The Future of FP&A with AI for Finance Professionals to Move Beyond Excel Analysis with Derek Baker

I don't think spreadsheets have a place anymore for analysis. And one of the big reasons for that in my thinking about spreadsheets is that spreadsheets are really horrible data format. It's too customizable. You can have a random number in cell, AEC, a million. And that will mess up your whole data structure whenever you try to do analysis in Excel. And that's just kind of how we interact with it. We've gotten used to having, you have a table in this part of the sheet and a table in this part of the sheet and this part of the sheet. It's just a bad data format. If you want to do analysis, the best way to do it is in my opinion, in a data warehouse, you can build a lot of semantic knowledge. Now there are a lot of tools coming around with adding a semantic barrier to your data warehouse. Welcome to another episode of FPNA Unlocked, where finance meets strategy. I'm your host, Paul Barnhurst, the FPNA guy. Each week, we bring you conversations and practical advice from thought leaders, industry experts and practitioners who are reshaping the role of FPNA in today's business world. Together, we're all in cover strategies and experiences that separate good FPNA from great FPNA. We'll help you elevate your career and drive strategic impact. I'm thrilled to welcome our guest this week onto the show, Derek Baker. Welcome. Hey, thanks for having me. A little background. I've known Derek now for what's been four, five years? Six. It was back in twinks. I'm getting old. Thanks for reminding me. But I've known Derek a long time. I've had him on FPNA today in financial modeler's corner. So I figured we should bring him on to FPNA Unlocked. And I'll give a little bit of his background here in a minute. But what I'm most excited about is we're going to get into some of the nitty gritty of how he's using AI, where some of the challenges are and how he thinks about it. Because I know he's doing some cool stuff and he's spent a lot of time here. All right. So Derek's background. Derek is the head of strategic finance at Circle, a growth stage SaaS company building the leading all in one platform for online communities. At Circle, Derek has built the strategic finance function from the ground up, tackling everything from pricing and packaging optimization to sales compensation design, investor due diligence, and implementing AI powered financial reporting. Before Circle, he was a customer of the product through the FPNA hub. A community he co-founded with some more on, I mean, some guy named Paul Barherst. And Dr. Laurent Edelest. The community was designed to help finance and FPNA professionals learn, network, and grow together. Derek's career has spanned FPNA roles at startups ranging from SaaS to Marketplace to biotech businesses. He lives in Utah, so he's my neighbor just down the road with his family and he's passionate about solving hard problems at the intersection of finance and technology. So again, welcome, really excited, love the background. >> Makes how you be here. >> You know, I hadn't read the intro yet and all of a sudden I saw my names, I had to have a little fun. >> Yeah. >> So I like to start with this question just to see the different answers. It always amazes me. So we're going to start here before we jump into AI. If I asked you to describe what great FPNA looks like, what would you be your answer? >> Lately, I've been hiring lately, so I've been talking to you, can't, it's a lot about this. Lately, I've been describing what we do is we are the bridge between executive strategy and the operations in the business. And at least that's how our role is. Maybe traditional FPNA isn't always that bridge, but if it's done correctly through the FPNA cycle, we're bridging the gap between where the executive leadership wants to take the business and where the people who are implementing that strategy are actually steering the business. And so through the FPNA cycle, through accountability and visibility into how the business is performing, we can really help to steer that trajectory of the business towards where executive leadership wants to go. >> I really like that answer and it made me think of two things. I saw a visual that kind of gave the three areas of strategy. The senior leadership is the strategy formation. But you have a strategy planning and a strategy execution. Finance plays a key role in the planning, particularly in taking the strategy and relating it to finance, then also helping with that execution and that bridge to the business. And so I think we pay a key role, like you mentioned in those areas, if you're doing it right. >> Agreed. It's a tricky problem because a lot of it is also translating operational data to financial outcomes and vice versa. When we're doing strategic planning, we're reverse engineering, financial outcomes, and understanding what we need to do operationally to create those. And I think that's where a lot of the analytical side of FPNA comes in is translating those two sides of the business. I'm working on a course right now where there's a whole section on taking financial performance and figuring out operational drivers and performance metrics and a value driver tree and all those type of things. So it's exactly what you're talking about. But trying to think of what our frameworks we can use to do that. Because it can be hard. Okay, I understand our revenues higher, but what's really the key drivers and which are the ones that the business can influence? Which are the ones that we can actually track? And how do I have those conversations with the business? And it always seems simple on paper. It's messy in reality. I want to ask one more question because I know you've mostly worked with startups. This is the first company where you've worked a little bigger. You start to see some of that growth. So how do you see FPNA changing? What are the changes you see as the company scales? >> It's a great question. And I'm figuring out as I go, this is the biggest company I've ever worked at besides an internship at a large construction company. And so I'm picking up as I go along. I think the biggest thing that is changing for us right now is the breadth of scope is just constantly expanding. And it's become impossible for just one of us to be generalists and understand the entire business and go beyond the surface level of the business. We still have to understand how everything fits together and understand how the business works together. But to go deep on specific areas of the business is requiring more focus. I don't like to use word specialization because I still believe that we're all finance athletes and could flex across all areas of the business. But it's important to have focus, analysts that are focused on certain areas of the business over others. And so that's where we're at right now is starting to really define not necessarily business units but focus areas for finance to bring the strategy. There's a whole. And the way that we're doing that is by thinking about how are we positioning our, you know, the strategy that the executive leadership team has lined up and when we try to adapt to that. And so right now we're focused on two different go-to-market motions. We have our PLG segment and our sales lead segment. And so we're thinking a lot about building out more rigor through focus areas between those two different parts of our go-to-market strategy. And then of course, we still have the GNA and the product and engineering side of things that becomes its own focus area, but it's a more broad but it's a less intensive focus area than the go-to-market ones are. Two things I like, you said there one. Yes, not specialization but focus areas. There's areas we have to go deep at different times in our business. But if you're going to be that athlete, you still have to understand