways, uh, my husband and a few other close friends talked me off the ledge and they're like, [00:06:29] **Erik:** you should How did they talk you off the ledge? [00:06:31] **Erik:** Like, what was the [00:06:32] **Sowmya:** I think my husband's a software engineer and he's like, you know, it's not what you think it is. And I think you should probably just like Keep on the path you're on. [00:06:40] **Erik:** Grass is not greener. You're, you're looking at a screen, but instead of in a spreadsheet, it's lines and lines and lines of code. [00:06:47] **Erik:** Because I've seen, I've seen some of those screens on the engineers at the company here. [00:06:50] **Sowmya:** Totally. But actually, uh, what he encouraged me, and I'm really grateful for this kind of exposure, is I actually took some like intro, you know, to like Ruby programming workshops in SF. There were these meetups that would happen and I just kind of like, Learned programming, uh, you, you build like a, you know, scheduling or to do list app. [00:07:10] **Sowmya:** It's like a very quintessential starter programming app. I also like try to build an app for our Alexa at the time to tell me when the next meeting was going to come. One of his friends had built something similar and I kind of built off of her code. So I think tinkering with programming on the backend definitely helped increase that tech skillset, but I'd say. [00:07:29] **Sowmya:** That has helped me many, many times over in all my jobs. So it's like, I'm an accountant in my day job, but I feel like the technical skillset is a nice back pocket thing to have as an accountant. [00:07:41] **Erik:** So this is something I experienced when I, cause I was at Pricewaterhouse for 11 years and I, I move off of the Brex. [00:07:47] **Erik:** This is still my first industry job. I've been at Brex for almost five years now. Um, there's a gap. There's a total gap between like how auditors are trained. And we're the industry and people in the field, like in accounting operations in my team, where we're going and how we're trying to improve our stuff. [00:08:05] **Erik:** And so, yeah, you know, I got folks that look at SQL. I got folks that are a little bit more advanced in some of the hard skills that that you might expect from an accountant. And, and was that a problem dealing with, with your auditors at Square? Like sometimes what happens is you're doing something so novel, the auditors can't get their heads around it, right? [00:08:28] **Erik:** And so it makes the audit more difficult because they don't fully understand the process innately. Did you experience that at Square? [00:08:35] **Sowmya:** We had a really great audit team, honestly, and they were extremely engaged. And we, we definitely sat down with them many times during the course of the audit to kind of explain, okay, like, let's do, Uh, let's take a few transactions or a few types of transactions. [00:08:51] **Sowmya:** Let's walk through like what happens throughout the life cycle. And there was a lot of like collaborative effort. It definitely wasn't like, I'm going to dump some PBCs in this folder and off you go. Cause I think if we did that approach, it would have just been like really painful for both sides, to be honest. [00:09:07] **Sowmya:** But I agree. And honestly, like the audit team pulled in a lot of technical resources as well. [00:09:13] **Erik:** So would you say your auditors at the time may have even recognized that there was some gap and they, and they brought in specialists from whatever group internally, uh, to help. That, that, that makes sense. [00:09:25] **Erik:** That's good. It's good to hear. You know, one, one other thing before we move on to the AI stuff, you, so you were at Square public company. Now you're at, you were at a private company before this at Rippling. Now you're at OpenAI, also a private company. Was there any step down in process? Was there any kind of like reprioritization of your day to day? [00:09:47] **Erik:** Being a controller or being an accounting, uh, lead at a public company versus private. [00:09:54] **Sowmya:** Absolutely. I think like at a public company, you know, your SOX checklist and your RCM drives a lot of like what accountants end up doing. And, uh, SOX adoption was a big part of my job at Square and not just my job, like the whole team, we were like living and breathing SOX for a long time. [00:10:11] **Sowmya:** Um, And then coming to a private company, the big difference is as accountants, we, we really like focus on the business and you see like what. Does the business need first and foremost, and you kind of augment the focus areas of accounting to that. And I'd say like at Rippling, we did similar stuff, like depending on, you know, new business models that will launch, that we're launching, uh, compliance challenges, uh, and, and like reporting, where do we need to bolster up reporting cycles so that we can get compliance reporting out without, you know, a mad scramble every quarter. [00:10:46] **Sowmya:** So like things like that. Um, and, and similarly here, like a lot of our process and controls is very much like self imposed because you're like, you're not in a socks world. No one's telling you you have to do this, but you're like, having a good control environment is the only way we're going to feel good that the numbers we record on are good. [00:11:07] **Sowmya:** And I'd say Hi. Bye. You can probably hear the bias in my voice, but I think it's a lot more fun personally