- 18 hours ago
Venture capital has always been equal parts science and art. Today, AI is rapidly transforming the science: due diligence that once took weeks can now be completed in hours, and sourcing platforms scan millions of signals to surface the next promising startup. With more than 90% of VC firms already using AI in their workflows, data is becoming central to how deals are identified and evaluated. But when everyone has access to the same tools and insights, where does competitive advantage come from? Does great investing still depend on conviction and chemistry? This panel explores how AI is reshaping venture decision-making: whether algorithms will enhance or erode investor judgment, how founders are increasingly evaluated by automated signals, and who ultimately holds responsibility when AI begins to influence investment decisions.
Category
🤖
TechTranscript
00:01Hello, it's great to be here in Paris today. I'm Maya, a reporter at Sifted, the FT-backed
00:08media publication covering European tech startups and venture capital. As this title of the panel
00:15suggests, the last few years have seen a wave of AI-powered investors. This new generation
00:21swaps human intuition and personal chemistry for billions of data points. One such extreme
00:28is the London-based firm Quantum Light. Launched by Revolut founder Nick Staronsky, Quantum Light
00:35claims to be the world's first systematic VC and growth equity firm. It's built its own
00:41proprietary model to track 700,000 VC-backed companies and is reportedly in talks to target
00:49$500 million for its second fund after closing its first in May last year at $250 million.
00:57At the same time, we're also seeing a wave of new AI startups, companies built around
01:03AI from day one, which investors are hot on. According to Sifted data, so far this year,
01:10they've received nearly double the amount of funding that they received during the entirety
01:15of last year. These startups, like Stockholm-based vibe coding giant Lovable and London-based
01:22synthetic voice company Eleven Labs, are raising at increasingly high valuations, automating
01:28work streams, and shipping new product features weekly. So, how does all of this affect the
01:35art and science of venture capital? Here to discuss it with me is Christian Nagel, co-founder
01:43of early-stage pan-European VC, early bird. Bets of the firm include Black Forest Labs,
01:49Aleph Alpha, and N26. We've also got Pauline Roux, managing partner at Alaya, a deep tech VC
01:56which has backed Miracle, Alice and Bob, Akima, and Vibe.co. We've also got Hiroshi Matsushima,
02:05senior investment director at Sony Venture Capital Europe, the CVC section of Japanese conglomerate
02:12Sony. His focus areas include gaming, sports, animation, music, and Web3 technologies.
02:19And last but not least, we've also got Claymo Aglietta, CEO and co-founder of EDA, an all-in-one
02:26AI-powered solution for investors to manage deal flow portfolios and contacts across private
02:32and public markets. Thank you all for being here. Thank you. Thank you. Thank you. So, to begin with,
02:38a sort of open question. Quantum Light's CEO told Sifted last year, we just believe that machines are
02:47able to do it better. Not only do they have perfect memories, they are also not swayed by emotion, by
02:54fear
02:54of missing out on a certain hyped deal. On a scale of 0 to 10, how far do you agree?
03:03Claymo, as the non-investor, you can go fast. I think it's true and it's right and wrong
03:11at the same time. So, from 0 to 10, if I have to go very straight to the point, I
03:15would say
03:16probably 6. Yeah. I don't think it's completely true because there is a lot of, you know, meeting,
03:24you need to understand people and whatever. But I will let the other people say, but I would
03:28say 6. I'm not going to rank, but I think it's part true and part challengeable in the sense
03:37that all these AI tools or processes that we can put in place have two major benefits. First one is
03:45to
03:46make sure you cover everything and you miss no ball. And also to be faster in the way you source,
03:56assess companies. On the coverage thing, I think it does it well, but from a certain stage. And I have
04:04an example with Quantum that has invested in one of our best performing portfolio companies. It's called
04:10Vibe.co. It's operating in connected TV advertising and where the agent and the AI was great, was into
04:20sourcing the deal in an industry that is clearly unsexy at the moment. At tech is not an area of
04:26interest for no one, but because they really managed to track and cross-read quality of execution, they
04:32were able to identify the deal. So, to me, yes, it helps a lot, but there are a couple of
04:39things on top
04:40that the machine can do itself. Well, I would say it's, with regard to the process, I would say it's
04:49eight, if not even nine. If it's about kind of the quality of the decision, I think it's the other
04:55way around. But if I look into the process, and this is my point, if we, in terms of identification,
05:02there we use a software basically. We see everything out there. So every company, even many companies,
05:06before they're registered, we see them today. And everybody has kind of built something in this
05:10direction. And so we see them very early. Then it comes basically to building conviction. Well,
05:15that's a little bit the thing with simply in between, because obviously there you have to
05:18build conviction on something. The engine can support you in this. And then it's about winning. And
05:24winning, I would say, the engine cannot help at all in the end. Because it's about relationship,
05:29it's about convincing, and so on. Then it's about support. I think the engine wins already again.