the big picture. You got to be able to flex to different areas versus sometimes we see someone specialization. They're great at just that. But if you put them somewhere else, it's like, oh boy. And so I think that's a good, good analogy or like that. And then you're dealing with all the challenges of scaling. As you mentioned, the business units and just trying to figure it out. I love sharing a little bit of that. I think we could probably do a whole episode on that. How you think about structure, how you scale, the challenges you're facing. And I'm sure people would like it. But I know we want to get into AI and that you're doing a lot there and that's kind of the not kind of that's the word of the decade so far, I would say. And it seems to be that hallucinations of this whole whole AI world. Before I get there, a couple of the questions I want to ask, you very much have leaned into tech startups. Why? Well, I think it's what originally interested me. I went to school in BYU in Utah and there are a lot of tech companies in Utah Valley, which is where the county that Utah's in. And so a lot of the internship opportunities were tech startups. And it was just what I was surrounded with. There's just a big, a growing tech ecosystem here. So I think that's probably the originally interested me is just that was where the opportunities were. And I tended to love it and I was drawn to it. I don't I think I probably could be very interested in other areas of finance. I used to think tech with the only industry for me, but as I've talked to more FPA professionals in other industries, there is a lot of complexity in other industries like manufacturing and PG even e-commerce. I've talked to some e-commerce professionals and there's just a lot of supply chain logistics there that I think can be really interesting. And so I'm not saying that tech is the like the best industry to do FPA and but it's the one I found myself in. I've gotten pretty in depth here and I just really like working on the bleeding edge of technology. I think especially now in the world of AI, there's a ton of freedom and really also a really high expectation, which can be high pressure, but to be on the front end of these new innovations. And I find that person really exciting like I'm having a lot of fun right now, experimenting and playing around with AI and starting to implement it in our daily workflow. And I think the tech industry in general is on the forefront a lot of that because it's a technology at its core. So it makes a lot of sense and I would agree with you text leading the way in the sense of right, it's technology at its core AI you would expect them to lean in heavily. So let's jump into that. I want to spend some time there. So I know you've leaned heavily into Gen AI kind of from the start. I know you've got you've always enjoyed technology. So what really got you started? When was it you said I should be leaning into this heavily? What was it that made you say this could make a big difference in my work? Yeah, I think a couple months after chat, TBT came out. I started using it to write code to automate reporting. And it was really bad. I was actually using it more. more as a tool to help me to learn syntax and how to write Python code, almost never ran with what it gave me back in those days. But I used it mostly just to learn how to, you know, get out of spreadsheets. That's kind of been the theme of my career so far as if I don't have to do this in a spreadsheet I won't because spreadsheets are not repeatable or automatable at least not easily. And also they don't transfer well to in this new world into agentic systems. And so I think that's where I started with AI was using it to help me learn how to write code and automate small one off processes in my daily workflow. And as time has gone on, it's just things have gotten better and better. Actually, I realized something crazy yesterday that it's only been six months since Cloud Skills was released. It feels like it's been two years since then. But it's only a six month old and that was probably where things really started to change was this concept of skills that you can train, cloud on a process and it can do it the same way every time because that was set of instructions and context to give it more information. And so I'd say up until six months ago, I was really just using the chat bots, you know, Cloud, chat to be and using it to in a Q&A type format asking questions, getting answers. And it got better and better throughout that time. And then six months ago, things started to really pivot when they can now learn and repeat processes. And from that stamp, from six months ago, things have just accelerated where they can now do things autonomously. Now my daily driver, I'm in Cloud Code way more than I've been spreadsheets or in a code editor. I'm using Cloud Code to, you know, do analysis to I use it more than I use the chat interface now. I use it just as my normal chat bot Q&A style. And so I think those, that's my current journey. I'd say I'm only a few months in and learning a lot and like what I'm doing today probably won't look anything like what I'm doing in a year from now, but having a lot of fun as the tools are being built, learning how to use them, implement them our workflow as it goes. Thank you for sharing that journey. I appreciate that. A couple of things I want to dig into. I'm sure one of me mentioned some people find controversial, but you mentioned, you know, spreadsheet. I think you said some of the fact of it. It's not good for kind of storing data, interpreting, pulling out the data in this AI area. It's not repeatable. So what's kind of led you to that conclusion and what role do you think the spreadsheet plays going forward as we see more and more AI? It's a great question. I've thought about a lot and I don't know. I don't just say it front. I don't know the answer to where it's going to be in six months or a year, but I can give a couple of thoughts I have. The first is you can have to separate financial modeling and planning from analysis. At this point, if you're doing analysis in Excel, even ad hoc analysis, you're going to get left behind like really quickly. That's the first big insight that I've seen because AI is really good at writing code and doing analysis is just writing code, whether it's SQL or Python. So on analysis side, I don't think spreadsheets have a place anymore for analysis. And one of the big reasons for that on in my thinking about spreadsheets is that spreadsheets are really horrible data format. And you know, that will mess up your whole data structure whenever you try to do analysis in Excel. We've gotten used to having like you have a table in this part of the sheet and a table in this part of the sheet and this part of the sheet. And if you want to do analysis, the best way to do it is in my opinion, in a data warehouse, you can build a lot of semantic knowledge. Now there are a lot of tools coming around with adding a semantic layer to your data warehouse. And that just adds context to AI that will query it more effectively. And so that's where we invest a lot. Like if we get asked to do an analysis ad hoc and we don't have a data or warehouse, we default to building the integration or a data warehouse and get the data in our warehouse. Because we believe we'll use it in the future. And more data is now compounding. The more data you have in your warehouse, the better it's modeled if that data is talking to each other through relational data modeling. And then you can add semantic knowledge on top of it. There's going to be compounding value from doing analysis outside of spreadsheets. Now in the financial modeling side, I don't have a lot of answers here yet. We still do all of our financial modeling spreadsheets today. And I've started to think a lot more about, well, I really like the concept of having AI doing