Stop fearing it, start building it: AI and finance talk, with Lydia Stone

Lydia Stone: M foreign.

Narrator: Leader wants AI to surface the insights that drive real decisions. But AI can only work with what you give it. And if your billing data is fragmented, your revenue recognition is manual and deferred. Revenue lives in a spreadsheet someone built three years ago. You're not getting insights, you're getting noise. Chargebee fixes that at the source. Billing that supports any business model from flat fees to usage based outcome driven or hybrid approaches, ASC606 compliant, revenue recognition and real time cash flow visibility all in one platform. No more month end scramble, no more audit anxiety. Just a finance team that is completely in charge of the numbers and empowered to support the business as it experiments and transforms its pricing. Find out more@chargeb.com so I'm delighted today

Anthony: to be joined on the $0.02 podcast by Lydia Stone. She is finance and Chief Accounting Officer at chargebee. Lydia, thanks so much for being with us today.

Lydia Stone: Thank you for having me, Anthony.

Anthony: So before we get started today, do you want to just tell our audience a little bit more about yourself and your kind of journey to your current position?

Lydia Stone: Sure. Um, I started off as a traditional accountant in big four accounting firms. And um, most of my career in the first 10 years was with large public enterprises which really provided a large, a big playground and learning, um, ground for me, uh, the rigor, the discipline, the technology, the transformation. Um, but eventually I found my home in the tech and growth side, um, growing with the company, uh, helping company through their growing pains, different phases of their, you know, life cycle. And that's what I really enjoy, um, doing. So on the charge B side, my job is, uh, three foes. One is I am the guardian of the numbers. Right? I'm in charge of the, making the numbers, reporting the numbers, having people understanding the numbers. The second thing is unblock um, the team using technology to make sure, um, we are not the bottleneck but we are accelerator of the company. The third thing is unique, uh, to the position here in chargebee. Um, we eat our own dog food. So I also am the advisor to our product team, um, to help shape the future of our product.

Anthony: It's great to hear. Yeah, absolutely. I love the, the phrase guardian of the numbers. That's, that's a, it's a great, great phrase definitely for, for someone in finance. Okay, so let's, let's get into the kind of meat of our discussion then. We want to talk today about what everyone is talking about with AI. Um, but I want to move beyond like sort of buzzwords, you know, it's become a Bit of a buzzword in finance, but in your experience, your vantage point of charge, be like, how would you describe, you know, where finance teams are actually at with AI right now? Are they embracing it? Are they stalled with it? What's your perspective?

Lydia Stone: Yeah, looking beyond the hype and the buzzword, the real grant of playing with AI, uh, where things are, um, from what I see is a lot of people are still in active experimenting and piloting places. Um, there's, you know, there are some successes at certain, uh, areas of finance, perhaps more of, you know, data extraction, data analysis and, you know, efficiency in reporting. Um, but beyond that we see, um, I see a huge urge and desire and anxiety to try to use it more combined with this huge hesitation. Hesitation. Because there seems to be a trust gap, if you will. We finance people are inherently, um, concerned about accuracy, concerned about, can I trust the process, thus trust the numbers. So are we there? We're not. Uh, I don't believe we are there. Now perhaps there's a small groups, more advanced, larger corporations that are more on the right side of the spectrum. The. But most people we see is in piloting or starting to pilot, but have not reached a scale yet.

Anthony: Yeah, absolutely. Yeah. So you think it's a bit of a mixed picture at the moment. There's a bit of hesitation, some curiosity and some old habits that aren't going away, I think are holding it back. Right.

Lydia Stone: So the mindset is there from, you know, most people I talk with, there is this worry about falling behind. Uh, there is this desire, strong desire of wanting to do something. Um, but either they have started piloting or in the process of piloting or have piloted, but not knowing what the next steps are is, um, you know, it's the, the minority, few on the right side that have figured out perhaps a midterm plan going forward.

Anthony: Yeah, definitely. Yeah. Okay. So what do you think then? Uh, currently, like, we get through all the kind of noise and hype around it. Where do you see it currently making the most impact in finance? And, and where do. What do you think that's going to lead to? How is that going to help evolve the role of the CFO or the finance leader?

Lydia Stone: I see, um, I would say the majority of the value that we saw is in AI in finance. It's not automating an entire job or entire process, but what it is helping with is solving some pain points with existing processes. For example, making forecasting better, faster by providing data insights. Right. Taking away the data, digging part of the preparation data. Uh, preparation or reporting Preparation. So we see individual tasked agents providing that value. Um, and that, making our reporting faster, making our uh, reconciliation faster, making our month close reporting faster and so on. Um, as far as beyond that, um, we are also experimenting on the day to day basis ourselves. We, we, we want more right where it, where are we headed? We're, you know, figuring out it every day.

Anthony: Yeah, definitely.

Lydia Stone: Yeah.

