From Hype to Workflow: How AI in Excel Is Actually Helping Finance Teams Today

Welcome to the future finance show where we talk about Future finances brought to you by Qflow dot AI the strategic finance platform Solving the toughest part of planning and analysis B2B revenue align sells marketing and finance seamlessly speed up decision making and lock in accountability with Qflow dot AI Welcome to future finance today Glenn and I talk AI news and We get up on our soap box for a few minutes as normal. So if you want to listen to two finance guys rant about AI This is the episode for you We're we're gonna get started with what is new out there? It's been a while and feels like things keep changing on a daily basis So Glenn why don't we get started with you? What's some of the new things you're seeing out there? I guess the biggest in the last couple of weeks would be Clawed for financial services, which is pretty interesting in just in that in the way that You see the frontier models making a move to go vertical and it's interesting that they picked finances as a first vertical the other one is Open AI continues to roll out Product that are I don't know honestly they rolled out open AI agent and you know my whole thing with agents And it's been the least impressive roll out that I've ever seen from open AI But we if we have time later we can we can talk about that I know we had some sort of specific things we wanted to cover But I think that sort of that capture the zeitgeist and interest of finance folks that clawed for financial services is a strong statement and Open AI agent is a strong statement on its own and maybe we can talk about that later about how There's ways to deploy and ways not to deploy agents and I'm doing agents in the air quotes right now as I say the word Yeah, and since we're audio only you won't see that but you if you see Glenn's face you'd understand What you're taking up you get chance to test clawed or so I reached out here's the interesting thing The biggest thing with clawed for finances. It's got a 500,000 token context window which is Which is great and that's more than what you have an open AI enterprise or pro It's got built-in Finance connectors that you can build this using API right now, but with this vertical it's they're making it more Off the shelf friendly. It's not that hard to set up an API connection and all that but this the clawed for financial services Comes with integrations and direct into snowflake fact set S&P capital IQ Morningstar box. It's got direct integrations that you don't have to do the API Bloomberg right it keeps in mind sort of the audit friendly Answers like it sites everything so that you can go back and you can you know retrace your steps. It's got some built-in clawed code that's specific for Python and it's got a finance specific prompt library so it's I don't know I mean all the stuff that they did any consultant could come in or internally in your company even someone could build these very same Connectors, but it's it's more of a signal. It's not really oh, it's super valuable right now You have to have over I think it's a minimum of 50 users on it and so it's it's not gonna be for everyone Truthfully for the price of it. I don't know why somebody wouldn't just build their own that is specific to what they're doing whether you're in high Finance or corporate finance and accounting whatever but it's I think it shows the push to productize these foundation models and Get things into the hands of people and it's super expensive to build these and train the models and all that So they're really trying to figure out ways to grow that top line to support their massive spend to keep these things Advancing right and we all know we're an FP and hey both of us and we've been a CFO as Almost always be to be productization is much bigger price tags, right? You're getting $20 per user per month and a couple cases maybe a hundred or 200 with open AI That's just that's a drop in the bucket compared to a $5 million dollar deal with someone like Amazon say you know and Throbic to use a company wider whatever it might be you know you got to have a at $20 a piece How many do you have to get to get to you know a hundred thousand or five or a five million dollar price tag You might get with some of these big huge companies. Yeah, and I guess what one thing I left out This is came a week before somewhere around the same time that open AI rolled out the agent They also have this new they're sort of following the Palantir model where they will embed engineers in a company Minimum price tag for that is is $10 million so what the idea yeah, but the idea is you know There's all these consulting companies out there that are doing what I do which is saying we can help you implement this and Put it into your company and make it so that it is usable and you've got these Agente workflows that actually do tie in to your regular workflow and you can trust the answers and you do have an audit trail and all that and Open AI is realizing that's cool outside consultants could do it But someone from inside the company coming in there and Customizing this for the biggest companies in the world obviously that are willing to spend 10 million a year It's a it's a great model and Palantir is shown that it works But think of how many companies there are out there that aren't spending $10 million a year just on their open AI integration Limited set by how many of those do you need to make some pretty good money exactly? Well, it's all relative right when you think about the hundreds of millions of dollars to For each training run and all the capex they're spending I did see something Paul. Let me make sure I'm quoting this correctly so AI related capital investment now accounts for over one third of US GDP growth I mean that's crazy and that's coming from you know That's not coming from these other companies that we're That are just using AI. This is this is the frontier models But that is an insane amount of expense that's really driving the economy forward right now