analysis and also being able to interpret what the impact of its findings are on the future or business. And the best way to do that is through a financial model. And this is for our business, mostly around the revenue side of things. You don't have very complex cost modeling, most of our expenses are headcount and hosting infrastructure costs and software. It's not a complex cost-structure. And on the revenue side, I've been starting to think a lot about, can we take some of our revenue forecasting models and can we put them into a Python package and give a CLI tool to AI so that when it's doing an analysis on a specific customer segment, it can say, if we grow this segment 20%, what is the impact on our growth over the next six months or NRR over the next 12 months, that type of thing. And I think that could be really powerful. And that's something we've just barely started scratching the service on is we built our first cohort forecasting framework and we're implementing that in a Python package right now that Cloud Code can interact with. So I think that the long-term direction is we're going to start to see more and more offloading of components of the financial model to AI, but there still need to be deterministic. And of course, there are always going to be the black box machine learning models. There's a good reason to have those for short-term predictive modeling, high accuracy type of thing. But for strategic modeling, trying to understand the business, I don't know if there's a better communication tool than a spreadsheet where an executive can get their hands in and they can easily manipulate an assumption and sure you can build that in lovable or in replet or Cloud Code and build your own app. But then who maintains it? And I don't know if I want my team to be, I don't know if I want one finance engineering team that understands how that app works and the rest of the team doesn't and can't interact with it or can't add things to it. And so what I think I see on our team, and this is totally conjectured at this point, I don't really know, but I think our spreadsheet models are actually getting simpler and simpler and simpler. It's going to be about finding what are the few assumptions that really matter in the business and then using AI to inform those assumptions better rather than trying to model out every individual customer segment, every individual marketing channel. Let's take all of that complexity and offload it to our AIS and in AI agents and then use that to help us to do analysis to inform our assumptions better and then document those things, which AI is also all that really good at documenting decisions that you make and helping to have a lot of the ability on why you made this assumption versus a different one. And I think there's going to be a lot of value there. You take these revenue models and you start to have four or five of them at some point. You start to have different methods of forecasting. Maybe you have a machine learning model that you don't really know why it gets you to answer it does, but it does. And it's like a time series model that's just trying to regress on what's going to be the most accurate short term forecast. And then you have a very deterministic driver-based model that's very similar to what we build in FPAN today. And then maybe you have a cohort-based forecasting model that's a little bit different and more granular. And you take those three models and you plot them against each other and you ask yourself which one do we think is right? And if you understand the models and how they're built, you'll understand which ones have their strengths and weaknesses. And you can start to make better judgment calls on why the assumption should be what it is in your main fan from all that you used to communicate to the board or to executive leadership. So I think that's where I see a lot of this going is they're going to be married together for the foreseeable future. Eventually, who knows? Maybe coding gets so good that we've reached the singularity and we don't need to interact with any tools ever again. I don't know, but for the foreseeable future, I think it's both and spreadsheets get simpler and code gets, it becomes more of a surface area that we need to build on top of. Yeah, I could definitely see that happening. I think there's some challenges in the training and the learning and asking everybody, but from just the logic of how things work, I can get behind a lot of what you're saying. We know AI is really good at analysis. If you could put the data in a data warehouse, you can have a look at millions of records. Why would you go through Excel? Now there may be some ad hoc. If you don't want it in your data warehouse, you can't get it there. Okay, you're still going to be doing analysis in Excel. So long time till that goes away, I totally understand the logic of what you're saying. This is like today, imminent that you never do Excel analysis and excel again, but we just didn't add hoc analysis. I can give it a really good example. Last month, our AI costs are ballooning, our cloud, open AI. And it's actually a great thing. We're very excited about it. We're encouraging this a lot in our business. But we really wanted to understand by team. And so what we did is we, the start, we ripped down CSVs from all of our AI providers. And we merged it in with our had count data and to bring in the team dimensions and things like that. And we did all this in spreadsheets, but as we were doing it, we were like, we can never do this in spreadsheets again. It took one of our analyst eight hours to do it and clean it up and get it all correct. And so now we're, we're had a cloud code start building data integrations with these tools to hit their analytics API and grab this usage data and bring it back in our warehouse. And that's something we'll be building as fast as we possibly can because we know a lot this in the future. So I think there's always a role for spreadsheets as like the MVP layer of analytics. But once you get past the MVP, it should be in the data warehouse as fast as possible. I think that's a great example. Like I said, the for MVP, her kind of wire framing almost so to speak, Excel can't be be the flexibility modeling at the moment. Although, you know, we are seeing I was talking to someone, they're creating a tool in their thesis. And we'll get your thoughts on it is, it's, hey, you speak natural language to what you want the model to do. It uses Python. And then the Python then translates it to the cells in Excel. So you get a formula built model. So you're going natural language, Python, kind of the magic versioning control and all this, they're building a tool. But you still get, because let's face it, it's really easy. It's nice to see it in Excel. It's a comfort. You can make an argument of, hey, when does that change? And is that just because of the way we were raised? But that's a whole separate discussion. But I thought it was an interesting thing. Thoughts on, on something like that. It's interesting. Yeah. So Excel becomes the UI, but it has a backend, essentially with Python and it warehouse. Yeah. I think that could be great. I think it's just your one step away from building your own software, though. And I don't know why you, like, you that. I know what you're saying. I think there's some shortcomings, but I just, I'm curious to get your thoughts. So I think these are all probably phases along a longer journey towards what the fans for modeling is. I think that that could very well be a few years from now, or maybe two years from now. I don't know. At some point in the near future, near ish future, that we could be interacting with Excel and our data that way. But one of the things that really requires is your data modeling to be really good. And in a place that it can interact with it. And that's the hardest part in any FPNA role is the data