Building AI for the Top 1% of Excel Users with Peter Fuller

The Mod Squad. We are the Mod Squad. The Mod Squad featuring Ian Schnorr, Executive Director of Financial Modeling Institute, Giles Mail, Humble MVP and co-founder of Full Stack Modeler, and Paul Barnhurst, the FP&A guy.

Start by telling a little bit about TraceLight, how it came about, a little bit of the backstory, what you're doing today.

I think it'd be fun to kind of build AI for Excel. At that point, no one was doing it. The key thing for us was that middleware between the spreadsheet and the LLM that made the problem more comprehensible. So since launching that Excel agent, we've really expanded the product. We're now really all-in-one AI workspace.

Meet Peter Fuller, CEO of TraceLight.

CEO and co-founder of TraceLight, here to unpack how his team built an AI workspace that's beating Microsoft at its own game in finance.

We tested TraceLight first, and then suddenly Microsoft came out with their first agent in web. Did you expect the LLM providers like Anthropic and OpenAI and Microsoft to do what they did when they did it, or was it a bit of a surprise?

I didn't expect. Anthropic and OpenAI to launch an add-in to Microsoft Excel. The core argument for us, the needs of the top 1% of users are different from the needs of the other billion people. So that's very much us as a company. We have no ambition to go fight Microsoft and Anthropic for the billion users. We'll fight them on the 10 million.

What is your plan or strategy to ensure survival and growth? Most people have discovered it doesn't really make sense to go head-to-head against Microsoft.

Call me an idiot, but I actually am fine going head-to-head against Microsoft.

Welcome to another episode of the Mod Squad. I'm really excited to be joined by my co-hosts, In and Giles. In, how are you doing?

I am doing well, Paul. Trying to keep up with the madness like everybody else is, I think. And we're getting ready.

Did we lose everybody else? Is that him or not?

Whoa, was that me?

Yeah, that was you. Yeah, I think you just run with it. That's a great start.

Love it. We're getting excited for the modeling conference coming up in 3 weeks on September 24th, just over. We've got an amazing lineup of speakers and going to be thousands and thousands of people at the world's first ever 24-hour online virtual modeling conference. So yeah, lots going on. We're busy, uh, having fun and excited to meet with Peter today.

Yeah, really excited for that conference. What are we, 3 weeks out, about 3 and a half?

Yeah, no, 3 weeks, September 24th. 24 hours.

Giles, what have you been up to?

Just working. I don't have a life outside of work at the moment. It's one of those things where I'm happy to be busy, but even like this week on its own, I'm doing Copilot training tomorrow with a team of about 100, doing some collaborative Claude training on Friday with some other quite well-known trainer with a large beard. That's interesting, isn't it, Paul? And then we've got the UK Excel Esports Finals. In London on Saturday. So yeah, like crazily busy and also trying to keep up with everything that's actually happening in AI world as well.

And I can confirm his comment on no life because the time he decided to meet with me last week, cause I moved our meeting, was 10:30 his time on a Friday night to prep for some training. So that tells you he's working around the clock. I can vouch for that as the big bearded guy he talked about that he was going to do a training with.

I should be rich.

All right. Well, we're thrilled this week to have a guest with us. We have with us Pete Fuller. He's the CEO of TraceLight. Some of you may have heard of it. It was the first— I believe it was the first tool we tested on the ModSquad series back in September, about a year ago. That platform has expanded and grown quite a bit since then. So we wanted to bring him on and just talk about the journey, a little bit of his background and where he sees modeling going. But why don't we start with just giving you a minute to just tell us a little bit about yourselves? Tell our audience who you are and how you ended up where you're at.

Well, yeah, thank you very much for having me. I'm very excited to join you guys. So yeah, I'm Pete. I'm the CEO and I'm one of the 3 co-founders of TraceLight. So we founded the business in mid-2024 when the LLMs could barely write valid formula in a spreadsheet. And before founding TraceLight, I was a venture capital investor, very, very early stage VC investor here in London at a place called Entrepreneurs First. And before that, I was at McKinsey 5 years, obviously did a lot of modeling there and was also part of Quantum Black, which is McKinsey's AI practice, basically. Before McKinsey, I mean, a bunch of weird and wonderful things. I ran a charity in East Africa. I pounded spreadsheets. Amazon briefly as well. One or two other things. So yeah, that's me.

Love it.

Fun journey. Well, why don't we start? Can you tell our audience, start by telling a little bit about TraceLight? Kind of how it came about, a little bit of the backstory, what you're doing today, and we'll just go from there.

