The Exchanges

Q Amazing. All right. Uh, so much to unpack here. So one, um, that you alluded to is the coordination. So if you have three agents and the future dozens of agents, that concept of supervisor, is that a supervising agent? Is that a human? How does it all work together in an orchestrated manner?

A Yes. So it's a supervising agent, but I do believe that in what we do, they will always be on top of that a supervising human. Because we deal with very, very sensitive data, and so you want a human to always be here to make sure that we are going in the right direction and put the right galleries around the framework that we're giving to the agents. So maybe I can give an example that might be easier to understand. So, um, if you think about these agents working together, so, uh, if you take, if you take, for instance, an analyst, so a financial analyst, let's say like, uh, so Tuesday morning, nine AM, and they have two days to run an analysis for, uh, their CFO. Um, I'm inventing that on the go. So we'll see if that goes somewhere. But imagine they have to do that. So what you will have, uh, is they will give a prompt to the agents and, and, and first that will trigger the analyst agent of understanding. So let's say for instance, they, they want to do budget variance analysis. We were talking about that before. That will trigger a first action, which is understanding budget variant analysis, perhaps across 50 products, 20 countries, I don't know how many cost centers. So already imagine the complexity of that. It's a lot, because that's things that could Perhaps take you months or maybe you would not even be able to do that. So that's the first step it does. And then, um, it…

AI assessment note: “it's a supervising agent, but I do believe that in what we do, they will always be on top of that a supervising human”

Q Partners. You have some, some, uh, big name partners from Google, but I'm almost most interested in the SIs, I believe Deloitte and others. It's one of the recurring topics that like every board, every startup talks about. Like at some point you want to start scaling through partners, but it's always harder than it seems. So any lessons learned there?

A Sure, that's a super interesting topic that is hot for a lot of companies to get their head around, I think, because it takes a lot of time. Um, it's, um, so we partner with Deloitte, we partner with EY, we partner with PwC, and then we partner with many boutique firms. What you see is, it's a little bit like when you build a business, is that you get velocity with the boutique firms first. It's exactly like you get velocity with your SMB customers before you get velocity with enterprise. It's the same here. So, boutique firms are the first thing to focus on, I would say, at the very beginning, to make sure that you have a network that is well-defined with clear guardrails around, you cover that geo, you cover that use case, and the, your friend over there, competitor friend, is not going to do the same because otherwise your guy's going to get upset or whatever. When you've done that, you need to build, start building in parallel the GSA motion, and that takes a lot of time because you need credibility, you need to see that you are serious about what you do. And for them, the other day, for instance, I was with a Deloitte partner, and they were telling me, for them, their yearly so-called quota is twenty-five million. And for those who don't know, in enterprise software, the quota is more between one and two million if you're lucky, right? So, very different numbers. So, you n…

AI assessment note: “boutique firms are the first thing to focus on, I would say”

Q In, in venture capital. How big is the company now in terms of a number of employees or any stats you can share?

A Sure. So today, um, we are 500 employees at Pigment. Uh, we are still growing, uh, We are, so we, we've been growing very fast since the beginning. There is real demand because of, of everything that I just explained. Uh, we raised four hundred million. Thanks to you, Matt. You were one of our first investors, and, uh, and I was so happy, actually, to, to, to get to know you, uh, now, about almost six years ago. Um, we have 60% of our revenue in the US. Um, we serve, uh, more and more Fortune 500 customers because our platform is really for large companies, so, um, We have really expanded there, and we are growing for X the number of large customers that we have at Pigment. And so, you know, that's just the beginning of the story, but very exciting for, for the future.

AI assessment note: “we are 500 employees at Pigment. Uh, we are still growing”

Q So tell us about, uh, yourself. What was your founder journey, uh, to, yeah, that led you to here?