comes. You spent, I think the statistic is like 80% of an ALS time spent cleaning data. And the rest of the 20% is actually analyzing that data. That's the part that is going to be really difficult in making that reality come true. I agree. No matter how you use the spreadsheet, how you use AI, and I think the spreadsheet for the foreseeable future of the AI, I'll bet it's off. I don't see the spreadsheet going away. I still get, see it being used. But there are definitely areas where AI may be better. If you can have it in a data warehouse and just do all the analysis, I agree. Why wouldn't you? But you can also do a lot of analysis, a lot of building. You can use AI in Excel and do it that way. And so there's, it'll be interesting to see how different people bridge all that. But I think you hit on the core of all this. You know, there's the data layer. And I think AI, and I've said this before, in some ways, increases the need for technical in particular around understanding data and data structure. If you want to build things in ways that you can really get benefits. I think there's kind of two things you have to understand with AI right now. We still hear a lot of learn how to prompt. Okay. Yeah. That's nice. And prompts make a difference. But I think we're at the point where the more important things is understanding how to give it context through skills, through instruction sheets. That can be through a prompt, but often it's much more than that. And I'd love your thoughts. So I think that's that context, the instruction, the skills, and then your data are really the two key things in prompt is probably third to getting good outputs, good results to really starting to build repeatable workflows and processes. So I'll let you speak to that. So managing the context that you can give to AI and not starting over from scratch every time. I think that's that's where we're talking about because I, one of the hard things and when you're just using a chat interface to try to automate tasks, it has memory, but like, it doesn't actually remember how it did something the last time that you asked it to. And so every time you ask it to, you give it a CSV and you say, do this analysis, you have to give all that context again. It's actually kind of exhausting. You try to do that. And the way that we're handling this is we're totally cloud-filled. If you've heard that term, like, you can choose red pill or blue pill, it's chat to be tier cloud. We're a hundred percent cloud over here. And I think it's because they have the best architecture right now for managing context, the concept of skills and hooks and agents and how those interact with each other. One of the key principles here is something called progressive disclosure, which I think is an actually really important concept to understand an FPNA. For anyone that's getting into AI, really is what I mean by that. The reason why it's important concept is because when you give a task to cloud, it has a context window. You can only have gifts so much information. You can't give it every report you've ever done or every notion doc in the whole business. It just can't handle that much context. And so what you need to do is you need to have a really good organization around how you give context to cloud. And the way that cloud handles this is through skills. And each skill has something called front matter where you give it a name and you give it a description. And it ingests every skill into every session. And so it can quickly search and find the skills that it needs. And once it finds the skills that it needs, it loads the entire skill document. And that's what makes clouds skills framework so powerful. And why things change so much last six months, in my opinion, is that now there's a way to manage the context layer and how you give context to the AI agent itself. And so we have skills that are we actually try to make them very modular. And so skills can interact with each other. So our skills look like here is how you do revenue analysis in our business. This is a very defined like these are the two tables that you use in our data warehouse. This is how you query them. Here's some example queries. We don't actually put that on the markdown. We actually have a folder within the skill folder itself that's called scripts. And we give it, you can give it reference files or string. Yeah. And into the skill. Exactly. First of all, so polls and when needed, it's better for context as well. How much it's using. So in the skill.md file, you explain what these queries are and when you use them and how you can change the granularity if you want to look at by day or month or week or whatever. And that's where I think a lot of power comes in in these recurring processes. So now once you define this once, it can do that the same way every single time. And so now when I ask Cloud Code, what is our NRR this month? I get the right answer 100% of time because there is a SQL query that is saved in this skill for MRR analysis and it will go and find that and it will load it and it will run it itself. And then it will just give me the answer. And it's never wrong. I think that's what's really exciting about moving from a chat interface to co-work or Cloud Code, you can use these skills to do them deterministically. And I'll just say it takes time. I think that a lot of us knew we should be documenting what we do before AI happened. There just wasn't enough value to actually do it because it was for somebody else. You were to hand off to whoever replaced you. And so it was an actually for a business decision. Yeah, exactly. Yeah, meet the need. All we're doing is now we have a good reason for every employee to be documenting how they do work. And the reason for that is because they can extend their leverage through AI and they can hand these things off. So the way that we take the approaches, we didn't try to boil the ocean. We started with one very specific thing and then the first thing we do is MRR analysis. That's what we get questions about a lot. That's what we're looking at a lot. So we just start with one MRR analysis and we actually what we do is we have Cloud interview us about our business and ask us questions. And then we go and we write the SQL queries and we give them back to Cloud and it says, oh, this is, let's go build a skill together and then it goes and builds a skill. And so like, Cloud is working with us as our partner to help us build skills that can use in the future, which just speeds up the process. And it also helps with like, there's a little bit of paralysis when you look at a blank mark down. You're like, I don't know what I'm supposed to write here. And honestly, I've never written a mark down file for AI. AI writes its own skills at this point. And I think that's, I read them. I make sure that there's good information in them. But I'm not going to start from scratch and write a skill file for Cloud. So that's how we manage context. We're fully captive to Cloud at this point. Eventually, we hope to move to a more sophisticated agent harness where we can swap in and out models. And I think there are a lot of, I've seen a lot of tools that are in development right now that will enable this. One thing that I'll call out is ramps glass, internal software. I don't know if you heard of that. But they basically rebuilt CloudCover and further organization. And they built it in a way that they can swap in and out models. They can share skills really easily. That's actually been one of the surprisingly hard things to do on my team is how do we share our skills with each other. We have a GitHub repo that we save all of our skills to. But it still is like requires you to pull it down every time and merge it into your own CloudCode and things like that. But when it's, you know, it's for your organization, you can share these things. There's a lot of network effects when you see