Yeah, for sure. So I started working with Alex and Yannick in mid-2024. So our backgrounds, I mean, I've just told you mine. Alex, our CTO, was a software engineer at Jane Street before TraceLight. And before that, he was at Cambridge, which is where he met Yannick. So Yannick's a machine learning guy. He did that at Cambridge and then did that in a bank for a while. So we got together and we were exploring different ideas, actually. I think it'd be fun to kind of build AI for Excel. At that point, no one was doing it, right? There was this funny little research paper from Microsoft. It wasn't anyone significant at Microsoft, but the research paper basically said like, yeah, LLMs have a really hard time with 2-dimensional data structures in general, obviously including spreadsheets. But if you were going to kind of represent a spreadsheet to an LLM, this is kind of how you might think about it, how you'd do it. That's where we started. Alex implemented that over the course of a few days and obviously Obviously we built on it very significantly. And yeah, the sort of the key thing for us at the beginning, which is now completely irrelevant, by the way, we've actually thrown it out of the codebase entirely, right? This is the journey for many AI startups, but the key thing for us was that middleware, I guess, between the spreadsheet and the LLM that made the problem more comprehensible to kind of GPT-4 at the time. And, um, we really built from there. I mean, our motivation to build for financial models in the first place, I'm sure you guys would agree with this, right? It's like, you know, I feel like people build financial models Whenever they have a really big and important decision to make, right? And we're very ambitious, right? We want to make our mark on the world and we want to build a company that helps the biggest and most important decisions get made. And so since kind of launching that Excel agent, we've really expanded the product. You know, we're now really kind of all-in-one AI workspace for our customers. So they can do all of their work in TraceLight. It's like a central web app and we have add-ins to all of the Microsoft Office suite, but Obviously, we still have that heritage as an Excel company, right? So, you know, we're really, really outperforming in kind of spreadsheets and spreadsheet-related tasks. So yeah, that's been the journey.

All right, I'll let Giles or Ian ask some questions so I don't hog all the mic.

I got just an initial thought when you were describing— is it Yannick you're saying, the other founder? Machine learning at Cambridge. I can imagine somebody like that must look around now at all the AI experts popping up left, right, and center and just be like, Good God, you know, machine learning, Cambridge. I said this at some point, like a year ago, I was like, there is a world of actual AI experts. He must be so annoyed with this.

I don't think Yannick feels that way about you guys, but he definitely did about me, I think, at the beginning, for sure. I think I've learned something.

You won him over.

No, he's an incredible guy, right? So he has this obviously deeply technical backgrounds. But, you know, he also loves like front end and like design. He's this like unicorn of an engineer who can do the kind of ins and outs of machine learning, but also loves an interface, you know. And also, you know, a lot of building with LLMs is not very sexy, right? Like it's, you know, so it is a lot of prompt engineering, right? It certainly was. And it's a lot of evals and it's a lot of hill climbing and you need to of tenacity and a certain mindset to be really, really good at that. And Yannick has that, right? Um, so, uh, it's, um, yeah, I think he's very, very well suited to what we do.

I've got another question, but I don't want to take over. Go for it, Ian.

No, go ahead, Giles. You go ahead. You can get 5 in a row.

The, um, I'll take it.

I'm gonna limit you.

I don't know if this is jumping ahead to like the meat of what we're going to get into, but obviously I guess what was interesting for us when we started last year with the Mod Squad, we We tested Tracelight first and a bunch of other tools. And then suddenly, you know, Microsoft came out with their first agent in web and Anthropic did and all of this. I'd be really intrigued to hear what that journey's been like for you. And I guess, like, for me, did you expect those, you know, the LLM providers like Anthropic and OpenAI and Microsoft to do what they did when they did it, or was it a bit of a surprise?

Yeah, no, no, that's, that's, I have to be intellectually honest about this because I actually went back to a memo that I wrote to raise our seed round way back in the day. And rereading that memo, it's clear that I didn't expect Anthropic and OpenAI to launch an add-in to Microsoft Excel, right? Of course, I talked about collapse, but I just didn't think that was a likely Product direction. I was obviously completely wrong about that. Like clearly, I expected Microsoft too, right? But you know, like there's a few things going on there. Like one, when big labs launch products like that, yeah, of course it's easy to be scared. It's also very, very validating actually for what you do, and you just. get orders of magnitude more attention in your space, right? So Anthropic, in particular, launching Claude and XL was a huge moment for all of us, right? And it was actually, it was fantastic for sales. And, you know, it's not like everyone was like, oh, okay, we'll just dump TraceLight and we'll go and use Anthropic. It was like quite the opposite. Everyone got interested in XL, right? So that was really good, actually. And like, the core argument for us, which remains the case to this day in slightly expanded form, is that the needs of the top 1% of users, right, are different from the needs of the other billion people using XR. And that's still true. So that's very much us as a company. We have no ambition to go fight Microsoft and Anthropic for the billion users. We'll fight them on the 10 million.

