An AI Excel Tool for Finance Teams to Speed Up DCFs and Scenario Modelling with Nico Christie
In this episode of Financial Modeler’s Corner, host Paul Barnhurst welcomes Nico Christie, co-founder of Fundamental Research Labs and creator of Shortcut, the superhuman Excel agent. Nico shares insights into his fascinating journey, from being a professional athlete to leading groundbreaking AI projects. Together, they explain how AI is revolutionizing financial modeling and making Excel smarter than ever before. Nico also discusses his work at Fundamental Research Labs, where they aim to build digital humans using AI, and the development of Shortcut, which aims to streamline the way we use Excel for financial modeling tasks.
Nico Christie is the co-founder of Fundamental Research Labs and the CEO of Shortcut, a cutting-edge AI tool designed to supercharge Excel. Nico’s diverse background spans professional athletics, corporate finance, and AI research. His work at Fundamental Research Labs focuses on building AI that mimics human-like qualities, such as emotional experiences and consciousness. Nico has a passion for both technology and the real-world applications of AI in tools like Excel, making complex tasks faster and more intuitive.
Expect to Learn
How Shortcut is revolutionizing Excel for financial modeling with AI.
The process behind building digital humans at Fundamental Research Labs.
Nico’s journey from professional dunker to AI innovator.
How AI tools like Shortcut are transforming the financial modeling landscape.
Insights on the future of AI in Excel and the response from companies like Microsoft.
Here are a few quotes from the episode:
"I just think Excel is the most beautiful design software ever created by humans." - Nico Christie
"The goal of the company is to build digital humans, giving AI all of the fundamental human qualities, not just intelligence." - Nico Christie
"People who were initially afraid of AI are no longer afraid once they see how it can be used in software engineering." - Nico Christie
Follow Nico:
LinkedIn - https://www.linkedin.com/in/nicochristie/
Twitter - https://x.com/nicochristie
Follow Paul:
Website - https://www.thefpandaguy.com
LinkedIn - https://www.linkedin.com/in/thefpandaguy
YouTube - https://www.youtube.com/@thefpandaguy
Follow Financial Modeler's Corner:
LinkedIn Page- https://www.linkedin.com/company/financial-modeler-s-corner/?viewAsMember=true
Newsletter - Subscribe on LinkedIn-https://www.linkedin.com/build-relation/newsletter-follow?entityUrn=7079020077076905984
Sign up for the Advanced Financial Modeler Accreditation Today and receive 15% off by using the special show code ‘Podcast’.
Visit https://bit.ly/3AK5dYo and use the code “Podcast” to save 15% when you register.
In today’s episode:
[02:15] - Introduction to Nico Christie
[02:38] - Worst Model Horror Story
[06:32] - Upcoming Projects and Exciting Roadblocks
[09:14] - The Journey to Dunking
[17:21] - The Power of Excel's Top Buttons
[20:17] - Reviewing vs. Starting from Scratch
[21:38] - Struggles with Formatting in Model Reviews
[32:35] - The Impact of AI Tools on the Profession
[37:26] - Favorite Excel Shortcut
[39:54] - Financial Models as Decision-Making Tools
[42:34] - Wrapping Up
Full Show Transcript
[00:01:32] Host: Paul Barnhurst: Welcome to Financial Modeler's Corner. I am your host, Paul Barnhurst, aka The FP&A Guy. This is the podcast where we talk all about the art and science of financial modeling, with distinguished guests from around the globe. The Financial Modeler’s Corner podcast is brought to you by the Financial Modeling Institute. FMI offers the most respected accreditations in financial modeling, and that is why I completed the Advanced Financial Modeler. I'm excited this week to welcome to the show Nico Christie. Nico, welcome to the show.
[00:02:06] Guest: Nico Christie: Hey, Paul. Thanks for having me. I'm really excited to be here.
[00:02:08] Host: Paul Barnhurst: Yeah, really excited to have you. I'm thrilled to continue the conversation we've had prior to this episode about what you guys are doing. So a little bit about Nico's background. Nico Christie is the co-founder of Fundamental Research Labs and creator and CEO of Shortcut, the superhuman Excel agent. Nico started his career as a professional athlete and corporate finance consultant, before becoming an engineer and starting a research lab out of MIT. All right, so we start this question for every guest. I'm sure you have a horror story. Worst model you had to deal with. Tell me that story.
[00:02:43] Guest: Nico Christie: In my corporate accounting days, we were doing, like a finance transformation for one of the fortune 500 companies. Legacy 50 year old company. You know, and unfortunately, I think they were doing a model redesign. They wanted to change the structure of the entire, you know, employee base there. And one this model was built. Through really combative interviews because no one wanted to share the data. And then it was built by a junior analyst that had hard coded everything across 50 sheets, hundreds of columns and thousands of rows. And it was handed off to me at this point where everything was hard coded in that fashion. It became my sole job for ten hours a day for, I believe, over two months to update and manage this deliverable and then at the very end to change it into using actual formulas and then to present it to the CFO. You know, the real truth is I handed that off and switched jobs before my job was done, and it was too much of a nightmare.
[00:03:34] Host: Paul Barnhurst: You're like, I'm out. I can't handle it.
[00:03:37] Guest: Nico Christie: It. It was inhuman.
[00:03:38] Host: Paul Barnhurst: Anytime you have a model with nothing but hard codes in multiple sheets, you don't really have a model. You have to let you have to rebuild. And that's just painful.
[00:03:47] Guest: Nico Christie: It was a data dump, and it was like sad, ugly data. Yeah.
[00:03:52] Host: Paul Barnhurst: I'm gonna have to remember that saying, sorry, I can't work with you. That is sad and ugly data. All right, so you're the co-founder of Fundamental Research Labs. Can you tell our audience what that is? Let's start there and what your experience has been like as the co-founder of the lab?