what other people do with AI. And you start to get more ideas of your own. And I think that's that share ability and that teaching each other. It just speeds up the process of innovation here. So those are a few fairly unorganized thoughts on how we're managing context and AI and building with AI and FBA. We've talked a lot about context. I think that's really helpful. I think that helps people understand skills. And one of the biggest things I've realized is you can have AI help you write all the skills. And it would not be great to start with depending on how much detail you give it, but start somewhere. Like everybody should be least experimenting. I've been surprised how easy it is. And if you're not using Cloud, you can still use any tool to help you write an instruction sheet. And if you're working in Excel and you're using an AI agent, you can still have it go against that instruction sheet. Yes, it's not quite the same as a markdown file, but it's a lot better than trying to put it all on a prompt. Sure. Yeah. I've been testing it a little bit with chat, GPT and co-pilot to see how good it can do with just an instruction sheet for different things, some of the visuals I like to build in Excel and things like that. So I want to mention something about prompting since you brought it up. I think early on in AI, there was this whole concept of prompt engineering and that seemed very important. I don't find that to be very important at all these days. In fact, what I do is I think a lot less about how I'm telling Cloud something and just try to give it as much context as I can. And it feels very weird at first, but I use something called whisper flow to talk. Yeah, I'm familiar with it. I don't use it, but I'm. familiar with it. It's just voice to this dictation. It's like you Siri talked like you know text whatever it is talk to text but that has been a big game changer and like I just worry about it to cloud and the more conflict context I get at the better the outcome is and it doesn't actually matter if I tell it you are a CFO of a tech company and all this jargon that used to everyone thinks is so important it doesn't actually matter it learns about you it knows that you are in fpna it's going to apply that fpna context by itself because it's gotten smart enough to do that at this point and really all that matters is just giving as much context as you can and so talk to text is something that helps me a lot with with that I appreciate that and I definitely feel like you know probably early on prompting this huge context to me is more important now but I want to talk about one other thing that I mentioned and just with all this AI there's the data layer data structure we all hear garbage in garbage out we heard that long before AI and I think it still applies here so what have you learned on the data side to get the most out of AI let's talk a little bit about that in your data journey I've always loved data like I think as some I think I probably could have you know pivoted my career at some point and gone to a data team if I had just hadn't started on fpna team so my my journey in data has I from the you know the first time I got access to SQL like or to a data warehouse and writing SQL like I just loved it like I it's a for me it's just such a more elegant way to do analysis then doing something excel don't be wrong I love spreadsheets I mean I know all the shortcuts I you know I love being a spreadsheet as much as the next finance person but I also have developed a love for writing code SQL and Python code and so I've been working on that for a long time I am officially a nerd and I think one of the things that helps a lot was understanding the data lifecycle and I love talking to our data platform team you know they're the ones that are architecting our data warehouse and understanding the systems design and the thinking around why do we have a bronze silver gold layer it's called medallion architecture why do we do it that way understanding the system buying that has changed the way that I think about our own data structures and finance because if you all on likeness to a medallion architecture because it's what I just mentioned to for those uninitiated medallion architecture is just bronze equals raw data silver equals basically where you take that bronze data and you you model it a little bit it's still in like the source format and then the gold layers where you do like your star schemas and you're reporting your reports that you send to a BI tool and things like that so in a finance data architecture basically all we've ever done is work with bronze data where we go into our GL and we go into our sales force or our hub spa CRM and we just export CSPs directly how it comes and we put that in Excel and then we might use power queries something to create a silver layer where we clean up the data we do some you know we fill the noils we merge a couple columns if they're duplicates that type of thing we create relationships you know make sure that we have the same IDs between both systems that type of thing and then the analysis itself becomes the gold layer and understanding how we can take that same concept and put it into a SQL pipeline it just changed how I think about how we take our own data and finance and we put it into a way that that can be used by modern technologies and be a query by SQL so I think that's how my my journey has been just like learning how the data team does things relating it to how we do it in finance and then copying them as much as I can and I think this is something that especially like I knew I have two people my team now and one of them is a finance data analyst and one of them is an FPA analyst and their roles are actually merging a lot like the the FPA analyst is learning dbt and working in data warehouse and billing or SQL pipelines and the finance data analyst has already proficient in that area but he's also learning the strategy side of things and and it's interesting to see these roles kind of merge just because of the tools that make it possible for us to do both sides of the work we can now cover more than we couldn't have passed so that's that's what I've learned about data architecture and in terms like if you want to get in a specific data structure as we can we can do that next but from a high level that's how I've thought about how we implement data and FPA before I get there is want to you know kind of call something I think you're in a great situation where it's apparent you're given a lot of trust to access and work with the data right many companies that's not the case data team and they'll give you what you want and so you know I think that's where often you know power query a lot of those things come in maybe they'll let you hook it up and you can do that you're still going to see a lot of that so I think it's great that's the if you're able to do that and work with the team to do all that always anytime you could do something at the source you can build the architecture you can automate it make it scalable repeatable that's the ideal and so I love that you're sharing that but just you know so I'm sure there's some people listening going my business will never let we touch any of that it's the nature of startups you know I have a big blind spot that this is the largest company I've worked at circle cross 50 million in revenue in December and so you know I don't know what it's like to work in a billion dollar company and governance in different teams you know how they manage that data layers completely different so you're totally right to call that out I think what I would still recommend for any FPA professional regardless of what size company you're at and what level of access you have to data is to become friends with the data team understand what they do why they do it and you you'll just be a lot you'll you'll partner with them a lot better if you