That's awesome. And listen, we love We love supporting the startup and the little guy, as it will, against the $5 trillion market cap. And we love Microsoft too, right? I mean, at FMI, we've got a wonderful partnership with Microsoft, but we love the innovation that comes out of nimble, creative, small, hungry organizations. I mean, we all work at and run small, nimble, hungry organizations. So, we're So, we love that. But we talked a year ago. I don't remember if you remember, guys, that it felt like at the time last year that we were in the early stages of the auto industry. We remembered that in the early 1900s, there were— well, we didn't remember this. Paul might have remembered this, but Giles and I just read the research on it that back in the early 1900s, there were about 3,000 carmakers in the US. And that quickly got condensed down either through consolidation or just companies just evaporating. And not making it. And we've already seen both extremes in the agentic AI space with spreadsheets. I mean, it's a couple of the companies that we interviewed and that we tested in early days are gone. They just threw in the towel. And others seem to have taken a more niche approach and said, you know what, we are going to focus on this niche and this target and this particular subset that we don't think We don't think the big folks will either focus on or they can't do as good a job on. And I'm just wondering, I mean, as Giles, just to build a follow-on to Giles's question, what is your plan or strategy to ensure survival and growth? Because as you said, I mean, most people have discovered it doesn't really make sense to go head-to-head against Microsoft. So what's the plan? I'd love to know how you're planning to develop something, you know, or what you're doing to develop some niche.

Yeah.

Well, call me an idiot, but I actually am fine going head-to-head against Microsoft. Like, in my user base, I have vast respect for Microsoft. Do I think they're a good product organization? No, I do not. Do I think Copilot is a good product? No, I do not. I think it's actually— and I'm sorry if you guys have allegiances there, but I think it's—

oh no, let it, let it out. I'll support everything you're about to say.

I remain shocked by how bad it is, right?

Um, I've—

anyway, I won't go on. So that, like, that's thing number one, right? Like, uh, I think, um, it's easy just to like roll over and say, oh, you know, these big companies looking at it, or it must all be gone now. I think there would be There are countless entrepreneurs who, if they'd taken that approach, then massive companies would never have been built. So that's the first thing. The second thing, it's interesting when you said that this is like the beginning of the auto industry, I was nodding along because I thought you were gonna say something else. I thought you were gonna say that the products are all very rudimentary. And I still feel that way. I feel that there's enormous product overhang in the space now, right? And by that, I mean, the capabilities of the models are fantastic, but the products that we built around those models, the harnesses that we built around those models still, to me, feel inadequate, right? When I look around the market. So I think there's just an enormous amount of work to be done there. And then the final thing is like, You know, TraceLight, we start at the application layer, right? We aim to build the very best product for our users to do all of their work with AI, including their Excel models. But the way to emote in our space and in many spaces is by making your products get better with every single interaction a user has with it, right? And so that means that you are not just building at the application layer, right? You are building memory, right? You are building, like, you're taking all of the data that's generated there and you're helping your customers to, or their instance of TraceLight to learn from that data and get better every single time they use it. And there's actually enormous entrenchment that you can build there if you get that loop right, right? So this is what we're working on currently with customers. And it's really, really exciting, right? And the needs of our customers in terms of not just the application level, so the kind of information architecture and everything that sits behind, the needs are very, very specific, right? And the labs and Microsoft definitely are not building for them right now. So it's a very exciting time.

Oh, I love, listen, I love that. I was not expecting that response, but listen, I applaud the fact that you have a plan and a strategy and you are, at least for the moment, not that it can't change, but that you are intending to take advantage of what you believe as weaknesses in the alternatives. I mean, good for you. That's, I love that. We'll be cheering you on. From the sidelines here.

And so just, just for our audience, I think to step back for a minute, you know, how would you describe TraceLight's core customer? I know you've done modeling. I know you help a lot in modeling and consulting, but you also now have a Word app, a PowerPoint app. So you mentioned the 10 million as an example, right? You're going after a very small subset. You're not going after the billion. So I know modeling is a big part of that, but What within modeling, like how would you just describe your ideal customer? Someone's listening and say, hey, I want to try TraceLight. Who's it really for?

Our customers are primarily in management consulting and high finance. And, um, the, I think it's, yeah, that's fair to say that, um, you know, the, the big early share of those customers work in areas related to private market transactions, right? So, uh, that's valuations and modeling practices in management consultancies. It's financial due diligence, commercial due diligence. As well as the investors and investment bankers who are kind of working on the other side of the deal, right? These are very, very high stakes, very schleppy, as you guys know. Yeah, they're just very important work, right? And those users benefit a lot from a product that is better than the kind of basic horizontal products you can buy anywhere else. Um, so yeah, that's where we focus.