for me to be here where, uh, you're much more kind of focused on the business and the growth and supporting non accounting teams in the business, um, without the pressure of, you know, we need to adopt SOCs like today. [00:11:28] **Erik:** So there is the practical elements of implementing SOCs in my, it is my opinion, but you implement SOCs, it has to get audited, right? People, your auditors are now doing tests of controls, et cetera. And you have to do all this work in preparation for that. There's also the spirit of SOX. The idea that you should have good control environment, good tone from the top, be able to have faith in the numbers that your team is producing, etc. [00:11:56] **Erik:** Um, and then to hit the spirit, I think, you know, that's more like, uh, just having pride in your work. Making sure someone's double checking it. Making sure there's the right checks and balances on this or that thing. And it doesn't necessarily have to be as formulaic sometimes as SOX is. You, you know, question for you. [00:12:16] **Erik:** If Rippling or OpenAI, if they were to be under SOX tomorrow, what would you say is the uplift in terms of hours and resources? Like, what is it, like 30%, 40%? [00:12:29] **Sowmya:** Yeah, probably north of 50%, mainly because, like, I know that's significant, right? Mainly because I'd say we tend to heavily prioritize the things with high impact, but that means things like documentation gets deprioritized. [00:12:45] **Sowmya:** I'd say at OpenAI, we've been like forced to get mature a little bit faster just because of the hyper growth. Uh, at, at, at Rippling too, like in the early days, You know, I, I was the first accountant and I learned a lot about the business, but I also had this like pretty messy instance of QuickBooks to clean up. [00:13:02] **Sowmya:** Uh, I, I can share some like craziness later, but, um, but you know, taking it from the early days, like if I had gone to Rippling when we were a pretty tiny startup at the time and gone in, like tried to implement SOPs, it would have totally been like, you know, read the room, like this is the wrong thing to do at this moment. [00:13:20] **Sowmya:** And as the company grows. And as the complexity of the business grows, like we always prioritize, what do you need to do in that moment or in that quarter to like help make the business successful? And I'd say, no, fast forward a few years, like I wouldn't be surprised if the team has actually been investing a lot more in the control environment and SOCs because it, it's not easy. [00:13:43] **Sowmya:** You know, there's a degree of maturity that comes with growth and, you know, SOX is part of that. I'd say when you actually do a SOX implementation, everyone's always surprised at how much more there is to do. So I'd say 50 percent is actually probably more realistic than people realize. [00:13:58] **Erik:** Yeah, it's a lot of work for sure. [00:14:00] **Erik:** And the documentation that it takes time. It takes time to do it consistently day over day, week over week, month over month. Uh, it takes time. Um, all right, let's get into it. So you guys launched ChatGPT. Reports are saying over a hundred million individual users. Hard to know, um, how many of those are actually true subscribers, meaning pay for the monthly service. [00:14:25] **Erik:** It's like 20 bucks a month. You guys have the fastest growing consumer application in history based on those numbers. And so I don't expect you to share like exactly how many subscribers you really have. It's not, you know, that's not what's important. Um, but you had a massive increase in revenue operations. [00:14:44] **Erik:** And so, how did you guys manage? That transition, because it was like two months that it got to this level or something. And so I can't even fathom how, uh, heavy the operational burden might've been to accept all those payments, to do all the work, et cetera. Just curious about your experience there. [00:15:03] **Sowmya:** Absolutely. And, and, you know, we haven't confirmed any of the numbers publicly, but it definitely was a period of Amazing hyper growth in user base, customer base. And I'd say we got really, we're really fortunate to have a revenue team built in here. And this is honestly like months before I joined as well. [00:15:23] **Sowmya:** But like we had a revenue team with already a very strong technical skill set. We were already like partnering with engineering to automate a lot of the billing transactions. We, we use Stripe here, uh, and, and, You know, the piping of usage based billing, like the internal usage tables to like, Making its way to actually generate an invoice in Stripe. [00:15:45] **Sowmya:** There's no way, even at the pre Chad GPT scale, it would have been really difficult to do this manually. Definitely post Chad GPT, it would have been just impossible. Like we would have just come to a grinding halt. And I think this is a good thing for accounting teams in general is to like, look at the things that are, you know, if you, your company would have grown like crazy, where would accounting break and I think injecting Engineering, uh, help into those parts early and frequently kind of revisiting this often is the only way to go. [00:16:20] **Sowmya:** Part of building out accounting functions at companies like. This one is, you really have to pre product manager a lot more than you realize as an accountant. [00:16:32] **Erik:** What do you mean by product manager? You're like advising the eng