understand how that how they do their job at least at a high level you don't have to write code you're not your right right sequel or get into the you know write the data pipelines yourself but if you can understand how to speak your language and ask them questions like how are you thinking about semantic modeling and how can we start to leverage that on our team how can we use AI agents to access our data warehouse and and give it enough context to to query the data tables correctly and ask them questions like this and you can start to guide how they're thinking out their strategy and hopefully it can also in the reverse guide or strategy as well and how you implement tools on your team so I think a tight partnership between the FPA is critical 100% agree I'll share a little bit of my journey here because I think it informs this I started out of grad school I wanted to my finance degree I started on team I was called a financial analyst was the data team I was writing SQL for a year and a half and building a lot of different reports some of it was in access you know this is 2009 time frame so you know long time ago and it was invaluable the data classes I took in grad school because I did a master science and information management having that data understanding knowing some basic SQL and then knowing how to do power query and understanding the idea of okay here I really should be building a lookup table versus writing this nasty old if statement I learned the hard way my boss was like why are you doing this is super long case statements you know just figure out the logic and put it in a table and I kind of push back at first and one of the best lessons I learned now looking back so I mean huge believer in understanding data I think it's a core skill and it's becoming more of one in FPA does that mean like you said you need to be able to write an expert writing SQL or pipe on no I agree with you you don't need to be that you need to be good enough to have the conversations with the data team and to get get the information you need to do the job you have because your job is not to be a coder not to be a data analyst per se it's to help the business make better decisions which you need data to and for right so that that's a little bit of my journey and you know I was fortunate because I was working for a big company where that was usually siloed but I was in one side where when I moved over to FPA they left my access to the database they continued to let me query stuff my boss loved it it's like hey I need this airline data right let me go right to query here you go you know I need never had that before so I've had a very different perspective because typically you don't have that at a Fortune 500 company price shouldn't say that publicly they're like wait how do we make that to the security mistake but anyway anything else I don't know we want to go detail in the data structure I want to get into maybe some of the kind of favorite things you're doing or what you're trying to do now with AI but anything else around data prompting any other advice that you would offer on you know context anything we've talked about there before we get into some of the things you're doing pretty much covered at all I think yeah I don't think I have anything else to say there that's kind of what I felt like I think it gives people some good some really good ideas so what's what's the coolest thing you're working on let's start closing you're working on the AI right now yeah we're working on we're calling it our AIS we haven't decided on a name yet so I won't name it we'll just call it AIS for now by the way I don't I really don't like it when people name their AI agents human names I just find it I just find it off-putting it's weird like Lucy or something and I just find it weird I like I like you know at list or he'll next there's a hot take for you we interview the guy he runs one of the biggest AI agent businesses that's growing very fast in the finance space and he said one day agents will vote this they will have the right to vote I was I was like where's that coming from so there you go that not only naming them but giving them the right to vote that is a wild hot take so there that's the hottest take I think I've had on on the pod cat that one will come out in a week or two here so I'm really fascinated to see what people say with that one that's funny so my favorite thing we're working on is building our AIS and the last couple months what that's looked like has been really tightening our our data architecture like we didn't have our GL data in our warehouse we didn't have headcount data in our warehouse because these things contain sensitive information and we'd never felt the need to invest in trying to figure out how to protect that sensitive information. But now that with AI, we knew that would unlock a lot of capabilities for us. And so we decided to figure it out. And we worked very close to our data team to figure it out. We're using encryption on the columns themselves. And only finance has the decryption keys type of thing. That's how we're solving it, at least in the current state. So we've been really focusing on just building out all of our data source of use and finance. And one of the hardest ones actually is the financial model data. And we're building a system around that, bringing in all of our financial automatrix from our spreadsheet models, and then creating a driver tree that's in a table format to interpret how these metrics relate to each other. So we can go in more depth on that next question if you want to. The idea here is with all these data sources modeled, talking to each other, they have relationships between each table, we can use AI to interact with that data and to do a lot of our monthly reporting. So our first big milestone is automating our monthly finance report from end to end, at least the initial seeding of the charts. And we're probably-- I would say we're probably like 60% to 70% of the way there right now. We still have more data work to do to make it 100%. But it's already starting to speed up our building of the financial report. And what's really exciting about this, I think that those beyond the scope of FPNA is it creates a framework of understanding business performance. And it can actually be like the way that you orchestrate further agentic analysis. So what I need about that is if our financial-- if we do our BVA analysis, or just budgetverse actuals, we see that we miss somewhere. What we'll do first is we'll trace it back to the most granular level of modeling that we do in our financial model. So for us, we forecast our new business on the sales side down at the quota level. And we'll-- so we can't actually go any further than just what's the aggregate quota of the sales team right now. What we can do from there is we can say, all right, we missed on our sales team, Miss Quota. But let's go dig in and find which sales reps Miss Quota. And once it finds which sales rep Miss Quota, it can go and dig in and say why? Is it because their win rate dropped? Is it because calls book to drop? And this isn't things that we do in our financial model. But it's things that is data that exists in our data warehouse. And it can do basically flux analysis for a term that finance you can wear of to look at just historical trends and what changed over the last few months versus this month and why something has been happened the way it is. So I think that's what's really exciting is we build this from an FPNA perspective first. And it becomes like a jumping off point of even deeper analysis to speed it up from there. So those are the things that I'm working on. I think about a lot with AI. And I'm probably most excited about at this point. That is cool. What I heard there when you said AIOS, so I feel like you're building the data layer so you can have the intelligent AI really to be able to