Got a question for you. I, and I thought, has the terminology that is now like, you know, in the kind of ether for most people, like harness and orchestration, have we just caught up to where teams like you were before? I'm assuming they aren't new words for this.

Yeah, maybe. Yeah, I think we've been talking about this and we've been talking about harnesses and orchestration for, for quite a long time.

But, but it is like harness has probably been the word of July and now it's all, I don't know what it would be now. but Loops was one recently, and God knows what else. But orchestration and harnesses are definitely buzzwords.

Yeah, and for good reason, right? They are really important. I mean, I think people— look, harness is a fancy word. It's like the environment that you build around the LLM, the agent, like the tools that you give it to do the work that you care about. But I think people have made mistakes there, right? And have probably suffered as a consequence. I think a lot of people built harnesses that just didn't price in the improvement in the models, right? So you can build a super opinionated harness where you really put the model on rails. And you do, back in 2024, early '25, when everyone was doing that, they their products were performing better, right? Because they had all these specifics baked in, but then the next model generation comes out and you've got to throw all of that stuff out, or usually teams actually just aren't intellectually honest enough to throw it out. So mistakes have been made at the harness layer. I think we've done a very, very good job on the harness. Orchestration, I think, Like, it's very important to be model independent, right? Um, that's another thing I was going to say when you asked about the, particularly about the big labs is like, there's a lot of nervousness in our customer base about like betting the house on Anthropic, like betting your entire organization just on Anthropic models when OpenAI could overtake them like next month. And, you know, maybe even Gemini one day, who knows, right? But like, I just, a lot of our customers just don't think that's a good idea. And I agree with them, right? I don't think you need to take that risk. I think you need to, you know, they buy us because they want to buy the vertically focused product that is built specifically for their needs. That's just going to be able to send them to the best model for any given task. So yeah, those things are very, very important.

So I have a couple of questions and I'll let Anand Giles go again. The first, when you mentioned the models, I'd love to get your thought on, are they going to all kind of become a commodity at some point? And the models will just all be pretty similar. Where do you think we're heading? Because it feels like outside of customized special, but the big, huge ones, how much can you differentiate yourself in a model when you're all training off similar data. So I just love a little bit of your thought there and you're already smiling. So I know you have an opinion.

I'm not going to claim to know the answer to that. I don't think anyone knows the answer to that question.

Sure. I just love your thoughts. I mean, if you're right, we'll come back and congratulate you. If you're wrong, you'll be like most of us. Yeah.

Yeah.

Well, look, I'll say a couple of things. The commodity thing, I mean, it's kind of baked into their mission statements, right? Like OpenAI always talk about intelligence too cheap to meter. And I read that and I think, is that good for your business model? Or like, they're always talking about that. Well, that means it's a commodity, right? So I think there is gonna be a lot of that and people will make a lot of money selling the complements to that commodity, right? Our customers included. And it's already happening, by the way, right? As As the power of the models grows, this intelligence grows, our customers, like our consulting customers, I think BCG grew AI transformation-related revenue at like 25% last year, something crazy like that. And they're just making so much money implementing AI in, you know, helping customers implement AI. So you're already seeing that effect where the complements are doing really, really well. The other thing is, I mean, I really don't know. There's been like architecturally, there's just been so little diversity in the frontier models. I mean, everyone's kind of pursuing the same thing, right? And scaling it. That's been the winning play for the last however many years. I don't know. It's perfectly possible that a startup or someone like Ilya releases a model that is architecturally different and has very different capabilities, maybe different spikes and troughs versus the existing models. And that would be amazing for companies like us, right? Diversity at the model layer increases the value of orchestration between those models, right?

Because your users can choose between them.

I mean, it would be so easy. Let's say Ilya releases this fantastic model, which is just very different from Anthropic's models. And then customers would be able to use TraceLight and we would use both the models for different tasks, right? And so the value of orchestration would grow. But people have been thinking that for a long time and it hasn't happened and it might not happen. Right? It might just be scaling all the way. I don't know.

And just to delve on that, because the way you just used the word orchestration, I'd never thought of, because to me I was thinking sort of within, I don't know, like a Claude environment, you have a sub-agent team of agents, but you're actually talking about the orchestration of potentially the use of different LLMs.

We already do.

Yeah. Okay. Oh, wow.

Okay.

Well, think about, if I can say one thing, I think Giles, when you think about it, you're selecting which LLM, which model to use within an LLM. They're just selecting which model to use of all LLMs.

But also by the sounds of it, but this is what I want to just make sure I understand. Orchestration for me would be like within a workflow, you're actually delving within different LLMs from different providers within a workflow.

Yeah.