Yeah, I mean it's why Navidia has such a crazy valuation right and question is can it maintain the market share it can probably not probably over price like every bubble but for the moment It's you know, it's the Wall Street darling so to speak So all right. Yeah, that's Amazing how much money we're spending it's like the number of companies and other things are out there without those few exciting One thing I want to talk about and I know you can Relate to this because you've mentioned this when you go out and teach people and you show them hey look You can do this in chat cheapy here. You do this in clods kind of hey, that's nice. That's cool But one I got to get I got to give it my data to I got to take it back into some other system most of the time Right, it's not really in the workflow We've talked a little bit about this before and I think we've coupled different things We've seen a lot of that you know the ERPs the FPNA is trying to add it to the workflow And I think that's for the mass of finance, but then there's how do you add it into the the workflow that everybody in finance uses? Excel, right? What is it a billion people globally and if you work in finance You've used Excel in your career. You might be at a company that's using Google sheets today And so there's even you know Google sheets, but how are we getting these AI in a way that they can really start to analyze the spreadsheet Because I think there's a couple things have to be solved there is you know, you got to make the data easy to get to You need to have some kind of LLM or whatever some model that's analyzing the data and then you got to have an easy way to Pull out the data from the spreadsheets and Compress it decompress it get it in the right context so it can be analyzed Because spreadsheets by nature you can't just throw the model on top and they're gonna look and rose and columns and really tell you what's going on Right, you have to have some kind of way to mark up the spreadsheet and so I've been seeing a lot lately. I've talked to a lot of companies and I think we're really starting to make Progress there toward hey, we can upload a file and it can answer a problem I've seen one that took you know financial modeling world cup Problem solved it. I've heard of somewhere you know, they're doing the work of a first-year analyst and most managers are Preferring the work of the spreadsheet another one where they found a big issue of Millions of dollars in a spreadsheet and so what I'm starting to see They're getting to that point where we could really start using some of these add-in tools Inside our spreadsheet As part of our workflow throughout the day or using some of these browser-based Data analytics tools that have data connectors that have different LLMs and they have Python and R all built-in So you can really kind of go back and forth I think sells the best right now and that because the browsers aren't as good as having that built-in Sandbox that you can just play in as you're asking the AI So I think we're starting to see some pretty good tools Have you seen much in that space or what's your thoughts there because I think until we can get it really integrated into the workflow It's great, but it's not Gonna have a huge impact. Yeah, the funny to me. This is like that The quote if the mountain will not come to Muhammad then Muhammad will go to the mountain This is this is we've been trying to figure out how to upload our spreadsheets in the chat GPT or and get chat GPT or or clod or whatever to to analyze and understand them, but instead it's about to add in and about bringing AI. And this is what I know this is Microsoft's vision. We say it all the time. They just are far from getting there right now. But when all that context that is in spreadsheets and you just think about even for humans, if you gave one of your models, we try to make them as clean as possible, but you go through so many different variants and versions and it's just it's very hard to follow these models and they get passed around, they get changed. Some there's always some exception cell where it doesn't fit everything else and there's a reason behind it or assumptions and drivers changing all that. And it's very hard for AI to have all that context and understand it. Even if you've got, you know, it's like having good good comments in your code, even if you've got a sheet that clearly explains what your assumptions and drivers are and what's driving all the formulas. It's still a lot for the LOM to take in, but when you inside the sidebar or inside the application, you can actually highlight an area and ask questions directly within the dock. I've seen better, much better results because anytime I upload anything to to chat GPT or or quad or whatever, I strip out all the formulas, I make it as you strip it down to a CSV, make sure it's as machine readable as possible and then work there, which is great. You can do some pretty cool stuff and it basically, basically I turn financial statements into a data frame so that the computer can understand them. But then basically creating it, you're loading it as a data table, right? Like it's a database almost. Yeah. But then to your point, how are you then going to get that out and have something you can use? You can have a great chat interaction where you do some cool modeling and scenario analysis and all that. But then you're left going back to the spreadsheet and having something that you can have as the permanent record of this and these chats outside of Excel, just their their ethereal. They're just, you know, they open up a Python session, run all the calculations, you get great results, you come back to the chat the next day, try to run the results, that Python session is closed and it doesn't have history of your data. Ever feel like you're go to market teams and finance speak different languages? This missile alignment is a breeding ground for failure. In pairing the predictive power of forecasts and delaying decisions that drive efficient growth. It's not for lack of trying, but getting