Anthony: Uh, I think it's, we're all in that position, aren't we, right now of sort of, it's still, there's a lot of excitement, but it's kind of experimentation time. And of course, you know, in a field that you're in, in finance, you know, it's very, it's quite sensitive. You know, it's uh, it's a lot of high stakes stuff. So in many ways like the, some of the fear and hesitation, I think um, as much as the excitement is, is justified, I think some of the fear and hesitation is also justified that

Lydia Stone: completely like act right and, and you know what, what also is, is exciting but also posing more challenges is yesterday I figured out how to use ChatGPT and you know, today is clawed, but by tonight ChatGPT came out with a new version. Right. So where do I settle with the tool? Which one's better at which tasks? Thus we draw the most value out of it. And while trying to experiment more to gain trust in that process. That's, that's the playground over there.

Anthony: I suppose every technological shift has that a bit, don't they? It's kind of like it's not just the uh, fear of the unknown, but it's like what we, we don't know how to use this yet completely. You know, we're still, we're still figuring it out. Um, it's like. And then eventually, I suppose, because I do think it's kind of like this is as big as like the Internet in a way. You know, this is like the big, the big shift that's going to completely change the way everyone's working. But um, and I think, you know, it will become eventually like a kind of thing. Like you know, the way we're just online now, we're always online and we don't even notice we're online anymore. You know, it's just constant. Whereas when it first came out it's like, okay, I'm gonna, I'm gonna, you know, go, I'm gonna log into the Internet. You know, I'm gonna, I'm gonna set up a web browser. You know, you would think about it more and I think maybe Maybe AI will be faster, especially in job roles. In terms of like, the point is like everything you're doing in the end is automated. Um, but I feel like we're not there yet because we, we don't quite trust integrating it with everything and we haven't quite figured it out yet. But I, I think it will get there eventually. It will become a point where we don't even realize we're using AI. Maybe eventually, you know.

Lydia Stone: Yeah, just like we don't realize we have our phones in our hands every day. Our smartphones are scrolling or you know, touching screen. Um, we're at the very beginning phase of that experimenting. Is it going to be Apple or you know, Motorola back the days.

Narrator: Right.

Lydia Stone: So now, you know, um, most of us are, you know, just touching here and there, going slight deeper, slightly deeper every week perhaps, and then um, shallower next week, you know, deeper this week on different areas. And I do agree with you, I think it will eventually get there. It will take a lot of experimentation and um, especially in the finance world, we want to see more proof, you know, it's on a day to day basis. Using AI is not as glamorous as everyone, you know, paint the picture of, because it's a lot of grinding with the actual prompting the actual data or data cleansing and then, um, questioning the output and try to prove it and prove it again. Um, before we, you know, carefully put that page in front of people and say, I have vetted enough to present to you all, uh, using this shiny new objects that I think I have seen enough to be able to trust this. Right. For this page only.

Anthony: Yeah.

Lydia Stone: So that's kind of the conversation.

Anthony: Yeah, I think there's a lot of fun and excitement in it though as well. Like, you know, it's the kind of thing of like I've had it a few times where like I've seen somebody has uh, said, did you, you know, people have said to me. It feels like people are saying it to me every day. They say like, did you know that this AI tool can do this?

Lydia Stone: Oh yeah.

Anthony: You know, it can help us with this. And I'm sudd. Suddenly I'm like, wow, you know, that's gonna like free up an hour of my day. Like I've been, or I've been doing this task manually and it didn't even occur to me that I could use AI in this way to, to help me with it. You know, um, I was late with meeting notes. I was still, you know, on meetings with people and I'd still be like typing notes. And then I suddenly realized I was like this, I was about like, I think about a year behind everyone else in the sense that it's like, you know, you can just plug AI in and you can just grab the notes, right? And it's like, and it will take notes better than you can always, you

Lydia Stone: know, so exc, you know, now I have, I build a couple of small um, agents now notion, summarizes my calendar events, warns me, you know, I have a conflict at a certain point in time, warns me that I need to have a, I have a pre read for another meeting and then summarizes all the emails that I did not get to read yesterday or last night and warns me there's action that I, that I need to take care of. You know, at this point in time I almost already take that for granted, right. And say like, of course I need that. Right. So, so hence the speed of moving forward. You know, I'm looking forward to the next thing that either chatgpt or clouds will come out that enables the next set of actions beyond my personal productivity, beyond preparing some PowerPoint and make that faster beyond, you know, identifying a couple of trends in, in my huge amount of data. Right. So where, where is that next big thing that I can quickly learn and then move on again. Right. So, and but one of the most exciting thing I think I've come to realization is that, you know, as a traditional finance person, the numbers person, the reporting person, I feel, and my team feels so powerful because now if we need a functionality, we need um, a small tool to solve a problem. We do not need to schedule a meeting with the engineers and to give them our business requirements, describe what we want, define outcome, define the success and wait for them to code it and then come out of the SAT process, sit process and the UAT process and then say let's schedule a date to go live.

Anthony: Right?

Lydia Stone: Because we are the business owner, the process owner, we have the knowledge of how things should work and how things works and we can code it ourselves in this no code coding world. So that power just so exciting every day when we talk about it. And that excitement has been going on since uh, May 2025. Um, it at our teams here, um, and with people that I talk with. It's just so it's a two professions in one type of um, feel. Feels amazing.