tie from finance back to what drove it, the operational metrics, and make it much easier. So AI can answer the business questions of what I like to call the SOA and now what versus just the what. OK, so we have these three Misses because these drop here's some things we might be able to do about it because AI is good at looking at those numbers and providing some ideas or different things that you then interpret versus spending six hours pulling down spreadsheets and a day later. Oh, I figured out it's Bob, Joe, and Pete. And I picked up the phone and I think it's related to this. That's an incredible value when you can reduce that time substantially, but not just reduce the time. You can help improve the decision. What's the one thing you couldn't go back to before Gen AI? If there's one thing that you could never give up that you now have, what would it be? Probably writing documentation. I mean, that was never very good at it, if I'm being honest. But now we're a lot better at it writing documentation because it's just part of our, you know, we merge a pull request for our data warehouse. It's part of the checks. You know, did we document everything we did and did you review it as well as the second important step there? So I don't think I could ever go back to writing documentation by hand. Now we have, yeah, I had to do that stuff for us. Oh, yeah, it's definitely getting a lot easier. All right. So, you know, a lot of people are trying to figure this out. Companies are all over the place from, we haven't started. Do we feel like we're getting huge value? You know, you're a long ways into your journey. I don't think you're, obviously, I don't think anyone's at the end point yet. You aren't nobody else is. But we have a lot of people listening that are trying to figure out how they really make sure they're preparing themselves for the workplace of the future. I'd love some of your thoughts. We're here to give one or two tips to people to help prepare themselves, to make them more value in this AI-driven future we're looking at. What would they be? My advice is the same as what I would always tell people when they asked, how do I learn how to build a financial model? It's do it for yourself. First, like my first financial model that I ever built were budgets for my own personal life. And they're extremely nerdy now. Like I still budget and forecast. And I have like retirement planning out 40 years. So I think that's where I, that's my main piece of advice is do it for yourself. I think there's a lot of really fun AI projects that you can do personally. I'll share a couple of my favorite ones. So I have my own personal cloud account. And I use cloud code. Something that's really helpful for your listeners is there's a framework called PI PAI. If you look up PAI GitHub, it's called personal AI infrastructure. It's basically just a set of skills that you can load onto cloud code. And it will make your experience the cloud code a lot easier than starting with scratch. So if you want to want to get started, that's probably how I'd recommend anyone to start. And so my favorite projects that I've done are I'm into lifting weights. And so I've never found a app that I'm happy with because it doesn't allow you to extract your data and do analysis on it. And there are some. But what I did was I vibe coded a-- I wouldn't call an app. Well, all I did was I created a set of, like a database for working out. It's got tables for sets, exercises, workouts. It's just three tables and those all relate to each other. And so I built this on my own laptop. And it's just a local database. You don't have to learn how to host or anything like that. There's just local databases called DuckDB or SQLite that you can go and start to build these without any technical experience. And cloud code will build off for you. And so I built this database. And then I built a system of being able to text cloud code and say, I just completed bench 205 reps. And it will go and log in the database. You use whisper flow to do that. Sure, yeah, of course. Depends on listening to music or not. Sometimes I don't like it to interrupt my music. That was a really fun project. And I think that's a really great entry point into learning how to build a database. And if you want to write queries against that database, do it with cloud code and say, what's my progress in over the last three months of weightlifting? And it will show you a chart and show you what your Estimated One rep max is and things like that. That's one of my favorite projects I've done. I mean, it's kind of nerdy, but I build these things on a weekend. Another one that is funny and also nerdy is my wife and the only time we ever argue is when we need to make a meal plan. And it's not because we can't decide on what to eat for the coming week or two. It's because neither of us know what to eat. And so we don't know what to put on a meal plan. And what we did was we built our own recipe manager app. We didn't actually build it. We found an open source one. It's called Mealy. And we-- I install Docker. I clone the Rebo. And now it's hosted on-- it's just host on our home network. We use tail scale to interact with it through our phones. So we can track all of our recipes as are our favorite recipes in there. And then using cloud code, you can take all those recipes and the ingredients and the recipe end is really good. It has a really good data model. And so it creates a lot of metadata about recipes, saying like, this is the primary protein, chicken or beef for whatever. It uses these ingredients like onions and garlic or maybe something more niche, like tomato paste, for example. And it will intelligently find recipes and recommend things that go together because they use common ingredients. So you're not buying a big can of tomato sauce and then throwing away three quarters of it because you can use the rest of it. It's smart and can start to help you plan your meals better that way. And now our weekly workflow is we text our cloud code, our AI. And we say, what should we eat this week? And it gives us options. And we narrow it down. And then it spits out a grocery list. And we take that to the grocery store. So those are just a couple personal projects that I've worked on. And I think anyone can get started with that. At this point, cloud will just tell you how to do something like this. You give it something that's very abstract. I started with, I don't want to meal plan ever again. What are some ways that I can do this? And you just go through the conversation and it helps you make these architectural decisions. And it does research for you. It goes and searches the web for things that already exist. And then it'll build it on your machine for you. And I think those are really great ways to start to learn how you can use AI at work if you're not given the freedom to do that. I work right now today. Great advice. Thank you for that. So I want to move into our FPNA section, looking for just some short answers here. I know we're a little long on time. So we'll run through these. What do you think the number one technical skill is today that FPNA professionals need to master? I think systems design is where I'm-- and that sounds like-- I don't know, that's probably-- people hear that and probably don't know what it means. And so me, what that means is like, don't go learn. I wouldn't recommend going to learn Python today. For example, I would recommend understanding the data lifecycle. How do you extract data? How do you load it into your warehouse? How do you transform it from there? And then there's now a new step, which is how do you give enough context around that data so that I can interact with it? So I think thinking about systems design from a data standpoint and then the process standpoint, how those interact together, I think, is probably the technical skill that I think people need to