team on how to make your own job easier with their help? Or how do you, how would you describe it? [00:16:39] **Erik:** I [00:16:39] **Sowmya:** think you kind of have to think about like, When you're building out an accounting function, you're building out a team, sure, and you're executing on accounting, but you're also building products, I say in quotes, that face your team or customers or someone else. And, you know, you, you kind of think about it like a PM would. [00:16:58] **Sowmya:** You, you say, what is the problem we're solving? Who are, who are the customers for this problem? If we didn't solve this, what's going to break? And then you actually define like, What is the, what is your ideal dream state experience? Like if this were working perfectly, what data would flow into what, what would be the outcome here, like maybe automated notifications, automated reporting, what have you, right? [00:17:19] **Sowmya:** Depending on the problem you're solving. And I think being able to articulate that and actually like build those partnerships with engineering to make the case to say, Hey, this is going to like be nuts. Much better for the company if we actually like put this investment in that. And, you know, people aren't like waiting in line, trying to help accounting. [00:17:37] **Sowmya:** You always have to like make the case. And I'd say there's like gives and takes. Pitch Rippling, one of the things I love doing was actually like working with engineering, um, and product on our own products because a lot of them were finance user facing, I guess, similar to you. You're in all the Slack channels, giving them feedback and I'd say like building relationships like that. [00:17:55] **Sowmya:** I did it for its own sake, honestly, I loved being in that world, but you also get to know names and faces and know who to, you know, whose shoulder to tap on to say, Hey, it's Hack Week, let's go build something that might actually help my team. I think having a tight connection there really helps in the long term. [00:18:13] **Erik:** Going through the hyper growth, there's different processes for revenue. There's collections, there's booking the initial journal entry based on revenue recognition, etc. You can go down the list. I'm curious, like. What were the processes that you look to automate with, without AI first, maybe. And then what were the processes that leveraged AI and you were able to get a lot of kind of like support from the technology? [00:18:40] **Sowmya:** Yeah, that's super interesting. I'd say like the core processes honestly is much more like SQL skills more than AI skills to the extent that Somebody can ask Chatubidi to debug their SQL queries like you're using AI, but for the most part, uh, you know, you are still like comparing like, hey, what we sent to this provider or to Stripe, like, did that match what they actually received? [00:19:02] **Sowmya:** And like you're, it's a lot more data heavy, uh, in terms of the workflows. Um, where we injected AI, we, so you've heard of Clarity. [00:19:11] **Erik:** And for, and for our listeners who, you know, are thinking about incorporating just, SQL to start with, without the AI piece necessarily, like that SQL helps you, how, like, what, what, what, what, what is the process that SQL makes faster? [00:19:28] **Sowmya:** So I'll give you an example. We use Stripe for billing. So invoices is really like living in Stripe, right? And say you want to reconcile, we have an internal source of truth, which is like, What are our customers? What is the usage? You have invoices that live in Stripe. And then you want to compare all of that to what you booked in NetSuite. [00:19:49] **Sowmya:** We actually don't book transaction level in NetSuite because my worry is like, if you put all of those transactions in the NetSuite, it'll just choke it. It'll get swollen. [00:19:57] **Erik:** There's a line limit. There's a line limit in NetSuite. [00:19:59] **Sowmya:** Yeah, there's a [00:19:59] **Erik:** line. It's like 75, 000 lines or something, but there's a limit. [00:20:02] **Erik:** Yeah. [00:20:02] **Sowmya:** Exactly. And, and. You know, we, we essentially only do like aggregate level journal entries in NetSuite, so we, we want to build this like recon of internal data to Stripe data to NetSuite. Now you could do this in Excel where you download everything into CSV and you would like compare CSVs to CSVs too. [00:20:23] **Sowmya:** NetSuite, CSV, and you get where I'm going with this. I'm saying CSVA lot, but like almost, and [00:20:27] **Erik:** Excel has a also a million row limit, so, yep. [00:20:30] **Sowmya:** Exactly. Yep. So when you're, when you're in a business that has like a large number of small dollar transactions, this will just break. It'll break before you even start, right? [00:20:39] **Sowmya:** And, and so where SQL can come is you can build these queries that essentially do this comparison for you. With Stripe, for example, we, we have this thing called Stripe Data Pipeline, which allows you to like. You know, whatever you can run as a reported Stripe, it sends you in like underlying data tables. [00:20:57] **Sowmya:** So then you can write SQL queries on Stripe data without having to like work with their reporting and download CSVs. Similarly for internal data, we, you know, have everything in Snowflake and I think now Databricks. So having data savvy in the team has gone such a long way, right? And like, you, you, you. [00:21:15] **Sowmya:** You can do this