Anthony: Yeah. So it has potential to break down silos a little bit as well, I think. Um, which is another thing, it's exciting. We always talk on the podcast about how, you know, we want finance leaders to become, um, what's the word? Like strategic partners. CFOs say they want to be strategic partners a lot of the time. And you know, they're tired of being seen as like the kind of like there's the rest of the team and then there's finance. And finance don't really have that much to do with the daily operations of the business. Like there's this kind of myth about it, I suppose. Um, but yeah, I suppose the potential for breaking down those barriers is exciting, you know, potentially.

Lydia Stone: Yeah, it's huge. It's cutting down so much prep time for the numbers, right? So we spend so much time just putting pieces of information together, pieces of numbers together to produce a set of numbers to try to tell a story, which is important to get to. But the amount of time to get there is now significantly shortened, thus allows us to partner up with other teams so, you know, impact other teams. You know what the trend has been? The CFO is becoming the new coo, right? So every part of the organization matters to get to the numbers, right? Um, it's the interactive, dynamic, um, process. Now in the AI world, not only that, what we also see, we're actually working on a few, uh, AI agents that are beyond individual task level, um, but became the orchestrator of multiple aspects of the company. For example, we have a deal desk agent that we're almost ready to go live. Um, it's pulling the sales and the marketing, the sales, the deal making team, the negotiation team with the back end. We're going to build it, we're going to revenue recognition, we're going to report ARR, and this, here's the board perspective to pull all that together. So in the deal making process, um, the very early stage or mid stage, you already see what the back end is going to be like. And on the back end when we're looking at these numbers, we see the front end, what they are doing every day, how things are shifting. And it took a number of smaller agents getting coordinated together, architected together to give us, um, that view. It's really, really exciting to see. And that definitely helped with what you were saying about the collaboration. We're collaborating without realizing we're collaborating. We're not scheduling another long meeting and hear everyone out before drawing a conclusion. Because the agent are giving us pieces of information every day, every minute. We are absorbing that without realizing we're actually actively collaborating on a new one

Anthony: that's absorbing the information. Like effortless kind of collaboration in a way, rather than having to Be okay, let's touch base about this. Let's do this. You know, the kind of thing, the information is just always feeding in, in a way. Um, you've kind of created like a kind of ecosystem in a way where the information is just spreading all the time, like kind of autonomously.

Lydia Stone: Absolutely.

Anthony: Clever.

Lydia Stone: Like a beautiful word, which is the ecosystem. Like, I think, you know, intuitively I feel that's where we're headed. We got like, uh, you know, Since June of 2025, we've started this experimentation of creating individual tasked agents. It feels like we need more. And this, this ecosystem, our network of agents working together, that is, you know, where the, the higher value relies to, to us.

Anthony: Yeah. So it's really interesting to me that chargebee is, you know, just kind of, um, you know, all in on the, on the AI agents at this point, because we hear time and time again, you know, people are still unsure about it and everything. Like, can you kind of maybe take us through your journey a little bit with that, with employing AI agents? Like, when did it start out and how was it scaled? You know?

Lydia Stone: Yeah, um, it is an interesting journey because they start out with, you know, us trying to solve a pain in the process. And, you know, then you hear about these AI tools, the hypes out there. We say, you know what, why don't we, um, do some demos? Maybe we're gonna buy a tool. Um, so we started looking. We, we went through like a journey of tools. And every time I was like, oh, well, I think we could build something ourselves. You know, as we're, we're talking, and one of our team members, um, and decided just to experiment. And one of the fortunate things in chargebee is we provide just about every AI tool that you need to do your job. Like, just feel free to go experiment. That, um, that really helped. We were probably one of the earlier subscribers of ChatGPT Enterprise then now we have Claude, relevance, notion, you name it, there's a tool out there, we might, we probably have it. So. So that really, um, helped. We always encourage our team members to experiment, right? And so one of our team members took a pretty painful manual process and just said, you know what, let me just see what I can do. And the result was quite amazing. That was like June, July of 2025. So when he showcased us how he was able to, within about three minutes, eliminate about three days of work, we were just, you know, flabbergasted. We, um, said, how do you can you go teach everyone else about what you did? And, um, next Thing, you know, that stirred up quite some excitement. So everyone uh, went into this informal, informal computation mode. It's like I could build something to cut five days out or six days. So, you know, without realizing we got a long list of little things. They are all each made everyone's job easier. Uh, we saw that excitement throughout second half of 2025. Now turning to 2026. Then claw came out and this truly no coding. Coding of building more complex agents. So, so as um, philosophy, we wanted to be AI first company. AI native company. So what does that all mean? You wake up, you have a question, you say, I'm going to go Google it. No go AI it. Right? So every vein, every cell of ourselves, by the time you wake up in the morning, you pick up your phone, you pick up AI. The first app I look at is not emails anymore. I go to Notion, I go to Notion, I go to ChatGPT because Notion summarized my email for me, Notion summarized my calendar events for me, Notion summarized what my day will look like. And then the next thing I jump into cloud because I want to ask another agent about something else. Um, it's a mindset, um, with, it's a mindset shift. With the encouragement and support both from a technology perspective as well as from a cultural, uh, perspective, I would say we are on our way to a real skilled phase, um, of AI. We're not there, but I think we've stepped beyond the early stage of building simple task level agents. Yeah, our first or first two, uh, collaborated network of agents to solve a much bigger issue is just about to go live.