all the data in one place doesn't mean you've gotten everyone on the same page. Meet qflow.ai, the strategic finance platform, purpose built to solve the toughest part of planning and analysis, be to be revenue. Qflow quickly integrates key data from your go-to-market stack and accounting platform. Then handles all the data prep and normalization under the hood. It automatically assembles your go-to-market stacks, makes segmented scenario planning a breeze, and closes the planning loop. Create airtight alignment, improve decision latency, and ensure accountability across the team. I totally hear you. Those are big problems. I think having platforms that can start to integrate all of this is huge. I've seen quite a few lately and I think I'll break it into a couple groups. There's those that are trying to create a new application to do financial modeling. We're there trying to say, hey, we're going to build our own whether it's a spreadsheet or application with AI and do it. That's one approach. I think the one that I'm most excited about right now is the sidebar within Excel. There are also some that are trying to keep create data analytic platforms allow you to store everything in notebooks and use different code and LLMs and those are great. Until you integrate that spreadsheet with them as well, I'm not sure they're going to get mass adoption. You have co-pilot, but co-pilot is designed for the masses. It's not for the finance audience. Can it do finance tasks? Sure. But I know you've used it. I've used it. The experience is it's just not there yet. I think even Microsoft recognizes, hey, we're really trying to build this for the billion people, not the 10 million or 50 million or whatever it is that are finance and generally more advanced. That's their challenge. You got Gemini for Sheets, but again, that doesn't work in Excel. Then I've seen some tools. You and I interviewed the founder of Rosie. I played with that. I asked it to, hey, right, a lookup formula. Did the Excel look up for me? I've seen some other videos with some cool things, but I'm starting to hear of a bunch of others and starting to chat with some of these over the next few weeks or have chatted. There's one that's being worked on called celery. The guy's built his own language to pull it out everything out of Excel with a markup language. Then applying the LLM to it, which to me makes a lot of sense because right, if you just try to read the file unless you're doing like you do, put it in this nice CSV format, which kind of defeats the purpose, you're not going to get great results. That's what I'm starting to see. That's where it has me excited because if you can do a good job and start to figure out context and font and how to pull it up in a markup language, they can read it in a machine needs to and then start giving us advice and feedback of what's wrong. That's where I'm starting to see some pretty good results from all these different tools and it has me excited that we are moving toward a workflow that could be very valuable straight in our spreadsheet instead of let me go out to my browser, let me copy and paste, let me get something that I don't know how I really get this back into my spreadsheet. I'm interested in seeing what celery does because I've tried multiple different approaches to try to see how much data I can jam into a single file and have the LLM be able to work with it well. I tried converting to JSON a couple of times for financial statements. There is a structure to that and I thought it would make sense and be easier to understand. I think agents, my favorite word, do understand JSON better but truthfully just in a straight chatbot, it actually did better with a CSV than it did trying to convert to JSON but I wasn't putting a lot of intelligence around it. I was basically just created the kind of like XML, you know, just created the structure of this, what you're looking at and all that and it didn't, but it seemed like CSV worked best but if they're actually spinning the effort to make sense to transform the data in a way that makes more sense to an LLM, I could see having really good results as long as it comes out, it's transformed, they do whatever magic they're doing, waving their AI wand over it and then it comes straight back into the spreadsheet where you have that is your system of record or you're aggregating everything, that's pretty interesting. >>And that's what I think they're trying to do. Microsoft has done their LLM whether they're trying to do things. I know with the celery example, I know he's using some Python markup language still very early, he doesn't even have the beta product but he showed me some, somebody early, I know like take shortcut, it will build the model for you, it doesn't get font and format right but it does a pretty good job, you know, it solves some world cup cases, I know Rowan's out there that's supposedly pretty good, you know, we talked to Rosie and so I agree with you, the key is and again, I can't say how good they are but I think the key is how do we get to the point where we can really efficiently and cost effectively pull out the data and some kind of markup language that the AI can really analyze it and then be able to bring it back into the spreadsheet where we can use it. I think that's kind of the key in all this as we start to get, as we get better at doing that, I expect this to see huge benefits in what we can accomplish working with AI within a spreadsheet. >>It goes back to where we are with AI right now, there's so much hype across the board and you see these really cool use case examples and they fall down when you try to go from what I call the parlor trick of look at this cool thing I can do to okay but I'm at a real company and we need, I'm getting pressured by management, board, investors, whoever that we need to implement AI so I'm trying to but there's if I'm doing this myself I can't find a way that I can use this reliably and I'm certainly not replacing any people with it and there was a see Gartner just published a prediction