without SQL, it will just be significantly more painful. And, you know, it's pain that's not one and done, it's pain that repeats. And I think my focus areas is like, you know, whenever I hear from people that like, Hey, this is something I do every month and it's painful, that immediately like rings a five alarm fire bell in my, in my head, because I'm like, this is never going to go away if we don't fix it. [00:21:38] **Sowmya:** And like, those are the things that we try to kind of clean up. As quickly as possible. [00:21:43] **Erik:** And then, and then what's a process that AI has allowed you to. Even further, um, automate's not the right word, but, you know, just make more efficient. [00:21:54] **Sowmya:** Yeah, I'd say one of my favorite AI applications internally is, uh, we, we're implementing Clarity. [00:22:00] **Sowmya:** We've, we used Clarity before. I think we're doing a re implementation now to do much more. But essentially, like, you know, we, we, we have a lot of Like in terms of deal desk review, Clarity can really help document like a, what are the nonstandard terms in a contract? Um, Clarity uses And what does Clarity [00:22:18] **Erik:** do? [00:22:18] **Erik:** What, well, like what, what, what, what does Clarity do for different companies, what is their offering? [00:22:22] **Sowmya:** I'm probably not the best person to pitch Clarity putting this out there, but in, in very simple terms. They, they have like generative AI technology underneath Clarity. So you can feed it a contract and it will read a contract and you can ask it to like kind of tabulate, you know, give me everything that has a TFC, like a termination for convenience, and it can basically like put that in a, in whatever format you need, ideally like a reportable tabular format. [00:22:50] **Sowmya:** So. When you have it, you can build like a much richer contract repository than just having all the PDFs live somewhere, right? In a way that you can then query for like, okay now give me everything that has non standard terms for a particular type of term and you're able to kind of kind of You can also say, hey, if everything looks standard and this is like our regular paper, then maybe you fast track that through deal review. [00:23:17] **Sowmya:** So you're not spending the same amount of resources. Um, you know, like a few weeks ago we had this like clarity panel and like, where do you automate and where do you make people superhuman? In my experience, where we are today with AI use is we're definitely in the place of making people superhuman and it's very tangible. [00:23:35] **Sowmya:** Uh, how much more people are able to do, uh, with having like a little AI assistant. The other place we've used AI pretty successfully is Honest, like in writing Python scripts for accountants. [00:23:49] **Erik:** What business does an accountant have with writing a Python script? Like what? Yeah, [00:23:55] **Sowmya:** it is so bizarre and like, it is not what you would think when you think of accountants, you're not seeing them run things in terminal. [00:24:03] **Sowmya:** So I'll give you an example, like we, we get these massive CSV files for our compute usage and like, we obviously want to track like where we're spending on compute, whether it's R& D or cost of revenue and like. So the marketing and, you know, we need to bucket it for accounting, but we also need to bucket it for the business. [00:24:20] **Sowmya:** We want to tell people like, Hey, this is how much like compute we're using for various projects. So there's a lot of reporting and filtering and pivoting that happens. It would be trivial to do an Excel if the files were small, but these files are like many millions of rows because of how granular they get. [00:24:37] **Sowmya:** So. And this is to compute [00:24:40] **Erik:** from OpenAI using like the GPUs that you guys have bought or, okay. Yeah. [00:24:45] **Sowmya:** Yeah. So it's like, think of it as like getting a window into. Where are we spending this money? And like, you can think of this as like, you know, any kind of like big capital investment, if you're a company, uh, that you might not use computer in the same way, but you're spending your money in something big and you want to know like where that's going, you could also use the same kind of logic for honestly, like Any large CSV file from tax to revenue to general purpose. [00:25:10] **Sowmya:** Um, so what we do, what we used to do because of the million roll limit, is you had to run like multiple files per month and then do this operation. Or you, you, you know, try to like, Download a big CSV file, run a program to split it into multiple smaller files, and then you do your filtering, pivoting, and stitch it all back together. [00:25:30] **Sowmya:** But it's like, one, it's just annoying work. Like nobody really wants to do this. And, you know, even like your best Excel loving accountant eventually just throws their hand up in the air and like, what is this nonsense? Like, why am I wasting my time on this? So what we as ChatGPG do is like to say, Hey, I have an input file that has these headers. [00:25:49] **Sowmya:** This is what I want to do with it. Now write this in a Python script where I can feed it a CSV file and out comes the output keyboard. And I think the most like interesting thing for me is like the interface is just English. You know, you're not, you don't need to know any programming or like technical skills. [00:26:08] **Sowmya:** You don't need to know