Anthony: So that it sounds like, yeah, it's become one of those things. It's almost like there's no stopping it now. It's like self sustaining almost. It's just grow, uh, growing out of itself almost. Yeah, yeah.

Lydia Stone: At one point we asked our team members, we say, hey, just list out the ideas that you have or your pain points that you want us get solved with AI. Now we have uh, a very long list, right? We are all learning, right? So the idea we first thought were like, okay, once we have this list of pain points and use cases, we ask our AI team, who has amazing AI engineers to build it for us. Then we look at how long the list is, we get it, uh, how long is it going to take, how many engineers to build that for us, right? Why don't we just do it ourselves and leave the difficult ones, the bigger ones, the ones that need lots of APIs, integrations with other tools to them and hence we just Say you know what everybody, you get access to these amazing building tools and build your own. And um, that's what we see now.

Anthony: So I feel like you've probably already answered this a bit with your own story of chargeby here. But let's just put it in like a nutshell then. Like if somebody like finance leader comes to you, says like we want to get started, our team wants to get started with AI, like you know, uh, AI agents, but we just don't know where to get started. We're starting from zero. What would you say? Like what would be your, your advice? Like in, in a kind of nutshell.

Lydia Stone: Start with your pain points, start with what you know the best and don't try to be perfect. Um, just you know what bothers you on the day to day basis. I know we had a bunch of manual spreadsheet where we were to manually stitch them together to produce an overall story and overall picture. We say, you know what, I bet AI can do that. It took longer than expected for us to find out. Okay, it does do it, but it made some mistakes that we didn't think it would at certain aspect is not as smart as you know, everyone was thinking. But at the end we made it work to just took a bit longer. But through that is a learning process. Then you learn, you know, can I trust this part? Can I trust that part? And you make you draw your conclusion. Just start with something that you know, it's, you know and, and experiment it. Because the beautiful thing is nobody has the ultimate answer to this question of how you use AI. Just start, you know, start small, start with anything you want. And um, just like we start a new journey into anything. Just put, take the first step and your foot will leave you there.

Anthony: I love the advice of start with something that you know, I think what I would say, my, my 2 cents is that I think people sometimes make the mistake of trying to do, get AI to do something that they don't know much about. I think that sometimes is a bit of a problem and that's how you end up embarrassing yourself a little bit. You know. Like I was, I was listening to, um, it's a completely different thing obviously, but it is kind of similar. I was listening to a, a YouTube kind of essay podcast about a historical event that I happen to know quite a bit about myself. But I was just kind of like, I just wanted to. Sometimes I like to listen to historical podcasts just to relax. I don't know why. Um, I was like listening to it and I was like about five seconds in, I was like, okay, well, this is AI generated. You know, this, this script is AI generated. I can just tell, you know, when you work with AI enough, you just get used to like the kind of speech patterns, almost like the way it formulates sentences and everything. Um, and then I was like, wow, this is really hallucinating, you know, like, this is just like, there's certain points where this is just like making stuff up, you know, because I know about the thing. And I was like, what's the mistake this person has made is they've made like, they're like, they've seen this like historical event and they're like, oh, this is interesting. I'd like to make podcast podcasts about that that'll get some YouTube views, but they don't know enough about the topic themselves so that they can oversee the AI, you know, and notice the problems that. So then you end up making errors like that when you don't always have knowledge. And I think maybe there's that mindset set shift there that needs to happen. Of like, don't think of like, AI as doing the work for you, you know, Think of it as like, you are the master of this. You are the one with the knowledge ultimately. And AI is, is sort of like your assistant, right? It's carrying out the tasks that like, you know, are just going to take up more time for you to execute on whatever vision you have. What do you, what do you think about that? You think that sounds like a sensible.

Lydia Stone: Hugely correct. So, um, what we realized was you use AI, you don't expect it to be a magician. You don't say, I have a problem. Give me an answer. Done. No, it doesn't work that way. It takes a lot more work. It's more like you hired a brand new staff, right? You say, hey, I have this goal of getting this thing down. But you've never done it before. You don't know what our process is, right? So let me, let me tell you. Step one, normally we do this way. Step two, normally we do that way. Step three, four, five, ten steps later, normally you get this, right? But if you see anything m better or a better process for step two and three, feel free. But the end result is I want an answer to that problem. Um, and then in, as you know, it's a new staff, right? As in this person is executing step five, you say, oh, no, no, actually, no, that's not what I meant. Okay. So, you know, I give you some, let me give you some context. Okay? This is how we normally deal with our customers, vendors, this and that. Now you understand that. Now you know why I said no. For step five, the way you're executing it, can you re execute with this new knowledge in mind? And then. Right, so you go step by step with them. Um, you, you verify, you course correct. You give more context for it to think similarly as you or better than you. Because it now has a lot wider context, faster for sure. So that's what we found is, you know, we, one of the project we did was we wanted to, we want the tool to read, um, many of our hundreds if not thousands of non standard contracts.