05:35So in terms of finding out really, and finding customers, and analyzing markets. So it's a little
05:41bit in between. And it depends obviously a lot in which stage you're investing. So this personal thing
05:48matters more in the very early stage, seed in early stage, rather than if you would invest in data
05:53stage, there the engine, I would say, could almost take over entirely. So I would say six up to seven,
06:02because I'm working for CBC, so I might have a different view. But you know, financial data,
06:09you know, all the hard data, you know, AI can work faster than us. But we as the CBC, like
06:15let's say me,
06:16so working for Sony, we also have to find out whether the startups is actually,
06:22we can assess, or, you know, work as a strategic partner. And also, we have to look into a
06:28characteristic of CEOs, founders, whether that guy can really contribute to our creators or not. And
06:34that is not what AI can, you know, look into it. So I would say six. Great, thank you. A
06:40nice
06:41sort of range of answers there. Christian, you spoke about kind of having software, what, how has AI
06:49sort of changed the nature of what you do at EarlyBird? Yeah, quite a bit, I must say. So we
06:56have been, first of all, we have this engine basically, which helps us, especially on the
07:00sourcing side and on the support side. But we have been organized, and most of VCs have been
07:06organized according to sectors. So covering fintech, insurtech, covering enterprise productivity,
07:11covering deep tech sectors. So most of my peers have been organized like this. And we realized,
07:16okay, well, then came AI. And then we said, okay, who's taking over AI? And then we thought,
07:20okay, one partner should took over AI. And then we realized that makes no sense at all. So two years
07:25ago, we completely changed that. We know our structure according to the AI stack. So we start
07:31basically with a compute layer, a lot of hardware, obviously, deep tech involved there. Then we come
07:36with the intelligence layer. So the foundation model layer where we invest becomes the infrastructure
07:40layer, the DevOps tools, and so on, and the application layer. And this is the way we organize
07:46our team calls. This is the way we organize our internal approach to the market. And I think this is
07:52a big shift in the end, because we basically, I wouldn't say we throw away all the knowledge,
07:57the sector knowledge we have, but we add this AI basically, because it disrupts across all sectors.
08:04If we take the most extreme in food tech, for instance, we don't look at cultured meat or
08:08new proteins anymore. We look for how can basically AI help reduce lab times and or basically develop
08:15new proteins. So just one example. So this has changed. And this goes across all sectors. That's why
08:21we organize ourselves according to this, to this, to the AI stack. Does that resonate with you?
08:28Yeah, yeah. Pretty similar. I think on the, within the investment team, people are doing,
08:34you know, the core part of AI, what we call internally joking, like the frontier deals sometimes.
08:42Then the middle layer that has the beauty of being very agnostic of every industry,
08:47but also has the challenge of at some point maybe getting into sandwich between the low layer and
08:52the applicative one. And about the applicative one, sectors are still a bit important,
08:59because at some point, those companies are successful, not because of the speed of being
09:03AI native, but the real expertise, our data, uniqueness of data you've gathered for decades.