A Yes, so, um, I, I'm an engineer, uh, by background, actually, so, um, I, I studied engineering, I studied fundamental physics, um, and I think I had always had a passion for creating and understanding the world and, and trying to see, you know, um, how could I have myself an impact on the world, so trying to really learn as much as possible. Um, and I also always had a passion for freedom, I would say, and, and, and very large impact, and so when I did my engineering studies, I actually studied entrepreneurship in parallel, um, because I knew that at some point I wanted to create a company, and so fast forward, I spent most of my career abroad, uh, came back to Paris for pigment, and I actually discovered the work of enterprise performance management during my time at Google, uh, when I was working as a data scientist for the CFO of Google Media and the CFO of Alphabet, And I discovered how much enterprise performance management could be difficult when it's managed on spreadsheets, and that's how we started. Uh, I did index ventures after that, and I saw, I would say, the other side of, like, founders, uh, struggling, uh, with everything, uh, with everything, uh, planning related, performance management related, and that triggered, uh, my, uh, my, uh, my real, like, the, the, the, the willingness to actually, uh, start, start the company.

AI assessment note: “that triggered, uh, my, uh, my, uh, my real, like, the, the, the, the willingness”

Q Why three and not one? Is that just easier for people to wrap their minds around or is that more of a Technical constraints.

A So I think, um, today, these agents really need to be separated because we are still at the beginning of, uh, of the, this agentic framework for, I think, any company out there, and they are working on very different things, very different sets of data, and very different ways of working. The analyst is working more, like I would say, on the past and present. Think about it more as a BI analyst, almost. The planner is really working on, with machine learning, with forecasting, and obviously Uh, handling very complex tasks to, to, to, to, to actually carry on multiple forecasts at the same time, et cetera. And then, uh, the modeler is really here to work on new models, on new formulas, think about it as cursor or other companies that out there augment or whatever for finance team, for HR team, for analysts. So it's a very different task, but over time what we want, um, and what the, the user will see is really we'll have the supervisor on top. That is really the only way that a user will interact with a platform and the user will just, you know, Give the agent, um, uh, a set of comments, and, and the supervisor will then decide, okay, uh, that is gonna be done by, uh, the analyst, and then we're gonna pass that task to, uh, the, the modeler, and then to the, to the planner. And so, the, the goal over time is that there is only one agent. But before that, I think we're gonna laun…

AI assessment note: “these agents really need to be separated because we are still at the beginning”

Q Meaning with AI, or what, what, what accelerates the timeline of a PhD?

A For sure, for sure, it's AI. Definitely, I think, you know, AI will trigger so many ways to discover great things, you know, in, whether it's in biology, in physics, anywhere else, it's like, you're going to be able to have, like, the fastest feedback loop to discover a new process, a new protein, a new way to do something, and that's so exciting. I'm pretty sure today, like, I, I would imagine, you know, you know, I was thinking about something is that, One of the reasons I actually went to Google, I think, was because of Demi Sassabes. It was very early stage of what DeepMind is today, but I was so impressed by everything they were doing, and I think that really triggers my, my willingness to join Google. But the fact is, um, I do think that when you hear him talking today, for instance, about the power of AI and on what potential AGI will bring to the world, I do think that every timeliness is going to decrease, and perhaps we're going to go back to a time where, you know, Nobel Prize will be won by People that are like, 25 years old, like Einstein or whatever, just because you are able now, in a PhD, to discover the unknown, and that's so fascinating and amazing.

AI assessment note: “For sure, for sure, it's AI.”

Q Yeah. And, uh, because you work on centralized planning, like you, you very much building the sort of operating system slash nervous system of, of any company, right? So you, you're ideally positioned to do that.

A Yeah, I think we are very lucky because we are the single source of truth. We have the data. We know how to operate on that data, model that data, and we know how to change things very efficiently. That's also the other thing is that we have a very agile and flexible platform. So every time you want to change something, You can change it super, super fast. And with a moderate agent, that means really making new models evolve. So, um, you have, uh, imagine you have a war and indeed your supply chain process change. Boom, immediately. If you need to remodel something, do something in the platform, it's super easy. So that's, that's, uh, it's, um, I think, uh, as we were discussing earlier, the, the, the problem will not be so much innovation, but the capability for enterprise to truly adopt it at scale. I think that's going to be one of the big issue, but, uh, I, I do think that, uh, the companies that will do that, uh, They will thrive. They will be the best companies out there, uh, in the very near future.

AI assessment note: “I think we are very lucky because we are the single source of truth.”

Q to code and do you need to still like study computer science in a very sort of classical way so that you can sort of Deal with the output of the model, or is that the wrong way to think about it? In the finance and planning world, like, do people, how does one become a CFO or VP of finance in a context where the models do so much?