Anthony: Right.

Lydia Stone: The standards in your historical systems. Right. You just download a report of all the data field, voila, you have it. But if it's non standard in the past, it's a human reading what extra clause the lawyers added in what extra terms the salesperson agreed with the customer or with a vendor and so on. So what we realized was it hallucinated so much and produced so much error. Um, the error rate was unacceptable at the beginning. So we course correct, we say this. You know, you literally would tell it, tell the tool and say please do not fabricate data. Just like you would tell your brand new staff and say if you're not sure, please don't give it to me. Because it would just go on its own. Right. It's uh, so, so but overall it's like train it as if you're training your brand new staff on day one. Let it learn about you, let it learn about your company and then gradually become more experienced staff too. And senior staff. Right. And like manager, accounting, finance manager eventually. Right? Yeah.

Anthony: I think it's weird, isn't it? Because it's, it's like we're underestimating AI, but we're also overestimating it in some ways I think a little bit. Like we, because it is so impressive at times of what it generates, we sort of get tricked into thinking of it as a person that's going to think like, you know, sometimes I think most of the time like a bright young intern you bring in would be like, they probably don't want me to just make stuff up, you know, because they're big, they'll be like common sense says that's probably not what they want me to do. But of course it doesn't actually have that. You know, it doesn't. It, it seems like it has human qualities, but it does, it doesn't, you know, it doesn't think like a human and it seems like it does at Times, but it doesn't. Um, so I think it's interesting that, like, you have to be or educate yourself on what AI's limitations are. You know, the fact that it isn't a person and things that might seem natural to a person at times.

Lydia Stone: Right.

Anthony: Um, are not natural to it. Well, you know, it won't necessarily think sometimes. Oh, well, common sense tells me that's not a good thing to do. I shouldn't do that. It's a good thing to bear in mind.

Lydia Stone: Yeah, we learned some hardcore, um, limitations of AI. For example, um, all the tools can read one spreadsheet very well, but you threw 20 at it. Well, it's going to make all kinds of mistakes. That's the thing as finance people that even me, huge pauses. I said, you know, the value, the real value lies in. You stitch together the 20 spreadsheets for me. But you started making stuff up. Come on. Like, I keep yelling. At one point, I was going to start yelling. If it was my staff doing that. I'm like, I. I told you. No, no, no. You know, this is how you join it together. No, no, no. You did it again. Like, why aren't you listening?

Narrator: Right?

Lydia Stone: So there, there.

Anthony: So, yeah, you almost want to talk to it like a human being, like, emotionally.

Lydia Stone: Like, why, like, if you were my kids, I'd be. Start yelling. Right. So. But you, you realize, you say, you know, a lot of times then I go back to it and I say, hey, can you tell me why you're struggling with this AI? Why are you, uh, it would say, you know, here's an article, here's a YouTube video that you can watch about inherent limitations, about Excel spreadsheet, data processing, volume of AI tools. I was like, oh, that's good to know. Thank you for pointing to me to that video, but only after I asked.

Anthony: Yeah, you do. This is the thing again, that is a bit like an intern sometimes, like somebody who's new sometimes does try and cover up their, um, skill gaps a bit.

Lydia Stone: So then you learn, you're like, so now I'm not gonna try to do that again. I'm not going to upload 20 spreadsheets and say, give me this thing. Right? Because I know it hasn't got to that point yet. It's demystifying a little bit for me in our actual work, because we are still held accountable to produce reliable information on a timely manner. Regardless what tools, what AI internal or AI manager were, or human internal, human managers are doing, the result still has to come out in the same quality. So Our accountability never changed. Perhaps what our eyes are looking for is a little bit different when it's a human being versus an AI being.

Anthony: Yeah, yeah. I love, I love that story of like, having to ask it, like, how. What are you struggling with? I've had to do that too. I've had to, like, when it's like making a series of errors, I've had to say, like, is this the thing that you're struggling with? Please confirm this for me, you know, or like, just let me know. Because it's kind of, I guess it is programmed in such a way that it's like it will try and give you an answer that pleases you.

Lydia Stone: Yes.

Anthony: As opposed to just tell you I can't do this.

Lydia Stone: Yes.

Anthony: You know, that's the way it's like it's program. So you have to be aware of that. Uh, I think. And I think that's the thing that they probably will iterate on, you know, uh, improve on. Sorry. With future models, AI and Claude is, is make it almost kind of less agreeable, make it a bit more kind of like if somebody makes a request of you that's like outside of its actual capability, it will say, look, I can do this, but these are the limitations of what I can do here.

Lydia Stone: Yes.

Anthony: You know, I think that would be

Lydia Stone: a good upgrade thinker, you know, as a flight, especially in the finance world, to be able to, be able to think critically, just, you know, to illustrate the full picture. That's really, really important. Uh, as many times as ChatGPT says yes to me, I get more uncomfortable.

Anthony: Right. Yeah, I know, I know, I know, I know.

Lydia Stone: It feels like if I ask enough questions, it will eventually agree with me. Even if it start.

Anthony: Yeah.