So, um, For example, if you upload 500 pages of PDFs into TraceLight and say, here's contracts, read contracts, transcribe, and build a model, right? It would be Gemini reading those PDFs. And then it would be today Claude doing the modeling for you, right? So already happens in our product. And On our kind of pro tier as well, and our max tier, you can choose between using the kind of frontier intelligence, right? Or we have efficient mode as well, where we'll use a smaller, cheaper model, right? But people get way more usage out of their subscription. And today those are from different providers.

Do you have a strong view on, and again, this has come up in a couple of places recently, do you have a strong view on Choosing the most powerful model out there at the moment versus having a really strong harness and setup, however that is kind of implemented. Do you get a sense of what is better, if that's fair?

I don't think we've yet reached the point where the tasks where you care about are saturated by the, uh, behind the frontier models, right? So, I mean, if you are building a big complicated spreadsheet, you definitely do want to be using Opus 5 or 5.6, right? Um, and, you know, I, we will get to a point, I think, where that saturation will occur and it will probably happen quite fast, right? Uncomfortably fast. Where like, yeah, I mean, Opus 5 is really, really strong. And so when you've got, when you've got Opus 6 or 6.5, I don't know if it's worth spending all the money on the tokens to get 6 or 6.5 to write your formulas, right? You should be getting it to do something else. So, but we haven't reached that point yet, in my opinion.

I mean, listen, I love, I can't even imagine waking up every day and wondering, is there going to be an announcement that is just like, you do even sleep? Like, Every hour something can happen that might, you know, cause you to, you know, flip over or, I mean, it's just, I can't imagine being in a place that is changing so fast. But surely you, I mean, you must be feeling you're doing the 100-meter sprint every day and wondering what's the newest release. What's that like? Talk about what that's like wondering, obviously anticipating, Any day a competitor or a frontier model might release something that's different. What's the feeling like to be in that environment right now?

I really look forward to the releases. I don't know. I think you guys maybe imagine that I might be nervous about them or scared.

I'm not nervous.

I really, really look forward to them because I know that if a really, really strong model comes out, I'm gonna— we're gonna clip that model into TraceLight, and next day we're gonna have really happy users, right? And so I feel really, really good about that. And, you know, we're at a point now, we've signed a lot of kind of big enterprise customers, right? And yeah, and making those users happy, and it is, yeah, feels great.

So I love it.

Listen, I love it. I think that's the right healthy mindset anyway. I mean, you'd be— you couldn't do it if you were just stressed and nervous and worried all day long.

You can't. You can't.

That's not— that's not gonna be— but no, I love that. It's, um, listen, obviously we are— we are in the fastest changing, evolving segment of any market that anyone is. We talk, we have conversations and talk about events as though they're like milestone events that were like over a 5 or 10-year duration as they would normally be in most industries. And yet, they were talking weeks, right? Well, remember way back in March when we thought this? Well, that was because it changed from February, right? Like, we're all in the same year here still. And we're excited about where it goes. Do you want to talk? Can you tell us a little bit about— I mean, I know we've been with you for a while. Can you tell us about what the What is the— what is that? Well, can you talk about the future a little bit? Like, what is the future of Trace like? What are you talking about? What are your short-term dreams and the long-term? Like, what can you talk to, you know, our audience about? What does that look like?

Yeah, so I think, you know, we are in the position now where, like I mentioned, we've signed a lot of very large customers and we have some others sort of very close in the pipeline as well. And so, you know, when you're building, I mean, you guys have used our kind of pro tier, our max tier, to be honest, like this is not the main engine of growth in our company, right? The main engine of growth for us is enterprise revenue. And—

As it should be.

Yeah. It makes sense.

Totally makes sense.

And when you sell into enterprise, you know, it's different from consumer or prosumer. There's this lead time, right? You've got to sign contracts, you've got to get through the InfoSec, you've got to, you know, get through all of this. And so that part, or we'll always be doing that, but a big part of that now is done for us. So now we're in that kind of super exciting growth curve now in enterprise, right? Where we're deployed and we're just expanding into these accounts. So The revenue's growing really, really fast now. And obviously, you know, such a large market that we're selling into, right? There's such a big opportunity here, right? The way I think about it, you know, is you've seen coding, right? You've seen the Verse, you've seen the Curses, the Cognitions of this world in coding. You've seen Lore, Legora, and Harvey and what they've done. Banking, a little, you know, Maybe Rogo has done pretty well in banking. And, you know, in management consulting, you know, we think we're going to be the next one of those, right? And so that's what the next kind of 12 months is about for us is like really, really scaling revenue and scaling the number of users in those places.

I love it. That's amazing.

And I'm getting— would it be— I'd be right in saying, you know, once you agree, know, to some sort of enterprise deal, I'm guessing TraceLight becomes quite sticky because it must be quite an effort to get TraceLight in. I guess on the reverse, to actually then step away.