09:11But internally, we consider those companies are like the old SaaS companies. There is no tech mode,
09:16most of the time. It's the proper use of AI that makes the company able to reach the speed of
09:22100
09:22million revenue, like lovable or others. But at the end of the day, it's, it's KPI analysis. And
09:28that's it, because probably those companies won't be as sustainable as the others.
09:32Mm hmm. Clemo, could you just talk to us about what you feel with header and what,
09:39how your customers are using it?
09:42Yeah, sure. So it is a platform where investors can manage the entire firm in a single place. So
09:51we think that right now there is a very big problem into this industry. I mean,
09:57I'm talking about the financial industry in general, is first of all, people are using Excel too much,
10:03which is not good for anything. And the other problem is to be able to manage a firm,
10:08especially investment firm, you need to subscribe to a lot of different softwares.
10:13So you need at least one for your deal flow, one for your portfolio, one for your contact,
10:17et cetera, et cetera. And so the, the problem with that is, first of all, it's very hard to
10:22understand, you know, what's happening exactly into the firm. And I'm especially for very big
10:26firms, you know, with many different teams. For example, we're working with BPI, you know,
10:31we've got like more than 13 teams using the platform. So it's actually very hard to understand
10:34exactly, you know, what was going on. And the second thing is AI is, is not about magic,
10:40right? AI is all about data. And so the beauty of, of data is actually where you can centralize
10:45all your data in a single place. So all, you know, every companies that you've seen during the past
10:50years, all your investment decision, all your notes, all your, you know, the different investment
10:55that you've done, the price per shares, your co-investor, your cap tables, everything is centralized
10:59in the same place. And now what we can do is then we can use all these data sets to
11:04actually get
11:04really, really good insight with AI. Because the paradox for investors is usually they have
11:10a small team, but they need to manage a very large amount of data. I mean, I think you guys
11:14can,
11:15can definitely agree with that. And because you need to get the data of many different companies
11:19and to analyze that where actually it's the opposite for companies. Usually they get a big team and
11:24a bit less data. So we try to solve this paradox. And I think that's why I think AI will
11:29have a
11:29really, really big impact on the financial world, more than what we think, I think. Yeah.
11:35And VCs are just kind of one of the customer groups that you deal with. How, how big are they
11:43as a proportion out of your whole customer base? And is it growing? Is it growing at a rate that's
11:49quicker than other customers? I'd love to hear more about that.
11:51Yeah. So it's, it's actually a very, very interesting question because we built a SaaS
11:57tool that can be used by any investors. So we got micro VCFM with, you know, a couple of
12:02millions on the, on the management. And we also got very huge asset managers with over a hundred
12:08billions and they're all using the same platform, but then we develop some specific
12:11AI application for, for each of them. Right. Yeah. So, so, so, so that's, that's basically what we do.
12:20Yeah. Perfect. Um, and Hiroshi, from a CVC perspective, how, how are you sort of using AI in
12:30your workflows and where do you see it sort of having the biggest benefit?
12:35You mean in terms of investment, right? Yeah. So, well, AI help us to like, um, uh, shorten, uh,
12:43workflow, right? For investment. Like let's say, uh, we find any interesting startup, we don't decide
12:50to invest immediately. We always go find like similar companies or competitors and see, you know,
12:58or otherwise we, we may fail and not also, we also look into like a past deals or past the
13:04startups
13:04companies, right? If there were similar cases in five years ago, 10 years ago, we always have to
13:11find out why they failed. And, you know, AI really helped us to do that job for, for us, right?
13:17Uh,
13:18however, I mean, uh, other than that, uh, we also manually have to confirm with our business units.
13:26Like let's say we, Sony has the PlayStation, Sony music and Sony pictures, Crunchyroll, right? And all those,
13:32uh, uh, our business partners. Uh, but it's, it's, you know, AI cannot really help, uh, to find out
13:39if the startups product or solution really matches to the workflow, existing workflow we have, or if,
13:46if our creators really would like, love to adopt those. If not, if we fail, like, you know, I mean,
13:53we really don't want to evade, uh, like disrupting the, uh, existing creators position because everyone
14:00know, like a few years ago, there was a strike happening in Hollywood and we don't want that
14:04to be happening again. Well, I think most of the, uh, IP companies or entertainment company don't
14:09want to do that. But, uh, so yeah, going back to your question, it makes, uh, you know, our, uh,
14:14lead
14:15time workflow shorten, but others in that it's, it's, uh, quite as usual. I mean, we work for creators.