start to master. How do you use technology to accomplish the work that we did in the past and what are the right tools and how you connect them together without a human? Is the technical skill, I think, is the highest leverage over the next decade. Thank you. You're not the first that's mentioned data. There's been a few, a little different angle, but I've heard that before. So not everybody's going to think you're weird, just most people. What about that softer human skill? This is a good question as well, because one of my thoughts that I've had about AI is it requires FF professionals to be both more technical and more strategic, the median in technical skills and strategic thinking is quite replaceable by AI at this point. And so I think we need to get deeper at understanding how the business operates, which means you have to understand what actually makes a difference. We talked about this a few times that was broadcast already, which is what levers can you actually pull and influence as a business? How do you actually influence them? And who are the right people to do that? And I think that's where the strategic impact of FFNA is really most important is to help and to create change without authority. We don't have any authority, but we can use the data that we have and our own frameworks for modeling the impact of decisions to help influence people to make decisions that will be best for the company over the long run. And so I think that soft skills are probably-- I have a wrap up, but I don't want soft skills business partnering, I think, for lack of a better term. But it's really understanding the business deeply, business, acumen, plus how to influence the people who are actually boots on the ground, making decisions and shaping what the business is going to be in the next year. Two skills in there, but I get what you're saying. The influencing the business partnering, the strategy, all those things. So thank you for that. All right, I'm going to ask one more on FPNA that I wanted to ask you. And then we're going to move into just probably two questions I'm getting to know you. So I know you use Google Sheets. I mostly hold it against you, not completely. But one thing you like better about Google Sheets over Excel. So I'm at Google Sheets Convert. My last company before Circle, I was die hard Excel. I pushed back on anyone that asked for Google Sheets. And then I came to Circle and I saw the light. And Google Sheets-- I mean, the number one thing over Excel is multiplayer mode. Like, the collaboration of Google Sheets is just far and away. It's just far beyond what Excel is. I really like the import range function, although I'm starting to hate it, our models are getting complex enough that import range is really breaking down for us. We're moving to a new spreadsheet provider called Ro0 because of that. Well, you aren't going to Ro0, I'm familiar with quite a bit. We'll have to talk offline. I don't want to take up a lot of time on the podcast. But for anyone who's interested, Ro0 is doing some unique things you can find them online. So you and I are going to talk on that. So the reason why I like import range is because it's a really nice way to create modular financial models. So you can have like, we model our community hub products separately from how we model our email hub product, which is like an email marking tool. And in Excel, you have 100 tabs to handle this type of complex modeling. But in Google Sheets, what you do is you create a load sheet in a single workbook that you import into a consolidation workbook. And so you can do this in Excel. You can do it through stitching things together with Power Query or the worst sin of all Excel sins, which is linked workbooks, which hopefully no one on your podcast is using the need to workbooks. And it's just-- but neither of those are a great experience. But in Google Sheets, the import range function is dynamic. As soon as you make a change, theoretically, as soon as you make a change in the precedent model, it updates in the dependent model. And that's probably the thing that really sold me on Google Sheets was that we could build these modular financial models using import range. What it reminds me of, but they only work within the spreadsheet. You have the camera function where you can make a change somewhere and see it instantly in the camera's screening Excel. But that's only within a file. Very different from import range and that you're dealing with the different worksheet. But that kind of that concept, but it's doing it across the Sheets. Yeah, yeah, exactly. I just never heard the camera function. That's really interesting. I don't know why you would use that instead of-- You take a snapshot of something somewhere else in your model. And if you make any changes to it, you could see it in a different area. So if you had two windows open, you work with one, you could look over the other and see what it happened there and impacts. There's some places where it records different things in your model. So even if you make changes and you know it's going to impact out part of the model, even though you're working in a different area, you could see what happened there. Anyway, all right. We've nerded enough on that. Roe zero, Google Sheets Excel. You're using them all. All right. To get to know you questions, what's your favorite musical album of all time? I know you like to listen when you work out. I did see this and I forgot to-- I know the album I just don't remember the name the second. I don't know the back counts. Can you currently claim that you know your favorite album when you don't even know its name? This is disappointing, Derek. It is this point. OK. I'm a big country, a music fan. I grew up in Houston, Texas, actually, in the process of moving back there, which Paul may not even know. But I love Loot Cums. He's one of my favorite country singers. And every song on the album, Father and Son's, is just so good. And I think it's because I'm a dad. I became a dad at the same time that album came out. And it was just like-- I related a lot to it. And it's still one of my favorite albums. So it's a two-all time. All right. Last question. I know you like to travel. You love credit card points, favorite credit card. My favorite credit card is the one that fits with what I want to do. You have a whole database that you vibe coded to track all your points. I do. I do. I do know. And I have skills in Claude, my personal Claude, to go and find the best credit card offers. And it's automatic. They don't apply for me. I don't give it my social security number. You don't let co-work life for credit cards for you. You have a limit. I do have a limit. But yeah, my favorite credit card is whichever one gives me the most points and the points that are valuable for what I'm trying to do. So I don't have one specific credit card. Love it. All right. Well, if someone wants to learn more about you or get in touch, what's the best way for them to do that? Yeah, you reach out to my LinkedIn. I check that at least a few times a week and I'm always good about responding. So unless you're a software vendor, then I probably won't respond to you. But feel free to reach out. Always happy to chat and talk shop. All right. Well, thank you so much for joining me, Derek. Enjoy the conversation and thanks for being on the show. That's it for today's episode of FPNA Unlocked. If you enjoy FPNA Unlocked, please take a moment to leave a five star rating and review. It's the best way to support the FPNA guy and help more FPNA professionals discover the show. Remember, you can earn CPE credit for this episode by visiting earmarkcpe.com, downloading the app and completing the quiz. If you need continuing education credits for the FPAC certification, complete the quiz and reach out to me directly. Thanks for listening. I'm Paul Barner's, the FPNA guy, and I'll see you next time.