based on their research they say more than 40% of in progress, agenteic AI projects will be canceled by the end of 2027. Think about that we're all rushing out there and this is Gartner see the hype cycle they see what's going on they see the pressure that people are under and they are saying now that over 40% of agenteic AI projects will be canceled by the end of 2027 and the reason for that I think is people are naively expecting so data foundation is huge but it's more than garbage in garbage out the expectation of what these agents can do is just wrong there was another reasoning level you're expecting of them is not there. Yeah so there was a Carnegie Mellon did a study where they had they created the agent company where it was a company it was going to be completely run by agents and no human involvement at all because this is all going on. Sam Altman's idea and a lot of people are saying this, the $1 billion unicorn, that's just one person and a bunch of agents creates a billion dollar company with no employees. It's all agents. But Carnegie Mellon's research, when they tried to turn this over just completely to agents, the best performing agent out of their whole study completed 24% of the jobs assigned to it, most completed about 10%. And it cost each agent on average of $6 to complete an individual task. And they were getting hung up by even the slightest exception, things like, I'm supposed to be going and doing something, but a pop up ad comes up on my browser and I'm completely shut down. I don't know what to do with it. I mean, that's the kind of stuff that when you just, again, that you've got all the hype men that are out there saying, how great I, and I agree with them. But if the idea that you're just going to take these off the shelf tools, and plug them in right now, until we get to something like a GI, that's not going to happen. You need, I'm not even going to say engineers, you need AI architects to go and map all this stuff out for you. Maybe if you have a small business that does things that you don't have to be auditable and you have a tolerance for error, maybe you can do it. But some people are being sold a bill of goods around agents right now. And even the cool stuff you can do in chat GPT where you have to have a data trail and you have to be able to reproduce what you did, we're just, there's a lot of gray area right now and people are getting frustrated because they can't make it work in the real world like what they see these 20 minute YouTube demos. Glenn, have you ever seen the Sardinay lives get with Lindsey Lohan and Debbie Downer? Do you remember Debbie Downer? That's what I didn't say. That's just out right now. No, I think it's great. We got to bring the reality. We're seeing amazing stuff, but yeah, like the 40% ahead and seen that number, not shocked. I mean, how many of these companies are going to go away? Like as I look at all these different workflow tools, I'm excited. I'm going to get ready to do some project to test a bunch of them, I'm working with someone else to let people know, hey, where are they really at? And what can they do? Because mostly I've talked to the founders. I've seen their videos, right? And you're going to get hype there. Then there's the reality and really starting to figure out where are the weaknesses? Like I'm excited for it. And I think we'll get to that point where we will be working in Excel. And I think we, some people are already back. What, what are the limitations? What can we really do? You know, what are the challenges? And I think that's where like what you're doing, what I'm going to do testing it here is reading these things and bringing the reality of just remind people, the billion dollar unicorn company with nothing but agents, we're not there. That's reality. Unless you, you had a company you can get to a billion dollars with a very basic path with not much reasoning. Then I guess you can do it with agents, but I have a hard time picturing a company to get to a billion dollars without much reasoning. If they can, then I could be a billionaire. I don't have the thing much. So, you know, probably a good spot then. I think as a wrap up is just a reminder to people look, it's moving fast. There is hype. There is reality. Sometimes it's not as easy as you think to cut through it, even glad and I struggle with that sometimes. And we're going to continue to look at these things and bring you our take. Where we think it's going, what we see sometimes we'll be right. Sometimes we'll be wrong and sometimes we'll be honest soapbox. But we hope you enjoy us for the journey. Final thoughts, Glenn. There are practical applications for AI that you could be using right now, but it's naive to think that you're just going to turn everything over to the bots and use off the shelf claw at or Gemini or chat GPT or whatever. But don't give up hope. The technology is the coolest I've seen in my lifetime. It's just we're AOL with the screeching modem in the early days when they were mailing the the discs out to everyone. And Adam in our college dorm, one of the guys used the discs he colored them each different. It's stuck them around all his cables to label them for his entire entertainment system. He had so many AOL discs. So I date myself. That was college, which tells you how old Glenn and I are because they're not that far apart. We're similar in age. Believe it at that, right? Well, leave it as being old and it being hot outside and me being cranky. That's all right. Well, there you go. It's hot, cranky and we're old, but we appreciate you joining us for another episode. So thank you so much. If you enjoy the show, reach out to us. Leave a review, leave a rating. We'd love to hear from you. You can find us on Lincoln. We're there plenty. So please drop us a note and thanks again for listening. Thanks for listening to the future finance show. And thanks to our sponsor, Qflow dot AI. If you enjoyed this episode, please leave a rating and review on your podcast platform of choice. And may your robot overlords be with you.