14:21Sure. Pauline, how are you using it sort of in your everyday to help you invest?
14:27Um, the primary usage is about deal flow sourcing. Um, so there are a lot of things,
14:34you know, scrapping the web, LinkedIn, detecting early signals and, and there, the complexities make
14:40things, make things that are on the shelf together with things you've developed internally and have
14:46all those stuff, you know, work properly all together. And something that is a bit more recent,
14:51but I think is also very useful is on, you know, monitoring companies port that helps decision
14:58making in portfolio management. Um, for instance, say how, how much time or how much money did it
15:04take to a successful or top quartile company of the portfolio in vintage number two to reach the 10
15:11million AR mark versus this one? What are the type of things that are, uh, real, you know, some strong
15:17milestones and validation. So you can really invest massively and aggressively into another round,
15:23uh, simulating front performance, according to decision making, when it comes to opportunity,
15:29opportunity of divesting a bit, a bit more or in all, uh, not all of those are AI driven, but,
15:37uh, but,
15:37but it's a mixture of, uh, uh, simulation on mathematical model together with a bit of AI on top.
15:44And are those tools that you've built in house and how long did that take? What was,
15:50what's the process behind that? I'm taking the examples of fund modeling that, that we have that
15:57helps on designing the funds, uh, in case the investment strategy from a generation, um, change
16:04to, to, to the other, um, then portfolio management and performance forecast. It started in Excel,
16:10actually, if we are fairly honest. Um, then Excel got to its limits, uh, and now it's, it's coded. Um,
16:18and it's been done by, by people within the team internally. Is that similar with you early bird?
16:25Um, yeah, that's similar. We have, um, a developer, um, basically working on this since
16:31six, seven years already. Um, currently we have five full-time developers working on this. And the,
16:37the basic idea was basically that we need support and finding out things happening. And since things
16:44have been accelerating in the last years, it's even more important. You can't do this with a,
16:48within, with a team of, uh, analysts in the end. And we started very early basically with
16:53machine learning approaches and so on. And then came AI that we use more and more AI. And
16:57in the end is what we try to do is basically try, try to, try to find patterns with entrepreneurs
17:02being successful. So that's something, um, I think we haven't found the final solution,
17:07but at least we're working on this. And, um, more, we have a lot of data, obviously,
17:11which we can use, but this is something what we want to do and we have worked on and, and
17:15we are
17:15getting better basically every, every other quarter with that. And we also use this, uh, to find also
17:21the best talent for our portfolio companies. In the end, it's about, um, finding the best talent.
17:27That's a war for talent in the end for, in the portfolio companies and they can succeed, can
17:33accelerate whatever, if they find better talent. And this is exactly the support we, we give them
17:38supported by, by our engine we have built in house. And, and that's, that's also something that,
17:43that can also become, well, is already quite, quite powerful in the end. And yet it's about, um,
17:51streamlining the process in terms of, um, identification of the right, um, either entrepreneurs
17:57or the right talent, which can add or can basically join our portfolio companies.
18:03And so is that tool sort of scraping LinkedIn and various other sort of.
18:07It's scraping all public sources, suppliers, subscribe sources, and then everything out
18:12there, basically registries, everything you can imagine social. So really everything out there.
18:17Interesting. What's your sort of pitch? Why should investors
18:23buy ADA rather than make their own in-house?
18:27That's the whole moment.
18:29Um, so, okay, I, I got five minutes to, to convince you guys. Um, no, man, um, you know,
18:36as, as you mentioned, so, so you, so you got a, you got a team of engineers, right? And actually,
18:40what is very interesting in, in your stories is, this is exactly how it has started. So I started my
18:45career at FG Labs, which is a very active firm in, in New York, and I couldn't find a tool
18:49that I wanted to use.