[00:04:09] Guest: Nico Christie: Yeah, absolutely. And before I do, I know I feel like I made some waves in the financial modeling community, and I wanted to set some expectations and introduce myself. About a couple weeks ago, I introduced a shortcut, which is a superhuman Excel spreadsheet agent. I actually worked with the CEO of the Financial Modeling World Cup, and he tried it and it was pretty impressive. And I quoted him in the release, and I think AI is a new scariest thing to a lot of people, and I think it definitely caught the world by storm and specifically the community that I mentioned. So I'm excited to introduce myself. I'll share more, but Fundamental Research Labs is a research lab that's spun out of MIT about 18 months ago. My co-founder is a former computational neuroscience professor at MIT, faculty professor, and was the head of a research group called the Meta Conscious Group, where we had 15 researchers all about exploring what it would look like to build a human brain using AI. The goal of the company is dystopian as it sounds, is to build digital humans. And when we say that, we mean giving AI all of the fundamental human qualities, not just intelligence, but there's long term autonomy. There's collaboration. The way humans do that there is from tool use to emotional experience to consciousness. All of these things that exist in the human brain are fundamentally malleable, and in neuroscience they're not dirty words, but in AI they kind of are. And the first versions of these digital humans, we actually built agents that could play Minecraft and Roblox, and we made some cool demos.
[00:05:24] Guest: Nico Christie: We let people play with them. We got millions of players. We ended up raising about 10 million from Eric Schmidt and some big funds. And then we moved to San Francisco, more specifically Menlo Park, and we started scaling up the research team, and then we gave them access to computers. And at that point, we broke the record in a very important benchmark for computer use called the OS world. And it measures basically the ability to do Excel, PowerPoint, email, video editing, photo editing, whatever. And we doubled the limit from anthropic. And six months ago they were even higher than what the operator has now. And at that point, I still felt, you know, no one wants an AI that's 50% accurate, right? What would it look like if I really constrained the action space to something I was passionate about? And I let agents do what they do best. And I as a, you know, as a software engineer, I know what that is. I've been building engines for a year and a half with this team. So I led a special team for shortcuts. And the result is, you know, I wouldn't say perfection. It's far from it. But it is the strongest AI on Excel by far. And we have benchmarks about this and we are opening it up. And I think by the time this is live, it should be open to everybody.
[00:06:25] Host: Paul Barnhurst: Exciting. No, I'm sure people are. Can't wait to get their hands on it and test it. You know I'm looking forward to playing with it. So I'm curious, what are some of the other projects you mentioned? The roadblocks you mentioned in Minecraft. Are there any other, you know, others in Excel that maybe you want to talk about that you're really excited about or that you're just like, can't wait to kind of get out there?
[00:06:46] Guest: Nico Christie: Yeah, yeah, it's a great good question. I'm not going to decide to just arbitrarily destroy the office suite. That's not my goal. I'm not going to also. And, uh, we definitely caught the attention of Microsoft.
[00:06:58] Host: Paul Barnhurst: What was their response? Can you share any of that? I know something you may not be able to, but they reach out to you. Or did you get a lot of questions from them?
[00:07:05] Guest: Nico Christie: I mean, this will be interesting to you and your audience, I'm sure. Yeah, we know them very, very, very well in my tweet. Not in my LinkedIn post, but my tweet also went viral. I tagged Satya and I used his famous quote, and I did it in jest because I have a ton of admiration for him. I genuinely think the strongest CEO that's a non founder as far as I know I ever lived. But I tagged him and I said, you know, there's a long way to go to disrupt Excel, but I'll settle for making Satya dance. And I tagged him and that was his quote of course for Google and Search. And yeah, he saw it. Funny enough. And I've met now with the head of the office suite, the head of Excel, multiple times, and they're a wonderful team. And we're thinking about how to actually more deeply integrate and partner with them to build out the AI ecosystem around Excel, because there's a lot of fun stuff we can do together. So that's the limit of what I can share. But it made a big splash of course internally. And. And we've come to know them.
[00:07:50] Host: Paul Barnhurst: Sure. Yeah. No, I figured you were limited in how much you could share there, but I was kind of curious.
[00:07:55] Guest: Nico Christie: But, yeah, we're very excited about what that could be. Of course, you know, and there was a second part of your question that.
[00:08:00] Host: Paul Barnhurst: Just any other projects that you could share about that you're working on right now or just something that you're really excited about?
[00:08:05] Guest: Nico Christie: Yeah, absolutely. Um, so it's kind of funny. We don't build products on a consistent product theme. Right? So it's not like the next thing is, you know, whatever, we actually make products on a consistent core theme technologically, which for us is building strong long term autonomous agents that can collaborate. Um, so things like ByteDance, they released 15 apps in their first year, and they're actually completely unrelated, like a jokes app, a car app, a video app. But they are related technologically because they believe in the recommendation algorithm. And like having the strongest would be like where all the flowers bloom. Um, so for us, our thing is building agents. So our other products are. Yeah, we have the top rated AI game in Roblox right now. Um, we have an events planning app. Uh, we have an app for readers that people can share their reads and think of like an agent first. Goodreads. Um, and there's one other one that's in the works right now. But yeah, there's a couple differences, all benefiting from the technology we're building.
[00:09:01] Host: Paul Barnhurst: Exciting. Well, look forward to seeing more of that. I want to step back for a minute before we jump more into shortcuts, because I think our audience will find this one interesting. So you used to dunk basketballs as a professional. So did you play college basketball or do you go straight into dunking first? Maybe a little background of how you got into dunking and talk a little bit about what that's like?