Lydia Stone: Ah, it will start what I really want and, and get me there.

Anthony: I know. I actually do think it's gotten better. I think the earlier models M were very much like, great idea, great idea all the time. Now I've noticed with the newer model, when I ask it something, sometimes it'll say, I'm just going to gently push back on that. And I was kind of like, oh, they've, they've, they're learning, they're retooling it. Like you can tell, you know, like, yeah, it's like, that's what you want. That is critical thinking, isn't it? Like, I can accept your point of view there, but let me just present you with this information that, you know, in, in, like, that's, um. That seems like it's the way it should go.

Lydia Stone: Definitely. A week ago 4.7 came out. ChatGPT, the newest version came out. I've been really impressed.

Anthony: Yeah, yeah, I think, I think they definitely are, they are improving on it. I hear a lot of people say that Claude is better, but I think maybe it's just better for certain things. This is what, this is going back to your point, isn't it? Like, who knows, you know, today I

Lydia Stone: thought, you know, cloud do this task better, but tomorrow, you know, try to pt, you know, bls all of that. It's like, okay, I'm shifting. I built a bunch of custom GPTs. Do I move it to Claude now? I'm not going to.

Anthony: So yesterday I was thinking about it,

Lydia Stone: so, you know, who knows what tomorrow brings brings. But that's part of the excitement. Scarily excited.

Anthony: Yeah, definitely. So we mentioned, we touched on, you know, the finance being like kind of high stakes in many ways, you know, like you can't, uh. I'm someone who comes from like a language and words background, very much so. I feel like with me it's like AI starts making things up. It's like, well, you know, creative writing, like you can kind of get away with that a little bit more sometimes, you know, but the numbers can't lie, the numbers can't hallucinate, you know, where you are. Like, it, it has, it's pretty rigid, you know. So I'm wondering, like, what where do we have to get to in terms of like, you know, with finance, like what do we have to do in order to make sure that that output is as reliable as a person going through it? You've touched on it already but like, what would be your answer to that question? Like, how rigorous do we need to be in our, in our um, supervision of stewardship of AI?

Lydia Stone: Yeah, yeah. Hence, um, that exact question also answers why the adoption speed is at a, uh, cautious speed. It's not as hyped as the AI capability development cycle is. There is the capability, you know, development hype. There is the much slower adoption speed, especially in the finance world, uh, for the same reasons that you talked about. So, you know, data is essential. Like ChatGPT, nor any of the AI tools, nor a human being can make, you know, a flower out of garbage. Right? So like you just can't um, so garbage in, garbage out. So um, cleaning up the data using uh, your, you know, whatever bottom line infrastructure, um, to create that data, make sure that is good. Now you don't have to start when the data is perfect because the AI, ah, can help you fix the data. Right? So Start with the data, make sure data is uh, of a good quality. Um, and then start with lower risk tasks to build up confidence. You say this set of data is somewhat. Okay, um, let's start here and let's see if the result is satisfactory. What can we learn from that process, what the tool can do, what AI, uh, can do for us or not. And there's always this concept of do I buy a tool to do data cleanse or data reporting or to solve another pain point in my process or do I build it with ChatGPT or Claude or notion or relevance and so on? Um, if you, you know, it could be both or could be either. Uh, if you were to choose to buy, uh, buy you know, domain focused AI vendors that knows that domain so well, knows what risks you're concerned with, knows the quality of work that you want to, you know, you're dealing with, knows, you know, what stakeholders that you had, you had, you have to satisfy. So they, they're playing that smaller domain really well where you want to choose those, those AI vendors, right? So you don't have, have to, you don't want to have to govern and audit their process, the vendor's process to make sure you're, you know, you can rely on them, right? So if that's the path you choose, um, you want to make sure you're working with the right partners. Um, let's say if you want to um, build your own instead, right? It was just, you know, we're doing that every day. Um, so you want to use tools, you want to use tools that you can experiment, of course, but you want to have a human in the loop, especially at this first phase. Um, I think we're all in the first phase and have more human in the loop than you would otherwise think and train the human to look for, not just review the outputs. In the past, you know, you look at, you know, I got a formula error, I fingered some data and then the output just looks so wacky. It says our growth rate was, you know, positive 3,000%. Right. So the output is wrong. So I then start looking at backwards and see where our fat finger performing on perhaps, right? But in the new days you got look at your input and then you have to audit the logic mhm of the AI brain you ask to spell it out for you, right? You give all the rules to make sure it's capturing the logic that you're thinking about, uh, and improve upon it. But you got to know what the logic is behind the scene before you can audit the Output, because it's going to give you things that perhaps a human would not be able to or, ah, not able to do as well. Like a beautiful trending chart of the past 15 years of the company history. Well, that might never be possible or possible within a year or two years with a human brain, human speed, but the machine can do it within a few minutes. Right. So with that result, what is your benchmark to say? Well, I think it's a great result. I can trust that chart and the trend that it produces. Me, you just don't know because you have nothing to compare to. But the thing that you can be sure of or put more eyes on is the one, data input, make sure that's good. Two, you agree with the logic that the machine is doing for you. So that's, you know, we, I would say the same about this process of making sure you can trust the data before you put it in front of the board. Right. In the auditor's eyes, um, this process of verifying, I think I've done enough to verify it, is also a learning process. Right. Just like we are learning how AI functions, we're also learning how to make it function in a more reliable way.