18:51And so that's why I decided to build it. And then after a couple of months, we had a team
18:56of seven engineers and, and I, I, I tell the partner that we got only two options, uh, in front
19:02of us is the first one is we keep it internally, but personally, I don't think it's a, it's a,
19:06it's a good idea. That's my, that's my position is because to build a very efficient platform,
19:11you usually need to have a lot of people. So some, some funds got the capability to do so,
19:15but that was not at least our objective and the way that we wanted to build the fund.
19:20And so the other option was to actually raise some money and then trying to build something
19:24that everyone wants to use. Um, and so I think, I think for a lot of, of VCs, it's kind
19:32of fine.
19:32So people can do it internally, but I got also some other examples about firm who tried to build
19:37the platform for five years and never succeed. So, I mean, of course, some, some VC we will do,
19:43but the thing is, it's much less true when you move to other class of assets,
19:47especially when you move to, for example, CVCs. So if you want to develop internally,
19:52uh, software for Sony that will fit with a level of security that Sony required, it's,
19:57it's going to not be possible. And then there is another thing, which is the integrations.
20:02So, I mean, you can, of course, lab code some stuff and you can do some integration,
20:06but then you need to maintain those integrations. You need to maintain the quality of the data.
20:10And to do so, you need, you need a large team. Um, so at least for now, uh, and even
20:15with a very
20:16new AI capability, because of course we also using AI to code, right? But the thing is, for example,
20:21in terms of security, um, we are working with a lot of banks and how can you convince your client
20:27that your platform is secure is you, if you don't even know what's inside the code. So right now,
20:32for example, we use, uh, we use AI to, uh, show to the client some new features in advance. So
20:38then
20:38we can show, okay, this is exactly how it's going to look like, but then to be able to integrate
20:43this
20:43into the security and the level of confidentiality and sovereignty that we have, uh, then we need our
20:49developers to rework the code, to read the code, to spend some time to understand every single aspect
20:53of the codes before to push it live. And even if we can prototype something in a couple of hours,
20:58then to be able to fit, you know, the requirements of the data sovereignty of a group like, you know,
21:04BPI or, um, or some others, uh, it can take months before to be able to, to make sure that
21:11everything
21:11is secure. Yeah. This is the week to be banging on the sovereignty drum. Um, Pauline, how has the
21:20sort of incorporation of AI tools changed, if at all, your hiring strategy?
21:27Actually not much. Um, because I would say it's something that enhances the transverse team when
21:34it comes to productivity, you know, workflow optimization. Um, and you just need people,
21:42you know, to be able to, to, to, to, to deal with those things and use them properly. So it's
21:48pretty
21:48easy on the front teams, to be fairly honest, I don't think much because one of the main criteria we
21:55have when it comes to investment people is the expertise in certain domain. And I don't really
22:01think that just drop a deck of a physical AI into chat GPT asking for a summary and the USP
22:08is the
22:08right way to go. Um, so, so to be, to be honest, not much for you, Christian. Um, well, yeah,
22:17I think
22:17to a certain extent, yes, Sean, it's true. This is the kind of the edification. I think there's
22:22something which is very much supported by that. And in the end, it's about, it's about our business
22:26is still a people business in the end, you know, so you have to have to, um, evaluate people. And,
22:31um, that's something which at least can be supported by the engine to a certain extent. But,
22:36uh, in the end, the final decision, it's about kind of, um, now it's, it's our business is a little
22:41bit like, like, like, like, like a marriage, you know, you're together for 10 years, you know,
22:45it's a long, long period of time, you need to get along quite well, you know, because there will be
22:48ups and downs in the journey. So that's why I think it's, it's very much a people, people decision
22:53in the end. If I can jump on that very quickly, I think it's a, it's a really good point
22:58that it's a
22:58people business. Um, and one day I had a meeting with a partner of a very big firm and he
23:04told me
23:04that, oh, you're gonna, you're gonna take the job of all of our analysts, you know, and I was like,
23:09actually, this is not what's going to happen because you, your analysts, you, you hire like
23:13very, very smart people, very brilliant people, very talented people. And now most of the time,
23:17they spend to read PDF and enter the data manually into a tool. Uh, and actually what they're gonna
23:23do is actually their job is they're gonna analyze these data for the first time and be able to
23:27understand what's going on. So, so you're right. It's, it's completely business people. And the point
23:31is now I think to use the most advanced technologies to be able to actually to let investors meet
23:37founders and, and take the human side of the deal. It's very, very different compared to,
23:43compared to PE. So PE I think can be, can be run almost by, by, by engines. You're right.