[00:09:23] Guest: Nico Christie: Yeah, I actually wrote a book called how I Learned to Jump Higher Than LeBron. Uh, when I was finishing college, uh, which is factually true. My highest vertical leap was 47in, I think it was like 46, almost. Uh, of course he's a different level of basketball player. Um, but in the book, I basically tell the story of my first day in sixth grade, and I was the shortest in the class. I wasn't on the height distribution chart. Um, and everyone was trying to touch the roof of the computer science class building. And no one could touch it, and I tried. I was like a joke. I thought, no way I'd touch it, and I touched it. Um, and it was the first time I realized that I was very naturally gifted at something. And it was a beautiful feeling. Right. I'm sure like everyone's got versions of that for them, for themselves. And I then realized what a powerful positive reinforcement loop that was. Um, because I started doing it. You do what you like, and as a result, you get better at it and then you like it more. Right. Um, I'm sure it's the same way for financial modeling, right? It becomes a part of your identity. So I became the local maximum. I became the best dunker in Miami. And that's where I grew up. And then that wasn't enough. I started traveling, and I started meeting other people who were like me. And we started performing contests and competing against each other. And what I learned is the ceiling for any given skill is so much higher than you think, right? Like, you probably have people here doing like nested lambdas on like corkscrews, like things that other people can't even imagine. It's the same way in every niche. And, um, I ended up partnering with like ESPN, Zion Williamson, McDonald's. Um, I would travel the world and compete in contests. And if you Google me, it's my prior life, but I can still dunk quite easily. I'm just not quite, you know, not quite the athlete I used to be.
[00:10:54] Host: Paul Barnhurst: Sure. You mean when you don't spend all day on it, you're not quite as good?
[00:10:58] Guest: Nico Christie: And then when you get a little bit older, when you start losing some hair and you work 14 hours a day. Yeah.
[00:11:04] Host: Paul Barnhurst: Just wait till you, you get close to 50. I, I definitely do not feel the, uh, athleticism I used to have. I still try to run every day, still enjoy it, but I'm nowhere as fast as I once was. I get it.
[00:11:16] Guest: Nico Christie: We are both as young as we'll ever be. Right?
[00:11:18] Host: Paul Barnhurst: Pretty much. Right. So, you know, you mentioned earlier, you know, when you released the tweet, the LinkedIn, both those went viral. You generated a ton of buzz. I know you got a ton of comments. I saw that, you know, some very positive, some scared, some wanting more information across the board. But what was that like for you? How crazy was that? I imagine it's the first time you've really had that strong of a response to a post you've done before.
[00:11:47] Guest: Nico Christie: Yeah, it's a fun question. So I've had one thing similar to that, um, before maybe 1 or 2 in my prior life as a dunker and then one in which I researched when we did something like a big multi-agent demonstration. But it's, it's a, it's a, it's an adrenaline dump. Right. I would refresh my phone every time I could possibly scroll. I would get the 20 max notifications right, like 20, 22. And um, you just couldn't quite possibly get through it all. And, you know, it was just really exciting. I had high expectations as someone in AI that knows Excel. I knew that what I have is really special, but I wasn't sure that other people would get enough information or that, like the spirits in the algorithm world would like, carry it as far as it did, but it by far exceeded my expectations. Um, and then in terms of the reaction, I would say I'm a little in a, I'm a little bit now, um, I kind of have my, like, my blinders on. And I'm in the AI world, like, I'm in a lab, right, in San Francisco. And I'm used to people who were afraid of AI but are no longer afraid of AI because, like, we've seen it and we've already adopted it in software engineering.
[00:12:49] Guest: Nico Christie: Um, but I underestimated the intensity of that emotional response in finance specifically, as in, like people, people are not really adopting AI except for ChatGPT, but for actually doing your job. Like, that's terrifying the first time, right? Um, we survived it two years ago when we started coding with it. Um, but I had anticipated that. And then I just surely miscalculated the numbers game. As in, like, there's another Excel modeler. Uh, Harry, I believe. Right. Um, who tagged me and he was, like, pretty upset, I think, for better or worse, I forgot we actually had a nice chat. We haven't. I chatted, but he was upset that, like, I haven't gone back to some people that like, I, you know, the common tech technique was potentially unfair. You know, I totally get it. And then he was like, I didn't even respond to Paul and I was and I had you know, what I did. I had to Google you. I was like, I actually candidly, I didn't know Paul. I've been out of the game for a little bit.
[00:13:39] Host: Paul Barnhurst: Lots of people don't mean they don't know me. You're okay.
[00:13:42] Guest: Nico Christie: Apparently, it's a cardinal sin. So I reached out, I actually saw I had that you'd reached out, I got back. We, you know, we set up time like, a few weeks ago. But yeah, it's amazing when you have 40,000 people commenting and DMing you that like as a sheer numbers principle, I can't go through them all. You know.
[00:13:58] Host: Paul Barnhurst: Yeah, I know somebody who I can't remember the exact numbers now, but they got like 11 million views in 48 hours on LinkedIn, and they had something like 10 or 20,000 requests to connect. It was in the thousands. And I'm just like, all right. He goes, I can never get through all this. He did learn eventually that after six months to go away because he's like, I want to look through some of them because they're gonna be a lot of valuable connections here. But how do you look through thousands of them?
[00:14:26] Guest: Nico Christie: I actually hired an intern for the week just to accept them all, and I want to at least memorialize the messages so that I can eventually make time. You know, and I chug out 500 a day, actually, is what I try to do.
[00:14:37] Host: Paul Barnhurst: Yeah. No, I, I totally understand. I get a ton of stuff. Not not that level of viral, but I get a lot of DMs and I try to respond to as many as I can. Sometimes it doesn't happen or I forget or they get lost in the collection. But I can relate a little bit.
[00:14:51] Guest: Nico Christie: You know, it's a luxury problem to have is the truth.
[00:14:54] Host: Paul Barnhurst: So as I like to say, if you know who a Weird Al is, he has a song called First World Problems.
[00:15:00] Guest: Nico Christie: I don't know the song, but he's got like 2020.