Yeah, no, you're right. You're right. So, I mean, it's so much work. It's so much work to get, you know, to go live, right, in an enterprise. And so you do have that kind of baseline level of stickiness, but that's not the stickiness we aspire to, right? I think user love is the real stickiness. And, you know, we, yeah, that's going like really, really well, right? And so that helps you also expand into the enterprise, right? Because, you know, you can have a beachhead into a very, very large company, but you are far from done, right? You have the whole company to go sell to. And so that's the kind of phase that we're in now. And then, yeah, like I mentioned, everything that goes underneath the app, the layers beneath the application layer, the kind of memory architecture, the way that you make TraceLight learn in a compliant way, right? I mean, these organizations, they have a very specific way that knowledge and information is allowed to run through the organization. When I was a consultant, I could do a project, And I could not tell any of my colleagues, right, what was— what happened on that project or who the client was and all of these things. And yet I could kind of learn generally from it, right? I could, I could bring that experience and that judgment to the next project. And, and that's kind of the, the trick here, right? And, um, and you need to build that same thing in digital form, right? If Tracelight is going to get better every single time that they use it. And it's just very, very deep and very specific work. And there's actually a lot of actually unsolved research stuff there as well. So yeah, we hope that that will be the next thing that will make TraceLight more and more useful and entrenched, obviously.

I love it. I mean, you are obviously aware of other firms, ancillary peripheral firms, whatever to you, that are not around today. They were around a year ago. You are clearly hurdled a number of critical hurdles by now. I think that's phenomenal, and I think the market will be better off with your success. And, um, and so we're excited to continue hearing about how you are doing and performing. Um, Giles?

A couple of things that jumped out to me, I think, already is number one, your excitement at new LLMs, which I, I suppose I'd never really thought about. Um, and I've forgotten what the second one was. My memory now, I'm in my 40s, is—

You got too excited.

Truly shocking. The second one was better as well. I'll pull— I'll hand it to you. I'll think about it. I'll, I'll figure it out.

Thanks for that moment, Giles.

That was helpful. No worries.

Yeah, for that seniors moment, it was great.

I know we all have it. 42. What was the second one? It's great.

Well, you keep thinking, uh, speak up. I would love to know— I know you guys have added to the product. Are there other tools you plan to go into? Expansion? You've mostly been focused, as you mentioned, kind of the due diligence, the modeling, whether it's helping with the, you know, Memos in Word or research there in PowerPoint. And I think Excel are the 3 main ones you have right now. Are you planning on expanding beyond that to helping with, you know, these valuations and these very specific niches? Is it always gonna be kind of that finance area in consulting, or do you look to go to other areas? Or are you looking to add more apps? Or how are you thinking, as you call this a platform, kind of that growth play as you continue to scale and grow and want it to be more used?

You know, I don't think it will be growth in terms of more add-ins, right? I mean, maybe there'll be an Outlook add-in or something down the line, but I don't think that will be the main vector, right? I think the, the main growth will happen within our web app, right? And so this is actually the central place where more and more, um, we expect users will, will be doing most of their work, right? And, you know, actually, I think You'll still want the Excel add-in and the PowerPoint add-in for the last mile. But I think that's going to decline in importance, right? As the LLMs get better and better, I don't think you're going to be wanting to spend so much time hunting through cells, right? You're going to have a system that gets that done correctly and shows you that it's been done correctly. And then you're just going to want to think at a higher level of abstraction. It's the same thing that's happened in coding, right? And so I expect the growth to be growth in usage in our web app, right? Where, you know, it's essentially a project workspace. You have all of your input files, you know, all of your agents and chats and all of the outputs that you're creating. And we sync it with SharePoint and, you know, you can send emails from there, you can bring in emails and Teams messages, et cetera. And we, you know, this will be the central hub for how these projects, uh, are run, right? And. Um, and so that, that, that's where I do expect to see the growth in terms of users. I mean, I, you know, we still have a lot of, you know, a lot more people to serve in our kind of core areas, right? So like I mentioned, valuations and modeling, FDD, CDD, um, these, um, you know, and, and the investors and bankers that these are, these are the people that we serve, but, you know, we, we also do have users in audit. now, right? And they love it too. We have users in tax and that's going really well. So I think those would be the kind of slightly adjacent user bases that we'd look to expand into.