Anthony: Yeah, definitely. It's, um. Yeah, I, I think one of the things that you touched on earlier is that there's too much discussion, I think, around the tools, isn't there itself. I think that what this comes down to a lot of time is a mindset, right, of how you, how you approach it. A lot of the time, I think is, um, is the key maybe to adoption or giving people confidence through adoption, I think is, um, yeah, 100%. Okay, um, so let's talk a little bit then about this anxiety around AI at a kind of broader scale. You know, um, AI, you know, um, replacing jobs, making human judgments obsolete. I want you to just kind of like, if you can like address some of those, like, you know, potential challenges that people have, especially in relation to finance, like, what is, is there validity to this and what do you think we need to do to kind of address these fears? Like, what do you think is the way around that?

Lydia Stone: Yeah, um, and it's real. Like every day you talk to people at work or at, um, a dinner table with your friends or whatever, people talk about their worries and then they think, you know, oh my gosh, AI is going to take the lawyers jobs away. AI is going to take, you know, um, finance job way. You know, that fear is real. Um, but if you really unpack it, what are they afraid of? What are they really fearing, we're thinking, you know, there's a few aspects as you unpack it, which is, is actually actionable. Right. They say action is the best fear buster. Right. Um, so one is number one, they're they're fearing about is skill, of their skill. Skills become outdated. Right. Which, you know, sure, when you, you know, when Internet came out, you, you no longer need to have a pen, a notepad, but everyone got trained on how to, how to use Internet, right. So uh, so the fear of job loss is really fearing about their skills getting outdated and they fall behind. Thus they can't no longer hold their jobs because you don't know how to do the jobs in the new way. And the way to, you know, one of the ways to bust that fear is training. You know, I've, you know, personally got trained and I sure the company can train people, but you can also train yourselves. What can you not learn from YouTube these days?

Anthony: Yeah, yeah, yeah, absolutely, I agree.

Lydia Stone: So it's the fear of skill, skill loss or skill outdating. Um, the second thing that they are feared about, which, which I hear quite a bit, um, from folks out there is the unclear guidance from leaders. You know, AI is already putting so much uncertainty, but our leaders don't know where we're headed. Are we doing it? Are we changing it? Are we not, Are we watching? Are we jumping in? Right. So I think, you know, from an organizational level it's really important the leaders set a clear vision, you know, and to push things forward or explain where things are and tell them that, tell employees that, you know, they have a plan or they're making a plan. Right. Um, and the third thing, of course, especially in the finance team, is the worry about accuracy, the auditability, the traceability. Um, and again, as I talked about earlier, having sufficient steps to verify the inputs to the logic and have a sufficient human in the loop. And don't imagine AI is this magical wand. You give it one task, voila, it's done. Next. Second, it takes real effort to build a real agent to, you know, real process to. That's the only way to produce your results. It's not, um, you know, it's like a dream. Everything. You know, you see those YouTube videos where it just produces beautiful, you know, videos at a click button. Unfortunately that doesn't, um, doesn't work in our day to day process. It takes real work to make AI work for us and eventually we'll work.

Anthony: I think it's true of everything actually. I do think to produce real quality work with AI I do think you have to train it. You know, I think that it's um, you know, there's a phrase that keeps getting thrown around, AI slop. You hear people say that like about like videos and stuff. And I think the slop, you know, I don't think AI is inherently sloppy, I think it's amazing. But I think it will produce slop with the minimal effort, you know, it will, it won't, it will create some things with no real substance to it. Whether you are talking about, um, you know, um, auditing, spreadsheets, connecting spreadsheets, integrations at work, or you are talking about, you know, creating a lovely uh, YouTube video as the example I gave you before. You know, it's um, I think with a bit of oversight, um, the long term results can be, can be amazing, you know, but, but you have to put in the work early on a little bit, don't you? I think that's important. So in terms of good, like, you know, again we talk, we go back to finance being uh, you know, really high stakes and you know, a lot of finance leaders are dealing with very sensitive data, other people's data, you know, especially in terms of things like fintech and stuff like that, um, what do you think needs to change? Obviously, like ordinance, audit and governance, you know, exists because human beings make errors, you know, they make mistakes. In the time of AI, where we know it doesn't make mistakes, but it maybe makes mistakes in a different way. What needs to change, do you think, in terms of governance, does anything need to change? What do you think is, um, what's your view on that?