[00:15:02] Host: Paul Barnhurst: Yeah. It's pretty funny, like things like, you know, one pixel went out on my 50,000 inch TV. He just kind of mocks the things we worry about. Like really that's a big problem. And he just seems to call them all first world problems.
[00:15:17] Guest: Nico Christie: Yeah, yeah. Um, exactly that. It's, uh, champagne problems. Even, like, I would kill to have this problem, right?
[00:15:22] Host: Paul Barnhurst: Yeah, exactly. If that's the worst problem you have in your life, you don't have much room to complain, right?
[00:15:27] Guest: Nico Christie: Yeah, yeah. So it was a fun, fun ride. And honestly, it seems like there's a lot of ripple effects that I get to keep riding. Yeah.
[00:15:34] Host: Paul Barnhurst: Sure. Yeah, I'm sure that the questions come, like just us meeting and being on the podcast, right? I'm sure lots of different things are coming out of this. And so let's jump into the shortcut a little bit more. You'll kind of talk about let's start with maybe some of the work you did to create it, what that process was like. And when you knew that moment when you're like, I feel like we were really on to something, this is performing better than they ever expected.
[00:15:58] Guest: Nico Christie: So let me tell you about building one. So we had an agent that can use the computer. And as a part of that, there was a hybrid approach of using the mouse as a human does, plus using arbitrary code execution in the background. Now that is like ChatGPT the agent that just launched, that's what they do is what we did six months ago. But I mean, I was going to come to their own principles, an approach of how I should behave optimally. And it became painfully clear that, like, you don't want to watch an agent move its mouse for two seconds and then click and then reason and then like the mouse is an abstraction we've had as humans because that's how our systems work currently. But we can do amazing things in concurrency, the way you know, code can. So I decided, what if we got rid of the mouse approach entirely? Um, what would that look like? And then we changed it because that means like, well, if you rely on code, you have to have a really strong API. You have to have code access to the environment you're talking about. But Excel only allows you to get so much right. I don't own Excel. So, um, what do I have to do? I have to, like, kind of create my own app. But if you create your own app, I have so much love for Excel that I think it would be naive to compete in a different direction with it fully. And to even do something totally different. Like humans have been using Excel for its Excel 40th birthday this year. It's a part of the general system prompt, right? Like that's like spreadsheets are our fundamental truth. And Excel has landed on a perfect product that's been here, right?
[00:17:20] Host: Paul Barnhurst: As I like to say, when it comes to Excel, the three most the three most popular buttons in all of software are okay. Cancel export to Excel and I'm not sure it's in that order.
[00:17:33] Guest: Nico Christie: Yeah, no, of course, it's just I really think it's the most beautiful design software maybe in human history. Um, so if you do a spreadsheet and you try and you have to own it, you should have something that looks as close to it as possible and has mirrors all of the same functionality as you can. Um, and that becomes a long rabbit hole. Google tried to do this with Google Sheets and like it's still, you know, decades, right. Couldn't quite hit parody and that's that's enough that like enterprises don't adopt it. even with additional features. So I had to build a version of this, relying on all sorts of frameworks that would allow that, which was non-trivial. Right. But given I had that and there's still not perfect parity, right. Like a lot of macros aren't supported. Um, there are things that break anything over 50,000 rows or 30,000 rows. You get web performance issues because we're in the browser, and that can cause other issues. So big, huge data dumps. That's kind of out. Some of the keyboard shortcuts by default, because we're on the browser like Chrome, are at a higher level of preference there. So like you don't have the command T for example, for a table.
[00:18:25] Guest: Nico Christie: Right. Like or control. So there are limitations. But we've gotten a lot of the parody in. So if you have access to the machine interface you need parity. And now how good the machine is. That's where we were good as a research lab. So we can make really strong performing agents. And I can get into the nitty gritty of what that means. But the result is that we have an agent that is way better than it has any business being. Now, in terms of setting expectations for how good it is. Um, we actually, I guess by the time this is out, this is down. We have done a blind study, um, against actual first year analysts from Goldman Sachs, McKinsey, BCG, JP Morgan, a couple of the big shops. Um, and then we've had them go head to head against not had, had had them go yet. So we had them go head to head against shortcuts on the same tasks and gave the Excel analyst the real humans ten times more time on the same tasks. And then we had a blind study where we took managers from those same companies and just judged in preference. Which model do they prefer? And we have over 300 preference selections.
[00:19:24] Guest: Nico Christie: And as we haven't released the video and more data is coming in, but as of this moment it's at like 89% win rate against first year analysts. So that is how good it is currently. Now you probably look at our first analyst and like, you know, your 20,000 hours of experience behind me. Like, you know, there's so many levels to this thing. And I'm sure most of the financial modeling community is way stronger than shortcut right now. But there's a delta that we're gaining ground on for sure. And then specifically where that is good, anything where it's creating things from scratch. It's going to be like orders of magnitude faster than even. But that's not that common, right? Like you have your templates that you love and you want to update them in light of new data or new assumptions or any questions. Um, in that capacity, I think we're at like this moment where self-driving kind of is where it's like there are a lot of magic moments, but we're not sure if we're ready to adopt it. Right. Um, and that's maybe the most important challenge we're tackling.
[00:20:17] Host: Paul Barnhurst: So, yeah. So reviewing people's work and making changes in the area. That's harder at the moment. Much easier if you're building from scratch.
[00:20:25] Guest: Nico Christie: Yeah, well, it's more specific than like, even building your own changes, but on the existing templates. So say you say you open up an existing like, super complicated, uh, you know, Monte Carlo simulation, right? Uh, and then from there, you want to be like, all right, change the simulation, make it 1000 rows instead of 100, and, um, pull new data from these other sites and, um, make it for a different company. And now, if you're really constraining it to the current template that it has to edit, there are some flakiness that happens. It's like when you're self-driving on your Tesla and like the left turns too hard or something weird, you know?