You know, I've got a question that might be a little bit crazy, but tell me what you think about it. I mean, often companies that are succeeding in one part of a vertical at some point get worried because of the, uh, because they, they're part of a chain. Is there any merit in discussion? Is there any feasibility one day to abandoning the existing Frontier models and creating your own or somehow leveraging your own so that you don't— I mean, you've just said you're excited when a new model comes out, which is great, but there's also presumably some risk that, that, that they can change their mind. I mean, it's conceivable, I guess, that, I don't know, maybe they'll find ways to restrict firms like you from using their tools. Like, is there a future where you don't have to rely on them at all and you can create your own frontier LLM? Or is that—

It's— never say never, right? I think there is a risk there. It's not a risk today. I think it would be a risk that we'd be thinking about maybe a few years from now. But OpenAI just did it to Cursore, right? Cut them off from their models. And that was a—

Is that—

would that be going down?

It's plausible that that could happen. I think you have seen vertical startups build their own models. I say it like, it's like, quote unquote, build their own models, like do a lot of post-training on kind of open source bases. And I think the Cursor guys did a really, really good job at that, right? I think Composer is a really strong model series. If I look at our internal coding model usage, I think there was quite a lot of Composer. I think that's actually now mostly been replaced by Groq. So, you know, people get Fable to do the kind of like big hard planning and thinking, and then they use Groq to do the work. But I think the Harvey guys are training their own model. I think usually when people do that, the first reason is actually cost, right? So your job is to get your customers the very best ROI on every token that they use with you, right? And I think it's extremely unlikely that you're going to train a model that is going to be at the frontier in terms of pure capability, right? I think that is the lab's job. But it is possible, right, with enough data and enough expertise that you can train a model that is on the Pareto frontier, right? And even if, you know, but it's much more efficient. I, you know, I won't commit firmly to it, but I think there's a, there's a, there's a world where I could do that.

Cool. No, no, I can't expect anyone in the space to commit firmly about anything beyond dinner tonight. So, I mean, I, uh, even, even that's a stretch, right, Giles? But I, uh, I, I have no idea. That's hours away. Anything could happen before then for me.

Yeah, I was gonna say, if you did commit, we would have been wondering.

Yeah, right. I But I know that it's helpful to think about that. I know you must have been thinking about it because, of course, as a responsible company, you're thinking about all the potential future permutations and what the potential risks are. And what does that mean? And where do we allocate our resources? I mean, you'd be no different than anybody else. So, I'm sure you are. But anyway, I just was curious to know what the major bottlenecks would be. Anyway, you've given a great answer. I appreciate it. And it's been fascinating to have a small window into your world and your lens right now. So, Paul.

I remembered my second question.

No, you did. Come on, you did. Wow.

I remembered it.

See, it just takes a few minutes sometimes.

Yeah, this happens too many, too many times in my daily life now. So it was just, again, interesting, like hearing you talk about the, when you get into an enterprise, and I don't think this is revolutionary what I'm saying, that the journey starts once you kind of agree that deal. And actually it feels like a really unique environment where you probably see what we're seeing as trainers, where people are genuinely like, we want to spend money on AI and we don't know what we want to do with it. And I struggle to think of any other topic or category where it's been so strong. Like, we are ready, we've got FOMO, take our money, but what do you want? We don't know. And I don't know if you are seeing that. It must be quite a nice environment if you are to be like, Okay, we've got the tool, we can now help you figure out how to use it.

Well, I'm not sure our enterprise customers are saying take our money. Sometimes they really negotiate, really annoying. But no, no, the way I think about it is, you know, our end markets are historically like very, very bad end markets for software, actually. Right? I think it's hard to think of true category winners in software that have been built selling initially into, I'm gonna say high finance and management consulting. I think there are some parts of high finance, like on the data side where that's not true, but selling into, you know, the way that we have been. And I think probably about the time that you guys first tried Tracelight, that changed and the world just changed. And after we launched, the amount of inbound from every single institutional brand name that you can think of was insane. And most of our funnel to this day is inbound, right? And amazing.

And wish I had that problem, right, Giles?

But that's what I mean, it's so interesting.

No, but you're right. So you're exactly right. All of these organizations woke up and they all put together task forces. And I was initially very skeptical. I was like, don't sell to the task force, sell to the managing director. But actually these task forces were real and they really bought stuff and their people were really motivated. So it's been, it's actually been amazing to watch that, watch this kind of category change in the last year or so. So yeah, I think now the challenge they have is there's a lot of vendors out there. And they need to figure out which one is best. And obviously, I think that's TraceLight. But so there's a lot of trials going on, there's a lot of pilots, and then there's a lot of change management on their side as well, right? And then you get the You know, the real worry that a lot of them have is that they'll just buy the wrong thing. That's the thing that they're most worried about, right? Because then, you know, 8 months from now, they'll have to do a whole change management program again. And that's not really what they want to be doing. So our job is, you know, we'll have our foot in the door, but our job is to prove to them that we are the long-term partner.

Nice. A fun time to be alive and running a business. I love it. We can live vicariously through you.