SQL. You're, it's like you had a buddy sitting next to you and you're telling them this is what you need to do. And, except, instead of a human, you're saying this to chat to BTN, it generates this Python code. And now, like, what used to be this super annoying manual process every month is done in two minutes. [00:26:26] **Sowmya:** And it, like, two minutes is, like, probably an exaggeration. Because computers, turns out, are extremely efficient at processing. So we've seen that like you can apply this kind of like application in a lot of different ways throughout the accounting closed process. So we've, we're still exploring. I wouldn't say we've come to the be all end all of what AI can do for accounting, but I've been like pretty impressed with how creative people can get. [00:26:54] **Erik:** And is, and is leveraging Gen AI to write scripts like this a way to. You know, there's these issues that we've heard of that it can't do math. [00:27:03] **Sowmya:** Yeah. [00:27:04] **Erik:** Right, because it's text based and the math is like different, it's all probabilistic in terms of the text, and so. But then, if you're asking it to do a Python script, or a SQL query, or what have you, Um, does that, do you avoid those issues with math? [00:27:19] **Erik:** Because it's running a script, And the script is what's doing the math, not the AI itself. [00:27:24] **Sowmya:** Kinda. So I think there's ways to do this within chatGPT. So you get a small enough file. It's actually pretty decent at analyzing that data on its own. But what it's really doing in the background is it's creating a script or it's creating some kind of code that it runs on that source data. [00:27:40] **Sowmya:** What I've found And, and, you know, we haven't gone through an audit yet with all this stuff. So let's see how it lands with our auditors. But, you know, our auditors are also really excited about the adoption of AI in accounting. So hopefully, uh, we will come to a good place, uh, as we, you know, get these things through audit. [00:27:58] **Sowmya:** But in, in my opinion, it's a lot easier to tell auditors, this is the script that happens every single month for this process. You can get your technical people to look at this script, or we'll just give you the raw CSV file. If you have your own, like, kind of internal tools that you want to run this through, you don't have to use Python. [00:28:16] **Sowmya:** Like, you can just re perform this in your own way. But it's a lot easier to say that in an audit, as opposed to saying, I feed it into a chat GPT every month that it comes up. But I think it's a little bit more of a black box that's Harder to get people on board with. Maybe, you know, as an industry, we'll all get more comfortable with the black box over the next couple of years. [00:28:38] **Sowmya:** But I'd say today, and especially being open AI, I want to play it a little more conservative in how we use this. So we've landed on like the Python script, uh, as a middle ground. [00:28:49] **Erik:** Let's talk about, you know, something you mentioned related to making your team super humans, right? With the help of AI. You know, there's obviously been a lot of talk about how AI will influence jobs in America and beyond. [00:29:06] **Erik:** Um, it's obviously doing a lot to improve processes. I think systems will ultimately catch up, like we all work with different systems, NetSuite, you know, BlackLine, you know, uh, what have you. And they're gonna probably figure out how to incorporate gen AI into those systems themselves. But, but what's your take on maybe first what it's gonna do with people and like, sure, it's gonna make superhuman or give people superhuman abilities now, But over the long term, isn't it kind of inevitable that some jobs need to be repurposed or they're going to get replaced by AI? [00:29:40] **Sowmya:** I think there's no doubt that this is kind of a, you know, it's a big moment for what's happening to accounting as an industry. And I think I feel like you're kind of on one side of a mountain and you're not quite at the peak yet. And it's hard for us to know from this side what it's really going to look like on the other side. [00:29:58] **Sowmya:** And I think I've, I've heard similar talks before where we've used a similar analogy, but it's like, are all jobs going to go away for accountants? Probably not. Are they going to look the same as they did now? Also, probably not. And I think what we're seeing in the immediate term is I think there's going to be a pretty high demand actually for accountants that are embracing AI and putting it into their daily workflows. [00:30:21] **Sowmya:** Going back to like my lessons from Square where I was surprised at how much more technical I had to be in industry. I think that's probably going to ramp up even more, except now your technical skills are probably not excellent SQL. It's like, how do I. You know, like even knowing what Python or a script, like scripting language could do is going to put you ahead above the rest of the industry. [00:30:45] **Sowmya:** I'm excited to be honest, Eric, because like, I hope that this will be the thing that makes all the manual and like routine work of accounting go away and really free up time to Engage with the rest of the business and the, the people that are building the products and like helping deliver like better insights, maybe faster, maybe close as we know it goes away. [00:31:06] **Erik:** Yeah. I mean, historically, let's talk