Lydia Stone: What needs to change? Let me tell you what doesn't change, which is accountability. Accountability. So if something goes wrong, you present the wrong numbers to the investors. Guess who gets the blame? It's not the machine, it's still you, right? Still the human. So hence we humans still have to figure out how everything works, right? It's your human employee or your AI employee that's doing the work. So let's say I have a piece of work that is done by a human employee. The way I review it, it'll be a bit different than how I review my AI employees work, where my human employee, I kind of know what kind of errors they tend to make. The fat finger, you know, they stitch the two columns, you know, that's, you know, not exactly aligned, you know, those. And they perhaps, you know, produced a chart that took a bit longer than, you know, my next employee that's, that's named AI. So I know that the review process to catch the errors of my human employee. Does it work the same way when the machine produced that sheet for me? Right. So that comes down to like, or still experimenting. How do I catch the machine's error? I realized something has to change because I don't necessarily see in Excel sheet that human formula is A minus B in parentheses times C divided by D. I said, no, no, no, it's not D. You put the wrong column. It's C, it's F. Right. Then I look at the output as the. Why is that 5%? I thought it would be a lot higher, right? So then the human would dig back in and say, oh, you know what? I pulled last year's number, not this year's number. Oh, I'm sorry about that. Let me fix it. Right. So through that interaction, you get comfort of the output and say, okay, look, makes sense. You fix some errors, review three times, I'm done. Can you do the same, uh, for the machine? Um, not the same way. What we found, right? So coming to the two points, at least two points that people need to focus on is once they input this huge amount of input, you really want to make sure that you're giving it the right context, the right input, and then audit the logic behind it. See, a lot of times the logic is not as clear as the human, um, um, formula. So how do you get the behind the scene, uh, logic to get there, to get to a point you say, oh, I've seen enough to trust this set of numbers you present it to in front of others. Um, it is um, a trial and error right now for us. We are um, one of the agents we build, which is what I mentioned earlier, to review contracts, to produce non standard contracts, to produce a set of contract data by reading those contracts one by one in a speed that human can never match. And we tested it for three months before we can say, okay, the error rate now is at acceptable level of 2%. It took some real work, real correction, real training and train, train, retrain to get there. So it's going to be different. It's going to be a few basic elements that are same for everyone, but new things for each agent depend on what the agent's job is. So keep that human in the loop to get the human comfort to a point of you can hold me accountable to that, right?

Anthony: Yeah, yeah, absolutely. And that's a, it's a great. Yeah, I think that's, it's definitely true. Accountability never changes. 100%. I agree. Um, okay, so final wrap up question I, What I normally ask people is, what, what do you think finance is going to. AI and finance is going to look like in the next five years? I think it'd be more interesting because that, it's like, it's speculative. You know, we can't really know the answer to that. So what I'm going to ask instead is what do you think? What would you, what do you not want to see in five years from. From, um, AI and Finance? Like, well, what do you think it be. Would be a sign that we weren't headed in, in the right direction?

Lydia Stone: Um, yeah, I don't want to see us, um, being stopped by fear. It's not perfect, but start doing something. Uh, it's not perfect, but it's useful. It is useful. Not everything, but it's useful in many things. And that is good enough. And if you have 100 pain points, it solved two, you solve three. Ah, a solved half. 20%. That's good enough for us, for me, right in getting started because, you know, the tools are evolving, so, you know it can do better in the future. But as you start with one idea and two ideas, three ideas continue with that journey. Not be, not be stopped. Of, oh, the tool is evolving, but who knows what tomorrow brings? Or I'm just so scared. Or I'm so scared of not where. Not knowing where to start. Right. And not doing anything. Because that's a sure way to guarantee our finance organizations become obsolete. And we need to do the opposite. Yeah, right. We need to lead the charge in the next cycle of, you know, utilizing it because, you know, this profession can benefit a lot by automating away the repetitive manual tasks. That ties us down. That takes way too much time every month that give us the reputation of like, oh, let's just wait for finance. No, we need to jump way ahead of that and become the accelerator and the enabler of the business.

Anthony: I love that. That's a fantastic way to end. Lydia. Uh, accelerator, an enabler. Uh, of the. Like that. Of the. I love that. Thanks so much for your time today, Lydia.

Lydia Stone: Thank you for having me.

Anthony: People want to reach out to you. Uh, yeah, people want to. Want to reach out to you. Find out more about the subject. Where should they go?

Lydia Stone: Uh, reach my email or, uh, yeah, LinkedIn. Send me a message. I love hearing from you.

Anthony: Thanks so much.

Lydia Stone: Thank you, Anthony. Thank you for taking the time.

Narrator: Every finance leader wants AI to surface the insights that drive real decisions. But AI, AI can only work with what you give it. And if your billing data is fragmented, your Revenue recognition is manual and deferred revenue lives in a spreadsheet someone built three years ago. You're not getting insights, you're getting noise. Chargebee fixes that at the source billing that supports any business model from flat fees to usage based outcome driven or hybrid approaches, ASC606 compliant, revenue recognition and real time cash flow visibility all in one platform. No more month end scramble, no more audit anxiety. Just a finance team that is completely in charge of the numbers and empowered to support the business as it experiments and transforms its pricing. Find out more@chargebee.com and that wraps up

Speaker D: another episode of the Two Cents podcast. Thank you to our listeners for taking the time to tune in. If you enjoyed this episode and want to hear from more finance leaders like today's guest, please subscribe to the show and leave a rating or review on your favorite podcast app. It would mean so much to me and my guests if you can show them some love and appreciation for sharing so many great insights with us on the show. Again, thanks for listening. See you in the next one.