23:47You're true. Yeah. There is more data in PE as well. Yes.
23:51Turning our attention now to how the startups that you're backing are changing. Hiroshi, how have the
23:58type of companies that you're investing in changed over the past three years or so?
24:02Okay. Thank you. So again, I, my main focus is an entertainment investment. So in that perspective,
24:10the, you know, the startups, how is they changed? Yeah. Um, so let's say three, four years ago,
24:17many people thought that the AI will replace the creators, right? But the reality is now the market,
24:24again, you know, they strike happening, how it was, the short answer was no. So I see lots of,
24:29lots of new startup that is more focusing using a, uh, how to use AI to support the creators,
24:36right? I mean, they don't want to, uh, burn the, the, excuse me, uh, they believe the, uh,
24:44creators first. So it's great. And also they are more focusing on, uh, I would say like monetization
24:50or after the post creation and after the content creation, we have to protect your IP,
24:55AI to, you know, find out the IP infringements, help the older background works, collecting the
25:01royalties, which is still, everything is done manually in entertainment, almost in entertainment
25:05industry. So I see those changes a lot. Yeah. And earlier you touched on the sort of significant
25:12disruptions that the creative industry has felt by AI and the backlash that it's had has,
25:20how has the sort of way that your founders or founders that you meet talk about the technology,
25:26how, how is that changing? So it's my personal view, but when I talk to any like founders,
25:33creators, real entrepreneurs, they, most of them are using AI when they create the contents. I mean,
25:39but that is because of, you know, because of this inflation in the debt cost rise, uh, they cannot create
25:46contents in the, you know, the old typical budgets, right? They cannot make it. I mean,
25:51let's say gaming, you have to spend 500 million to create a single contents. It's too, too risky. So
25:57they have to save the, uh, labor costs. So people are using it, but, but, um, yeah, I think that's,
26:06I think nothing major has been changed because, you know, if you create AI, if we go ask chat,
26:11GPT, I think all the outputs are similar, right? Yeah. I think same thing is happening in the creative, uh,
26:17industries. So people still have to modify the output came out from AI. So ultimately I think it's not,
26:24yes, it's giving an impact, but the fundamentally the core thing has not been changed yet.
26:30Pauline, we spoke briefly about sort of vibe coding companies like Lovable.
26:35Uh, how, what does the dynamic of people being able to vibe code, whether it's sort of products,
26:45how does that kind of change your role as an investor?