A Well, so first of all, I think if you're a software engineer today, I mean, and if you, if you, if you want to study, like just go, I think you still need to understand fundamentally what's happening. And I'm sure over time it will be new languages, et cetera. But, but I mean, the fundamentals of what coding is today, you are not able to use cursor. Um, if, if you don't understand coding and I know the cursor will evolve also towards, uh, probably a more advanced version of lovable or whatever, but like, It's still, you need to understand the fundamentals in order to build like an app, to build a complex model. Um, and so for finance, I would say almost it's even worse. You are in a heavy regulated industry where you cannot afford mistakes, where you report your results to the street, where you, I mean, if you don't understand the concept of a balance sheet, if you don't understand the concept of a P&L, if you, if you don't understand these things, you are, it's, It's gonna be very, very hard for you to know if these agents are going in the right direction, and, and also, um, it will be, you will probably use more of your finance skills, because instead of, of going back to trying to, uh, collect the data, plug your systems, try to make sense of the data, now you will actually have the time to properly analyze it, understand it better, finding insights that you might not have f…

AI assessment note: “more than ever, You need to be highly, highly skilled”

Q And when you were in school studying Quantum physics. Did you consider doing this as a career, or was it always clear that you'd be doing something else?

A Absolutely. Actually, I, I thought first, uh, that I was going to do, uh, probably a research career, um, and at the time, um, so I think, so I studied that in France, and at the time, we had one of the best, I think, master of, uh, of, um, machine learning in France called MVA, and I actually wanted to take that role, but then I realized that, um, the problem with research, especially fundamental research, is that the time lines are very, very long, Um, and usually it takes you, you know, more than 10, 15 years to start seeing the results, and the problem is that I think I don't have the patience. I have too much energy to wait 1015 years to actually get to see the results. I actually think that maybe today there is never a better time to do a PhD, because I actually think that now, probably in two years, you can actually, actually achieve things that you would have done before in 15 years. So that's quite exciting, actually. It's a good time to be in fundamental research.

AI assessment note: “I thought first, uh, that I was going to do, uh, probably a research career”

Q Yeah, Elizabeth, you mentioned, uh, automating some mundane tasks and then like doing things that humans cannot, cannot do. So like, what, what, what is that?

A There are so many things that humans cannot do today, and there are so many tests that would literally take a lifetime to carry on. Um, so, you know, if you really want to do, like, uh, think about, like, again, um, uh, any large company out there have a combination of, of products and, and countries and business units and cost centers that, you know, makes the combination really, really hard to understand at scale. That's just impossible when you carry on an analysis to do that. Or if you really want to, let's say, um, Improve your plan. You want to improve your margin, and you want to actually understand all the potential levers to get there, and then you want to understand the variety of scenarios. Maybe you're gonna literally want to run 5000 scenarios in parallel on an infinite quantity of data. That's just impossible to do for anybody. And so, um, there is that, and there is also, uh, as I was saying earlier, sometimes the problem is just people are time constrained. So, Perhaps, yeah, maybe in three years they could get to a result, but they just don't have that time. So I think this is, uh, what is fascinating, and these are just, like, the first example that we see today with, uh, with, uh, with our customers on how they are using the product, but I think it's gonna unlock a lot of use cases that we have no idea about today. We have no idea about, and that's, again, go…

AI assessment note: “run 5000 scenarios in parallel on an infinite quantity of data”

Q Actually, while we're on the topic, um, presumably, uh, you guys are heavy users of AI productivity tools like coding and other things. Maybe one or two thoughts on that. What do you use and what do you find helpful, not helpful?

A Yes. So we try to push it. So we have an internal AI committee. So we really try to push it across teams. And we actually have a committee to make sure we know What we are doing within the company, because we could end up buying 10 times the same tools and, you know, everything. Yeah. So we try to put guardrails and also, as you can imagine, with Pigment, we handle very, um, important data and strategic data for our customers. So we need to make sure that it's okay to use an AI tool and buy it kind of bottom-up when you're writing a blog post, but it's not okay to use on Pigment data or on a, on a customer call, for instance. So we, we have this incredible committee now. And so I think, you know, we really use it across every single team. So obviously from cogeneration in, uh, in engineering to everything in marketing, like generating content, generating SEO, helping us with literally every team in marketing. Um, we have built also our internal AI tools with the growth team that we have internally to help actually feed the right leads to the right seller at the same time, know exactly. When someone might be ready to buy, etc.