about the spreadsheet and automating the spreadsheet. You know, like when we automated the spreadsheet. I wasn't there, but people were like, Oh, job, like what's going to do with jobs? Like, we're, we're no longer going to need all these people to do all this work, um, at their desks and putting invoices in, et cetera, et cetera. [00:31:27] **Erik:** And I think what ended up happening was like. Accountants became more useful, and then you got more of them as a result. Because, right, because their skills expanded, and they could do more, uh, with that technology. And I'm with you. I think, I'm hopeful, certainly there's going to be, uh, folks that are getting a lot of, um, leverage from the technology in the near term, and then long term, we're not over the mountain yet, so it's hard to really know, but I'm hopeful that it's going to show the worth, worthwhile. [00:31:55] **Erik:** Uh, skills of an accountant and of all these folks in finance, um, and, and the profession will get a boost, actually. Let's talk about, uh, systems just, just a little bit more. You're on NetSuite, we're also on NetSuite. I haven't necessarily heard of anything about how Oracle is going to implement Gen AI. Um, and I don't know what other What are these other like kind of brand name companies are going to do about it? [00:32:20] **Erik:** Um, but what would you love to see, uh, as an incorporation of Gen AI in kind of like everyday systems like that in an ERP? [00:32:29] **Sowmya:** So actually, fun fact, when, before I joined OpenAI, like back in January, one of the side projects I was tinkering with was this idea of what if you could just talk to your GL, right? [00:32:40] **Sowmya:** And it, not in like a weird way, I know that sounds kind of crazy and something an accountant would say, but imagine your CFO, uh, wanted to ask a question about how much should we spend on this vendor? You didn't have to like run a GL detail report and pivot and like look for that. You could just. Ask your accounting system, how much did we spend on blah? [00:33:02] **Sowmya:** And it just gives you an answer. And like, it maybe gives you a CSV to support that. I think, and I hope a lot of the systems that we use today move to that model, because I think the whole idea of like, Reporting and filtering and manual stuff. Like it can all happen in the background very quickly because of where we are today with AI. [00:33:23] **Sowmya:** Um, and Eric, I don't know if you've come across Numeric yet. Um, Oh, that's awesome. Like, I, I think they're one of the products that I was seriously impressed by, uh, early this year. One, they just like, Got up to speed on implementing AI in their tool very, very quickly. But my favorite feature is their autoflux, where, you know, it ingests data from, uh, NetSuite and you click one button and it just like, writes a flux explanation for your PNL. [00:33:53] **Sowmya:** And it's, it's a little magical and I think, but I do hope that there's much more magic in the tools we use. Because of all this, like, I think over the last five, six years, we, we all as a content saw a huge upgrade to the UX of the tools we use. Uh, and now I hope that it extends a lot more to like, just doing more and, you know, requiring less, like, training on how to use the tool because it's just like talking to it. [00:34:21] **Sowmya:** Does that make sense? [00:34:23] **Erik:** And, and maybe it's, it's also a little bit of the incumbents may not implement it as much as we are thinking they might. Um, and it will be new players that do this work like with Numeric or with Clarity, etc. Um, and, and that, you know, but the thing is like, it's really hard to rip out an ERP, right? [00:34:43] **Erik:** And so maybe those companies start partnering. Maybe just to close off this segment, You know, one more question, is there anything in the finance and accounting world that you think AI won't touch or maybe won't be as successful at? [00:35:00] **Sowmya:** Um, my answer is probably going to be pretty biased to like people working in the startup and like the younger growth stages of companies. [00:35:09] **Sowmya:** I'd say the hardest accounting things I've had to work on, honestly, are Not the things I do in front of the computer, but it's like understanding what a new business model is going to look like, sitting down with your C suite and explaining how a particular decision might play out in your financials. [00:35:27] **Sowmya:** And, you know, a lot of times the questions aren't going to be okay. They're going to be like, why does it need to do that? What is the principle behind this? Is there a different way to do it? And I think you could, sure, you could use ChatGPT or similar tools to help you with the gap research, but, you know, A CEO is not going to like, be like, oh, okay, I guess I'm okay with that because Chachi said so. [00:35:49] **Sowmya:** I think there's this, uh, this layer of needing to kind of break accounting terms to non accounting people. And that might be one of the areas where I think You probably want a human in the loop, uh, it's, it tends to be really critical. You're maybe like negotiating a partnership agreement or something with a customer and like you, you know, the stakes are pretty high to not have a human in there is how I think about it. [00:36:15] **Sowmya:** Um, [00:36:17] **Erik:** what you're saying is this nuance of Just dealing with people, soft skills, relationship building, credibility, you know, like, it, like, if I