[00:20:59] Host: Paul Barnhurst: While my background is in FP&A. I am also passionate about financial modeling. Like many financial Modelers, I was self-taught. Then I discovered the Financial Modeling Institute, the organization that offers the advanced financial modeling program. I am a proud holder of the AFM. Preparing for the AFM exam made me a better modeler. If you want to improve your modeling skills, I recommend the AFM program. Podcast listeners. Save 15% on the AFM program. Just use Code Podcast.
[00:21:39] Host: Paul Barnhurst:And I think something you've mentioned before last week is how does it handle? Because I know almost every model that reviews files is there pulling apart that we have right now, you know, to be able to feed it into an Ala and whatever markup language you're using to, you know, kind of analyze the model struggles from what I've seen with color, with font, with the formatting nuances. Are you guys struggling with that as well?
[00:22:04] Guest: Nico Christie: Yes and no. It's a great question that shows you definitely are in the know for AI and models. There's a couple reasons for why that's the truth. And it's because if you're using an Excel API or you're building onto Excel, the information you can get is a little constrained and they're actually a little bit it's a little difficult to get formatting information as in colors specifically. And even like is it using general.
[00:22:26] Host: Paul Barnhurst: Or even Microsoft Shapes? Lm has struggled with that for sure.
[00:22:30] Guest: Nico Christie: Um, but knowing how important it is and knowing that we own our own API, um, it's like one of the main priorities for us. Um, so currently we can get all really all reasonable visual based formatting and information and then extrapolate from there. I will still say formatting is probably where it's where it might be the weakest, as in like you'll get a model, but then like say it's an income statement. Like you'll get the um, everything will be at the same font, the same standard thing. It'll be Ariel Tan, whatever. Um, the right format isn't like accounting formats, accounting or currency. But then that, like, bottom line item, instead of having the double border, it'll do like no border. And then it might make the numbers smaller and like or bold and smaller. And there's a couple that flakiness at the end that um, it's more a matter of just prioritizing than like any real fundamental limit for us to solve. But yeah, it is. It is funny enough, one of the hard problems for me.
[00:23:17] Host: Paul Barnhurst: Yeah, I find it interesting, you know, when you think of all the things you would struggle with, that's not what people typically think of.
[00:23:25] Guest: Nico Christie: No no no no. Well, the other reason is because most of the approaches to solving that rely on vision based models. But vision based models are not great at pixel coordinate truth. So what happens is, you take a screenshot of a model and you're like, oh, well, why is this cell green? And then it's smart enough to know there is a green cell, but then it will actually infer that that cell is like C4 when it's really D6. And then that actually confuses the model because it sends that information. And now it's conflicting ground truth data about your model. Um, so it's like there are very funny edge cases about this.
[00:23:55] Host: Paul Barnhurst: Well, interesting. Well, I know, uh, someone asked Microsoft, one of the Microsoft MVP's, when they came out with the dynamic arrays. Right. You're spilling formulas like. All right, I put in a new input and now goes from 50 to 500 rows or to 30 rows or whatever it might be. And none of the formatting follows a spilled formula. So if you had 20 rows and you'd formatted 20, and then it goes to 500 that you have to go in and change the format from 21 to 500. It doesn't spill. And someone asked one of the Microsoft engineers about that going, you guys need to develop it this way. He's like, how difficult that is to develop. Like we haven't found it. You think we would have it out there if we had an easy solution?
[00:24:35] Guest: Nico Christie: Yeah. Um, you know, there's a long tail of formatting that's hard. I think, um, like, the very shallow fix there would be, like, just include the, um, the function for a, like, copy and apply formatting as, like the tail end of that same step. Sure. The reason I'm sure they don't do that is because, like, every row might actually have a separate format.
[00:24:52] Host: Paul Barnhurst: Well, often you have totals you want. Exactly. You might. Yeah. Right.
[00:24:55] Guest: Nico Christie: So like you can't really arbitrarily flood fill formats unfortunately.
[00:24:59] Host: Paul Barnhurst: Um, yeah. And I mean, and I'm sure they could, but they'd create almost as many problems as just not doing it at all.
[00:25:05] Guest: Nico Christie: Yep. Yep. And it's easier to not get something wrong. Right.
[00:25:08] Host: Paul Barnhurst: Yeah, exactly. So. All right, well, I think we spend enough time on formatting. Our audience is like, isn't this a modeling show? Okay, so I'm curious about writing formulas because most of the tools I've seen a lot of times are either hard coding or the formulas they've written are generally not logical. Like if you're auditing them, sometimes you know, dollar signs or you may not need them like they work, but they're often in an odd syntax to how a human reads and a human thinks about it. Are you seeing some of that as a build? Models like think of the financial modeling World Cup example. Did it just hardcode or to build models with that? We'd love to get into a little bit of the actual formula writing, because that's something we're all wondering about, is you audit and review or change these models.
[00:25:52] Guest: Nico Christie: Yeah, I'm, I mean, ideally I would share my screen even to show you.
[00:25:54] Host: Paul Barnhurst: Yeah. Let's go ahead and do that. So this is my plug. If you're watching this audio, switch over to YouTube and watch it.
[00:26:02] Guest: Nico Christie: Yeah I mean you get to see the coolest, the cool model stuff, right? Um, so here.
[00:26:05] Host: Paul Barnhurst: You don't want to see the, uh, a world champion dunker after all.