Well, it's interesting when you compare, again, I'm not going to say too many nasty things about Microsoft because they will kick me out of the MVP program eventually, but Microsoft are so lucky in the sense that they are the stickiest company. And I agree with you, like a year, especially a year ago, or even more. Copilot was awful. I mean, I'm now in a position where I can train it because it's good enough in lots of areas. It's certainly not perfect, but my God, I don't— I'm not sure any other company other than Microsoft would have gotten away with how bad it was and how overhyped it was by them for so long, which is crazy. So if you're going in and from day one, like you say, you're focused on like give the customer a good experience. It's doing what we said it would. That's awesome.

But I think, so yeah, I continue to be confused about how bad Copilot is, given how strong the models are that are powering Copilot, right? But I think Microsoft does have that innovator's dilemma problem now. And let's just take Excel. I always compare Excel with coding. And one big difference I think between those domains is coding was really nicely set up actually, even before ChatGPT, coding was really nicely set up for AI to come along, right? You had IDEs, you had a really nice collaboration environment, you have diffs, you have Git, you have version control, All of these things are really important if you're going to start working meaningfully with someone or something else. And in this case, that something else is AI. Excel is just, it's just missing all of it. And, and that's, that's a huge problem. So, you know, now if you go into the TraceLight web app and you say, here's an Excel, I want you to make the following 25 changes to this Excel. What you'll get back is you'll get a diff. You'll get, okay, here's, you know, here's an overlay kind of thing of what the file was like before. Here's what the file is like now. And as you build up these diffs, you can then build up versions of what you're working on. You can roll back and you do all of these things. And yeah, I know, I know there is some versioning, et cetera, and diffing, but it's completely insufficient in my view in Excel. So yes, it's Copilot being slightly disappointing, but it's also the product right around, you know, I think that some big changes have to be made to make it really nicely set up to work with very powerful agents. I think they have a bit of a problem there.

It'll be interesting to watch. I think there's some good points there. And I've said to get what they really need, you almost need that kind of online whole new environment because it's 40 years old to truly have that modern innovative platform. But given the constraints they've had, they've done an amazing job, in my opinion, overall on Excel. Obviously, AI side, there are some challenges.

When you're a $5 trillion company and you have 7 or 8 billion customers, we should all have such problems one day, but you know, it's a whole different set of criteria and internal machinations, and it just obviously is different, and you have to be careful about People all over the world using things. I didn't even realize until recently how many people are not even working on Excel 365 yet. I mean, they're still using older versions of Excel that don't even have. So there are lots and lots of things that factor in when you are, you know, the large player that the whole world relies on. But that said, I mean, but that's why there's opportunity for TraceLight and others to take advantage of speed and nimbleness. And you're obviously doing a fantastic job with that. that. So anyway, I thank you for spending time with us on an evening obviously as busy as you are.

I think we'll go ahead and kind of wrap up here, but before we do, any last thoughts? Anything you want to share? Any thoughts about the future modeling? Anything you want to say to our audience here?

Thoughts about the future, or wherever you want to go, it's up to you.

Just kind of last thoughts.

One thing that I think about and talk about most days, you can be Quite scared, right? Looking at how this stuff is evolving. And it is quite a scary experience, I guess, if your career to this point has, you know, your unique, you know, value, right? You've been the best at writing the formulas and building these complex models. I can imagine that it's very scary then watching LLMs, you know, the curve is just like this, right? And even if they're not past you today, they probably will be past you quite soon, right? And I totally get that. But I think people, your viewers, our users should all just be tremendously excited about what the future holds, right? Their jobs will be different, but the amount of value that you'll be able to add in the world is now, it's just like increasing exponentially. And I think by this time next year, if your job had been building big Excel financial models, and then handing that over to a customer, I think your job will be building those models and then that becomes code and then that becomes a persistent part of how your customer is operating. And like, just the amount of value adding to them is so much greater. So I am very, very optimistic, right? Even about the users that most of the rest of like Twitter, if you read it, everyone's like, oh, these, you know, these people are so cooked or whatever, because the LLMs are doing their work. I just completely disagree. That's how I feel about our users. And I hope that's how your viewers feel too.

I love it.

Love it. Well, thank you. Really enjoyed it. Appreciate you carving out some of your evening for us. And Giles, anything you want to add before we say goodbye?

No, exciting to chat with you. I'm thrilled and rooting for you and excited to hear how things go. So thanks for spending time with us.

Awesome. Thanks for having me.

Echo that. It's, uh, I'm happy you're, you guys are doing so well. It's been really interesting just delving into your mind and your outlook a little bit. And there's some things I probably wouldn't have thought of. but make a lot of sense. So, so yeah, wish you all the best with it.

Well, thank you so much. We really appreciate you carving out some time.

Take care, Pete.

Thank you.