think about putting ChatGBT as a replacement of myself, I don't, I don't know, maybe one day ChatGBT will be as credible as someone like you and me, but I think we're still pretty far away from that, um, and then they may not, it may, it's based on some LLM that may not have the specific context that you or I have. [00:36:45] **Erik:** And, and so, but, but that's interesting. Yeah. I think these timeless, the timeless art of people management, dealing with other people, living in the world, actually, like, I think, uh, that makes sense. [00:36:58] **Sowmya:** Yeah. I will say though, that AI is going to advance like really quickly, right? And so when I think of, like, I often catch myself in the same thought process that we are in right now, and then I have to remind myself, like, maybe can't do A, B and C today, but tomorrow or in six months, I can. [00:37:17] **Sowmya:** It might be able to do this. So like, I think we kind of have to keep evolving what our role looks like, uh, and what our team does. And maybe teams will look different, right? You don't need the same roles we need today, but maybe you need different goals tomorrow. Um, and maybe, you know, we, we all get more efficient. [00:37:35] **Sowmya:** Software development gets more efficient. Accounting teams get more lean. And as a humanity, you get to do a lot more faster. And, you know, it's certainly an exciting place to be. That's my only caveat though, cause like the advancements in AI is a little bit like, I feel like it's happening a lot faster than most of us. [00:37:56] **Sowmya:** Really grok today and as a non technologist, I, you know, I don't really understand the science and the math that goes behind it, but I see the impact of it. And I, I'm really cautious to say what can AI definitely not do. [00:38:10] **Erik:** I'm going to share, uh, one example of a use of chat GPT, and this is personal, so it wasn't, it wasn't related to Brex or anything at all. [00:38:19] **Erik:** But my dad, he runs a, a, Uh, he runs a restaurant in the Minneapolis St. Paul International Airport. And so when you get a sublease into the airport, there's certain data reporting requirements. And so he actually went through a point of sale system change. And what ended up happening was over the course of that month, The system stopped sending the daily sales reports to the airport commission, and they require that on the lease. [00:38:46] **Erik:** And so he didn't do that for a month, it was like 30 days, they fined him, 200 a day, 6, 000 bucks for a small business. That's deep. Yeah, it's not a small thing. And so, but my dad's been in that airport for over 20 years, and so he's telling me about it, he wants me to like help him fight the fine, I'm like, Oh, I have to write this letter. [00:39:07] **Erik:** I was like, there's all these, okay. So tell me what happened. What do you want to write in it? And I'm like, you know what? I'm going to have ChatGPT4 write it cause I'm a subscriber and I give it a prompt. I said, write a business letter. Uh, these are the three reasons why this fine should be waived. I kid you not. [00:39:23] **Erik:** It wrote this whole thing. I changed like 5 percent of it and I sent it and he got, he got the, he got the fine waived. And, and like this save, [00:39:33] **Sowmya:** this task, [00:39:34] **Erik:** like using ChatGPT literally saved probably 45 minutes of my time because I would have been, you know, trying to search for the right words in every sentence, I had the right tone, blah, blah, blah. [00:39:45] **Erik:** And, you know, for the most part, it nailed it. Like, this isn't an accounting related story. Um, it is, it is kind of like on the other end where you're dealing with people and how to communicate with people. But it really did a good job and that's when I got kind of sold. Like, oh, this is, this is going to change the world a little bit. [00:40:01] **Erik:** Yeah. [00:40:02] **Sowmya:** Yeah. You know what I mean? Yeah. I mean, that's a great outcome, but I agree with you. I've used it for all sorts of things. Like, um, Yeah, lots of research. Uh, recently, I've been really excited about the new Dolly release. And my daughter loves, like, she's five, and she loves coloring and artwork. And, you know, I used to, like, keep looking for, like, weird coloring pages of things she asked for. [00:40:25] **Sowmya:** And now, I just, yesterday, she wanted, like, a unicorn fairy with, like, flowers in her hair, with eyelashes. I don't know, something, like, super random. And instead of, like, me Googling, trying to find a coloring sheet, I just asked Dolly, I'm like, print a, like, you know, make a coloring, line drawing for that. [00:40:43] **Erik:** That's amazing! [00:40:44] **Sowmya:** And she's like, holy crap, like, I mean, she didn't say holy crap, but like, you know, It came to life, and it was a little bit of magic, but I'm also like, this is going to be a lifesaver, [00:40:53] **Erik:** actually. That's crazy. [00:40:55] **Sowmya:** So, I don't know. It's pretty cool. [00:40:58] **Erik:** Alright, we're about out of time, but before we leave, we have a last segment. [00:41:04] **Erik:** We call it Controllers Are Fun Too. And so we typically ask our guests if they have the best, like what's their best accounting joke or what's the oddest expense report they've seen, or maybe just like what the biggest debacle in accounting they've been a part of. So curious if you have anything. [00:41:22] **Sowmya:** I would say