[00:26:09] Guest: Nico Christie: Oh, well, I'm not so great to look at. Um. Here we go. Uh, let me know if you're seeing this, okay? Oh, this is a shortcut, right? Look, obviously, it looks extremely familiar. What you can do is open up an Excel file. So, like, I'm going to look at some files I just created. Um, let's see what this is. Boom. Okay, so this was a ten. This was a 10-K that I asked it to basically build a DCF model over with a bunch of charts. Um, I'll zoom in just a bit. Now, the reason I'm bringing it up is because a shortcut made this. Right. So all of this says you should hold the investment. It has projections DCF valuation scenarios. Monte Carlo simulation cap table summary. This was a live demo right before this call. But your question is like what about formulas. Right. Um, so it turns out, um, so whatever it can be, a formula is a formula. Um, let's see if I show you the DCF. So everything that can come from a formula is from a formula, and then everything that has to be hardcoded because it's definitely a driver or an assumption comes from the driver or assumption, but it turns out in the limit. There's no reason that models like these can't use formulas, right? Like, look, this is a real formula, right? B14 times one. Like, you know, like these are formulas, even nested formulas. Um, oftentimes like circular references to the extent that they can still work. Um, it's really something that is actually quite solvable because it's so in distribution, we as humans have been using models. Um, have we been building Excel models for so long and formulas are so important.
[00:27:42] Host: Paul Barnhurst: Okay. So we definitely show I'm seeing all the formulas there and it looks like it use a lot of funk. So not so much formulas but you see different functions like let's say it's doing an NPV. Does it do the NPV function or just kind of the math behind it. It's kind of curious.
[00:27:57] Guest: Nico Christie: Yeah good question. No, it can use any formula. In fact it was I was trying to get it to try using lambdas and some of the more complicated ones. But yeah, there's no I haven't seen any formula that it can't quite do in principle.
[00:28:09] Host: Paul Barnhurst: Okay. Yeah. Because I've definitely seen some I've seen some other tools, you know, and I've seen them write some formulas. But so it's just kind of curious how you're, how you're seeing it. No. Right.
[00:28:18] Guest: Nico Christie: No. But it actually becomes a matter of like one. If it's capable of doing it, then how do you really just strongly emphasize that it chooses to do that? Um, because of the same reason that like, like first of all, this could say is C19 divided by seven minus one, or I can just have my Python interpreter or I can have, um, my code gen environment, like, do that same math and just print 1.0.1378.
[00:28:40] Host: Paul Barnhurst: Yep. And I've seen that. I mean you see that from ChatGPT, right?
[00:28:44] Guest: Nico Christie: Right.
[00:28:44] Host: Paul Barnhurst: And that's like you're very specific to tell it to use Excel. And it's not as good when you do that.
[00:28:49] Guest: Nico Christie: That's funny. Yeah. So you've clearly tried and we can talk about that too. Um, yeah. It really struggles to use formulas at that point. But the same reason that it can use a formula means it can just use any arbitrary formula. So it could just use NPV for NPV like that. That doesn't quite seem to be a real limit for us.
[00:29:04] Host: Paul Barnhurst: I know that that makes a lot of sense and this is helpful, and I appreciate you kind of sharing a little bit there. That all makes sense to me because I know that's an area that I think in general it feels like, and we're moving so fast. But historically a lot of tools have struggled with actually showing the function, showing the formulas, or showing them in a way that makes sense to a human. Yeah, maybe computer written and you're looking at it going, what does it even mean? I would have never.
[00:29:31] Guest: Nico Christie: Yeah, well, that's my problem with some of the top human modelers too, is they do these pretty convoluted formulas. So yeah, there's that's a human problem also.
[00:29:39] Host: Paul Barnhurst: Well, 100% agree. I've attended the Financial Modeling World Cup the last two years. And watching the lambdas and the formulas, they write, I go, I'll never be able to do that. Like people consider me good in Excel. Yeah. And there's good. And then there's the top in the world. And I'm like, all right, I watch the top of the world. And I'm just like, nope, I just don't have the time to figure that out.
[00:29:58] Guest: Nico Christie: Yeah. But also I don't believe that, like, we should get to a point where, like, everyone feels like they have to, right?
[00:30:04] Host: Paul Barnhurst: Like, oh, I agree. I'm not disagreeing with you at all. I mean, and that's one of the beautiful things about lambdas is you can download anyone else's, download them to get hubs. The reality is, within a company, if you want to have an environment where somebody is writing a ton of lambda isn't really valuable. Just write good instructions with them and let others use them. We don't know how everything works anyway when we're writing other functions.
[00:30:27] Guest: Nico Christie: Yeah, I mean, this is the kind of thing that Excel starts to look a little bit more like programming, like these conventions exist, right? Like we have a Readme in our GitHub. Right? Like for the same reason I had to set up your local environment, right?
[00:30:38] Host: Paul Barnhurst: Bad financial models can lead to bad decisions or worse. So, how do you minimize the risk of a bad model? You make sure the models you build are great. The Financial Modeling Institute developed the Advanced Financial Modeler accreditation program to help modelers like you. The AFM program offers a step-by-step approach to building world-class financial models. The program ensures that you know the best practices in model design and structure, and will help you brush up on your Excel and accounting skills to be the one on your team to build great models. If you want to impress your boss and your clients, get AFM accredited. Podcast listeners. Save 15% on the AFM program. Just use Code Podcast at http://www.fminstitute.com/podcast.
Yeah, exactly. Because, you know, as Excel continues adding Python Lambda becomes much more of a coding environment in addition to what it's traditionally been. You have to have ways to manage that. And how do you help everybody take advantage of it, knowing that 99% of users are never going to want to write a Lambda. 90% are probably never going to want to write Python, maybe even more. And how do you meet both needs? And that's what I think makes SEL so impressive, is they've managed to do that for 40 years and continue to develop with some areas that are very old and some challenges, like VBA is an old language.
[00:32:19] Host: Paul Barnhurst: It's great. A lot of people like it for certain things, but it's not what Microsoft would use today. All right. So you know, we've talked about how shortcuts work. We've shown a little bit of the tool. What I want to get into a little bit more is what this means for the modeling community. What do you think about it? So, you know, how do you see these AI tools, whether it's shortcuts or other kinds of changing professions? Do you see it as more of a point agents are going to place a lot of people, or it's more people are going to be using all agents all day to build their work and have these all add-ins, or are you more talk to and work with it. And you're not writing as many formulas. How do you kind of see it?