AI assessment note: “from cogeneration in engineering to everything in marketing, like generating content”

Q All right, so all of this is the, Perspective of Pigment as a builder of Gentic AI. Now let's switch to the customer side. So what's the vibe or the mood currently? Do people, people are certainly excited about all of this, but how do they react when you start talking about, are we going to have a planner agent and the analyst agents?

A So I think I would probably distinguish, uh, uh, two, two types of, uh, companies slash, uh, customers or prospect is on one side, you have people that are using pigment today, and I think they all get incredibly excited about the technology because they already love the technology. Otherwise, uh, uh, you know, they would not have adopted, uh, pigment. So, um, I would, I would differentiate that from like some enterprise that are, Further away in their journey to adopting technology where, uh, you have to start from a very different point of view. So I would say clearly today the analyst is used amongst our customer and the excitement is incredible because they really see already that not only it helps them save time. Uh, we have so many examples of customers that are already giving testimonials about, you know, how much time they save on that. I even got, uh, last week just to, it's a funny example, but, um, Uh, the super self-founder, uh, who, uh, is still the CEO of the company, and, uh, so it's a company behind the clash of clans and, uh, bro stars, uh, and they, they, they have been happy customers of pigment. Last week, he was literally walking past the finance team and he overheard a conversation of the finance team singing praises about pigment and pigment AI. So he recorded a video and sent it to me about, you know, how much he loved pigment and how much he's changed t…

AI assessment note: “the excitement is incredible because they really see already that not only it helps them save time”

Q Yeah, so this implied in a couple of things you said, including now, so there is a concept of constant feedback loop, where the system gets better. Can I sort of declaratively input what my policies are, you know, at Coca-Cola, what company XYZ, this is how we do planning, this is how we look at data, or does a system, uh, has to learn?

A Yes, so it's, I would say it's two things. So I would think about it when you start as a, as if you were going to train a new analyst in your, that has just started in your team. So you need to train them on the concept and on your constraints as well. So I think it, it can be, it's not as powerful as what you describe. I think It could go there. It's not there yet. I think today, um, it's more about to really act under constraint. So to know that, you know, uh, you are never going to be able to hire, you are maybe a 200 person company, you're not going to hire a thousand person next year. That doesn't make sense. So it's going to be able to work through that set of constraints, and I think over time we'll be able to input even more rules around really how we work. Uh, already today also you can fit them into the way you do workflows, and it can follow the workflows you want. So, Today, really think about it as, at first, when you're going to set up the agent, you're going to, you're really going to talk to the agent, and, and explain to he, and, and, to him or her, I don't know, to it, a couple of things. Um, um, uh, you know, there's a problem in French, it's because everything is a...

AI assessment note: “It's not as powerful as what you describe. I think It could go there.”

Q It sounds like UI UX is a very important part of how it all works. So, um, just for, for people to understand and visualize what's happening, you, you expose all the steps and each time you have the human in the loop, Validate. Do you, do you expose a score of, um, um, you know, explicit confidence that this is a hundred percent accurate?

A So yeah, so the way it works today is, um, really every time you carry on the calculation, whether you do it with agents or without, you can see the anti-OD trail. And you have to even understand that Pigment, sometimes with some of our customers, we, we power a lot of public companies, is used as a governance platform. So we log everything and our agents log everything. So they log even more than humans in a way, because I cannot tell you the number of them. I'm sure you see that with your portfolio companies that sometimes you come to a result, you don't even know how you came to that result. With agents, that's amazing because that's the magic. You can really log everything very, very precisely and gather all that information. So that's really how pigment works. And then obviously we carry accuracy tests along the way. And we, we make sure that we expose that also to our customers and that we work with them to obviously improve that over time. But the goal for us today, uh, to give a rough idea is to say we want to be more accurate than a human being, which is not hard by the way, I'm joking, but it's, it's, uh, it's, uh, it's, it's, it's really what we try to do is to, and this is the feedback we're getting from our customers, by the way, is that we are more accurate than when they were doing before.

AI assessment note: “we carry accuracy tests along the way. And we, we make sure that we expose that”