where, okay. [00:41:23] **Sowmya:** But this is one of my like startup. Clean the QuickBooks job, right? Uh, and it made me vow like never to take a job like that again, where I have to clean up QuickBooks. I'm like, are you guys on QuickBooks? Cause I don't even want to talk to you. Um, but so this, this outsource team had basically been claiming that they reconciled the bank statement, uh, for every single month that they were outsourced for, right? [00:41:47] **Sowmya:** Uh, and then I was like, okay, like, you know, for a startup, that's like, That gives you a lot of comfort that most things are okay. And I was really surprised because a lot of the accounts that they had referenced out were actually like platform accounts and like things where it's not just like AP payments, but also like customer activity. [00:42:03] **Sowmya:** And I was surprised, but I was like, okay, great. Like, you know, they've figured it all out. Like maybe there's no like crazy balance sheet things. And then when I started digging into it and I looked at the account, And I saw that they basically created like phantom asset accounts and they would just like credit the bank geo and debit this phantom account. [00:42:22] **Sowmya:** So like you haven't actually done any accounting, right? You've just like credited and debited the balance sheet for like the same thing. And then I was like, what are these random, like, you know, other current asset type accounts there? They're just like buried there. And it's like, And then when I got into it, I was like, Oh my God, this is going to be like years and years of like, dealing with every transaction and nothing had been done. [00:42:46] **Sowmya:** And I was like, I don't know how to feel about this, but I will tell you there's a big opportunity and like, actually being an outsourced bookkeeper that gets stuff right. It was bonkers. For the listeners [00:42:58] **Erik:** out there, what Somya is describing, like, you can do a bank rec, right? You figure out cash coming in, cash coming out. [00:43:06] **Erik:** Uh, for cash coming out, you would debit some other account. For cash coming in, you should credit some other account, right? Because both sides need to, need to equal. Those credits and debits, it sounds like they're just putting in suspense. Like forever, you can't even call [00:43:21] **Sowmya:** it suspense. [00:43:22] **Erik:** They just booked it to a random account. [00:43:24] **Erik:** Oh, I hope it all nets out. Maybe that's maybe what they're thinking, but. [00:43:28] **Sowmya:** It definitely did not. And we found like big stuff after it. I was like, great. Yeah. I mean, it was fun slash really kind of once I got into it, I was like, okay, this is great. Because now I know every single transaction that ever happened in this company. [00:43:44] **Sowmya:** But let's just say, I didn't know I was signing up for that. [00:43:48] **Erik:** It really gives you an appreciation for. Just like having the right bookkeeping and how much time it saves you later, like just to avoid that, yeah. Well, Soumya, it was an honor to have you on the show. Thank you so much for being here and sharing your stories. [00:44:03] **Erik:** Thank you for having me. And for all of us here at Controllers Classified, my name is Eric. Thank you, Soumya, again, and we'll see you all next time. [00:44:10] **Narrator:** Thanks for tuning in to Controllers Classified, presented by Brex. Brex is an AI powered spend platform with global corporate cards, expense management, reimbursements, and travel. [00:44:25] **Narrator:** Visit brex. com and follow Brex on social to see how they can take your accounting game and your company to new heights. ### Image ![Closed-loop-dark](https://brand.brex.com/transform/d2f2341f-92ea-4983-a67d-417a52921900/Closed-loop-dark) ### Content See what Brex can do for you. Learn how our spend platform can increase the strategic impact of your finance team and future-proof your company. [Contact sales](https://www.brex.com/sales) | [Open an account](https://www.brex.com/signup) ## Related Articles ### [Episode 01](https://www.brex.com/resources/controllers-classified/episode-01) Brex’s CAO Erik Zhou dives deep into the evolution of accounting and finance with guest Ron Cook, the VP/Controller at Lightcast. ### [Episode 08](https://www.brex.com/resources/controllers-classified/episode-08) Brex’s CAO Erik Zhou welcomes Barrett Daniels, the co-leader of US IPO services at Deloitte to talk through considerations and best practices in the IPO process. ### [Season 2 Episode 04](https://www.brex.com/resources/controllers-classified/season-2/episode-04) Brex’s CAO Erik Zhou welcomes Jessica Peng, Controller at Invoca, who uses a recent revenue recognition system overhaul to highlight change management best practices. ### [Episode 10](https://www.brex.com/resources/controllers-classified/episode-10) Brex’s CAO Erik Zhou is joined by Patrick Lynch, SVP and Controller of the Boston Celtics, to do a deep dive into the world of accounting and finance for sports teams. ### [Episode 09](https://www.brex.com/resources/controllers-classified/episode-09) Brex’s CAO Erik Zhou welcomes Katie Slattery, VP of Accounting and Corporate Controller at Fivetran to discuss high-performing teams & global operations. ### [Season 2 Episode 03](https://www.brex.com/resources/controllers-classified/season-2/episode-03) Brex’s CAO Erik Zhou discusses public company accounting considerations and best practices with Eric Van Cleve, Controller at 8x8.