[00:33:00] Guest: Nico Christie: Yeah, let's break it into two things. For one let's address a principle, and then I'll break it into two things. In principle, I believe what's going to happen is, the input cost for modeling goes down. As in like it's easier and faster to build models. Now, in principle, I think whenever that happens, the ROI to build a model goes up, or whenever your input costs go down and the outputs stay the same like the ROI goes up. Meaning if there's a higher ROI towards doing something, more of it will be done. That's kind of the market hypothesis for everything. So I expect there to be more value for financial modelers and for finance in general as a result of it becoming easier to model. But now let's go even deeper. There are different kinds of people who will benefit from this differently. There are people who so most of the tech world are kind of excel and like, neophytes, right? It's really powerful to watch people who were bad at Excel become really good at it pretty fast. You know, good's a long tail, right? Compared to you, they're not good. But, like, they get a lot better fast. Uh, thanks to it. Now, that's happening in software engineering right now where people can build apps now. And, you know, they weren't classically trained engineers. And it's a beautiful thing to watch. Hundreds of millions of people become good at something and like, contribute to the world in their new specialty or their new interest.
[00:34:14] Guest: Nico Christie: So I think that'll happen for Excel. I think hundreds of millions of people will get pretty good at it. Then there's the second thing, and this becomes up to the community. Will the community decide to adopt frontier tools? And then what will the upper bound look like for people who happen to figure out how to really wrangle it to their advantage? So there were their senior engineers, great engineers who when, uh, you know, GPT four or sonnet 3.5 came out, they were in principle opposed to using AI to code. And what's happened is a lot of those people have fallen a little behind. Um, there are like, junior engineers or like, really just really savvy context engineers who can out just outdo, do like, you know, really strong form of programmers because they refuse to adopt technology. Now, there are also classically trained IOI medalists, incredible engineers who now also use AI. And the upper bound is unimaginable. Like they are so good and like they use it in ways I can't even quite describe or write like. Um, but what I expect is something similar in the Excel community. As in some financial modelers, they'll know the ins and outs, but like, no, it's actually garbage at this other thing. But for this one use case, it is ten times faster than me and I run ten of them in parallel. Right. Like, um, there will be some people in the financial modeling community who can find ways to, like, become hyper, um, better at it because.
[00:35:30] Host: Paul Barnhurst: Right, almost like exploit it, for lack of a better word. Right, right.
[00:35:34] Guest: Nico Christie: Right. Right, right. But that's how it should be. Right. Um. It's just a tool. Like, do you want to get, like, are you going to be in principle, opposed to a tool? Probably. That's probably unwise. But maybe, maybe there are places that can help you, and you owe it to yourself to find out, because this technology is not just changing things for finance, software engineering. It's like the biggest thing since the internet. So yeah, my recommendation is like definitely, definitely learn to use AI and it will and consistently, consistently reassess your opinions on it because the ground truth trains every day. You know, I know people that were like, right, rightfully said that I wasn't ready for coding in 2023 and they just weren't ready in 2024 for when it was. Right. So that's kind of the attitude. But I think, I think in general the fear is misplaced. It's just it's just obviously it's a natural thing to be afraid of, you know.
[00:36:22] Host: Paul Barnhurst: Yeah. No, it's an adjustment for everybody and it's new technology for a lot of people. Scary. So it's just taking time to adjust. And this is move so fast. I mean you think what it was doing two years ago, you know, what you're seeing publicly three years ago, you know, 18 months, sometimes it feels like it's measured in days, not weeks or months like we're used to. It's like, oh, I tried that yesterday and it was no good. I'll try it today. Oh, it actually answered the questions today.
[00:36:50] Guest: Nico Christie: It's also not just getting better faster than ever than everyone's used to. It's getting better and faster at doing things. And that's just a different feeling, right? Like we haven't seen machines do things before, you know?
[00:37:01] Host: Paul Barnhurst: Yeah. It's weird when you think, okay, I can do my task for me. And that's an area I need to get better at using. It is I, you know, I use AI a fair amount, but I'm realizing a lot of things I could do better with it. It's crazy.
[00:37:13] Guest: Nico Christie: Yeah, for sure, for sure. So I totally understand and I think, I think in six months the sentiment will become much more positive towards it as like people realize it's not quite as exponential.
[00:37:24] Host: Paul Barnhurst: Yeah I agree. All right. So I have one more question. Then we're going to move into our rapid fire section. So this is when we ask everybody your favorite Excel shortcut at AMC. I know the H. Obviously that's the home. The M is trying to figure out what that is on the menu. I'm almost ready to pull up Excel. Tell me what I'll see. I don't know.
[00:37:46] Guest: Nico Christie: That it's the Merchant Center.
[00:37:49] Host: Paul Barnhurst: Oh, that's why nobody does that.
[00:37:51] Guest: Nico Christie: I know it's so bizarre. Um, I'm always like, uh. I just always find myself, like, combining certain cells, centering them, and, like, trying to make, like, nice, nice, pretty tables to display information.
[00:38:00] Host: Paul Barnhurst: Why not center across selection?
[00:38:02] Guest: Nico Christie: I'm a bit of a madman.
[00:38:04] Host: Paul Barnhurst: You and Jordan Gold Meyer, he's an Excel MVP. That he. He told me his favorite, uh, Excel feature was the merged center. So you and him.
[00:38:12] Guest: Nico Christie: Did very good.
[00:38:13] Host: Paul Barnhurst: And everybody blasted him on LinkedIn and said he was. You know.
[00:38:16] Guest: Nico Christie: I had warned you that my my answer to this would probably be very unpopular.
[00:38:19] Host: Paul Barnhurst: Uh, yeah, we'll probably make a short video out of it just so we can get a lot of hate mail. Good, good. You might get some, too. All right. So here's how rapid fire works. It's a quick yes or no to each question. We have about 15 of them. So we'll take, you know, just kind of a minute run through them all. Then at the end you can elaborate on 1 or 2, maybe three that you're most passionate about. Because we recognize when it comes to modeling, there's nuance to everything. There's rarely a right answer. So you ready? Circular references in models. Yes or no? No VBA. Yes or.
[00:38:53] Guest: Nico Christie: No? No.
[00:38:54] Host: Paul Barnhurst: Horizontal or vertical models. Horizontal Excel. Dynamic arrays in models. Yes. What about external workbook links?
[00:39:05] Guest: Nico Christie: No no no no no no.
[00:39:06] Host: Paul Barnhurst: Yeah I've got a few hell nos on that one. Named ranges. Yes or no?
[00:39:11] Guest: Nico Christie: No.
[00:39:13] Host: Paul Barnhurst: Do you follow formal standards like Fast or Smart? One of those formal standards boards when you model?
[00:39:18] Guest: Nico Christie: No.
[00:39:20] Host: Paul Barnhurst: All right. Do you think financial modelers should learn Python in Excel?
[00:39:24] Guest: Nico Christie: No.
[00:39:25] Guest: Nico Christie: No. And caveats.
[00:39:27] Host: Paul Barnhurst: All right. We'll get to the caveat here in a minute. What about Power Query?
[00:39:31] Guest: Nico Christie: Uh, yes.
[00:39:33] Host: Paul Barnhurst: Power BI? No. Okay. Will Excel ever die?
[00:39:39] Guest: Nico Christie: Yes.
[00:39:41] Host: Paul Barnhurst: All righty. Alrighty, this is always the million dollar question. The Excel one. Next one. Will I build the models for us in the future? And I have a feeling I know where you're going to answer on this one. Yes. Yeah, exactly. If you said no, I would have been a little worried. Do you believe financial models are the number one corporate decision making tool? Yes. Okay. And then for you personally, what's your lookup function of choice? I choose Vlookup index match X, Vlookup.
[00:40:06] Guest: Nico Christie: Something else I use vastly in superior lookup.
[00:40:11] Host: Paul Barnhurst: Uh, well we gotta end the interview now. Good. So I think the first one you mentioned has a caveat and then you can pick another one, Python in Excel. You had a copy out there.
[00:40:20] Guest: Nico Christie: Yeah. Um, so the caveat is that I believe all modelers should learn Python. Um, I just don't believe that Excel is the right abstraction layer. I think they're serving it to you probably in a serverless environment where you can do pandas manipulation on a data frame, but like that's you should just learn Python and then use whatever programing language you think is the best for your modeling.
[00:40:38] Host: Paul Barnhurst: That makes sense. And you're not the first one who has said that. So I've had others give a similar answer of like, if you're going to learn Python, learn Python, there's no reason you need to learn it in Excel per se.
[00:40:47] Guest: Nico Christie: Yeah. Yeah, exactly.
[00:40:48] Host: Paul Barnhurst: All right. Well, are there any others there you wanted to elaborate on? Maybe we'll excel every day.
[00:40:53] Guest: Nico Christie: Yeah. Let me tell you about this. So this is a discussion we've had with Microsoft, right. I just mean I'm uncomfortable with the future predicting out more than like, five years. Uh, I see how fast the world is changing. And I think, you know, even ten years with Excel die, that would be another quarter of Excel's life. Like that's all. That's a meaningful time. And for any software that's like an entire lifetime. So, um, the answer is yes. The question is when is it? Ten years, which would be awesome. 20 years would be fantastic. It could be 50 years. I just don't think so. My guess is probably closer to ten. And Excel is just a fundamental truth. It's the best way to observe, you know, matrix spreadsheet data. So there will always be some abstraction that is like that. But as you know, as technology gets better and faster, as we've seen, um, it's really hard to imagine it would look very similar to it. It looks to be.
[00:41:37] Host: Paul Barnhurst: Yeah. And I think for sure it's going to look different, whether it's Microsoft or replaces Excel with some other version, whether it's somebody totally different. Remains to be seen. But I'm always a believer that nothing lasts forever. I think it's at least ten years out, but nothing would surprise me.
[00:41:55] Guest: Nico Christie: Yeah, I'm in the same boat for sure.
[00:41:57] Host: Paul Barnhurst: So last question. If our audience wants to learn more about you or, you know, get in touch with your tests, shortcut any of those types of things. I know you mentioned here, soon you'll be opening up. Opening it up. What's the best way for them to do that?
[00:42:09] Guest: Nico Christie: Yeah. Um, I'm on LinkedIn and on Twitter if you want to even email me at Nico at Fundamental Research Labs. Um, I'll try to get back to everybody and. Yeah, reach out to me. I think Twitter's probably the most natural way to use LinkedIn. I had to go through a friend request, so.
[00:42:22] Host: Paul Barnhurst: Sure. And then, uh, as far as the platform, what's the website for? Shortcut.
[00:42:28] Guest: Nico Christie: It's try shortcut I. By the time this goes live, the product will be launched.
[00:42:34] Host: Paul Barnhurst: All right. Perfect. Well, thank you so much, Nico. I appreciate you carving it out an hour here to chat with me. I really enjoyed our conversation. So thanks for coming on.
[00:42:42] Guest: Nico Christie: Absolutely. My pleasure. Paul. Thank you for having me.
[00:42:44] Host: Paul Barnhurst: Financial Modeler's Corner was brought to you by the Financial Modeling Institute. This year, I completed the Advanced Financial Modeler certification and it made me a better financial modeler. What are you waiting for? Visit FMI at www.FMInstitute.com/podcast and use Code Podcast to save 15% when you enroll in one of the accreditations today.