26:49We, we have a different eye when it comes to more mature companies in vintages of seven or eight
26:56years old that had AI already, but the previous generation, and the other question is, uh, defensibility
27:03of their positioning. Um, and there, there are a couple of good or bad answers. Uh, a lot of people
27:11talk about the ownership of data, which we think is not the right word is the unicity of data. That
27:17is more important. Um, things that, uh, you know, are too much integrated to be unplugged or things for
27:25which, um, the best probability of is not a good answer enough. So most of the regulated, uh, businesses or
27:32industries. So out of those, then we have different moves and we made some proactive
27:39decision of early MNAs and company that we thought wouldn't be defensible enough, uh, in two to three
27:45years from now, um, companies in HR tech or some other stuff for which AI is a short term, massive
27:55disruptor. And when it comes to new investments, um, I mean, probably the same for you, all of the
28:02companies we finance now are a native, um, at different layers, uh, and we don't look for the
28:08same type of differentiation. When it's core, you need a truly differentiated IEP that you think
28:15is sustainable over three, four, five years. When it comes to things that are more of execution,
28:21investment thesis that then you want people to be well financed to, to, to, to have this mile in
28:29advance to become a category leader. So the mindset and the stages at which we invest differ a lot, uh,
28:36depending on how AI is the core of what they do or what they use, uh, and the approach of,
28:43of, um, investing
28:45more or less capital is not the same as well. Yeah. For, for us, adding to this, I think what,
28:50what has, I think has changed quite a bit, um, for us, it's about cutting through the noise in the
28:54end
28:54because there's so much, so much noise out there and it has also, it's also changed. So when we have
29:01been investing, I would say in typical SaaS companies, you could invest in almost any point
29:05in time. You either very early, say you did a bit, but very early or a little bit later or
29:10later.
29:10Now in AI times, you hardly can invest in the end because companies scale so quickly. You mentioned
29:16the level ones. If you don't invest in the one quarter, you cannot invest in the following
29:21quarters because you're totally out, uh, basically they are totally out of range. So that, and that
29:26has changed to a vast extent basically. So this is exactly why we have to be faster in the end
29:31in
29:31decision-making, uh, that, that has certainly accelerated. Without the FOMO. Yes, exactly.
29:36I would invest in the end and deal with the FOMO in the end. And you just mentioned Lovable. I'm
29:44pretty sure it announced it had sort of hit 500 million dollars in ARR today. What do you guys think
29:50about ARR as a metric? You mean for evaluation? Well, I think it, it, it's, it's, it's very valid.
30:02I would say KPI still, it depends a little bit on the business model in terms of how much money
30:07are
30:07they, are they, what's how the, how do the margins look like? So how much money are they burning and,
30:13and spending on compute? So that's, I think a very decisive factor in that respect. So you can't
30:18just look at, in SaaS, you would just look at it because you know, the margins are good. You
30:22would just, just look at ARR. I think here's a little bit different. So the ARR is a little bit
30:26different compared to the typical old SaaS KPI. There's also a M&A market right now that is
30:34completely distorted between companies that are revenue making and the buyer buys a P&L, a position
30:42in the market. A, I don't know, coverage of a geography doesn't have and IP preempted companies.
30:49Uh, prior labs was acquired a couple of weeks ago for a billion, uh, out of barely no revenue.
30:56Uh, but when, when we take the, the track record of what we did in the past, those are,
31:03you know, very exceptional deals that, that, that is not a playbook for AVC. Investing in IP,
31:08praying for someone, putting a billion on the table, uh, and make a fun return out of there.
31:13It's great. And we do deep tech, but this is not the main playbook. At the end of the day,
31:18historically, the biggest exits we've made are P&Ls in hundreds of millions.
31:24In addition, what, what is different here in this ecosystem is that basically the growth capital is
31:29missing to a last extent. Obviously some companies, they will get the growth capital and the better ones
31:33will get this mainly from the US, but in this ecosystem, this is missing specifically when it
31:38comes to, comes to deep tech, this is, this is missing. So you always have to think, okay,
31:42who could be the next investor in the next, um, round? Um, whereas in the US it's much easier
31:49because much more liquid specifically in the, in the, in the growth phase.
31:54Is that something that you see between customers?
31:58I wouldn't, I wouldn't do any comments on that.
32:01No, no, I'm kidding. Uh, no, I think, I think the, the IR of course is, is still, uh, I
32:05mean,
32:06this is, I mean, when you, sometimes, you know, we forget, especially because there is a lot of
32:09fundraising announcement and those kinds of things, but sometimes people forget that actually if you
32:14build a company to make money and, and the real way to make money is your IR. Um, and so
32:18we shouldn't
32:19forget that the revenue must be, I think, um, one of the first metrics. And then of course you
32:23mentioned the different margin. Um, so the margin will be different between AI application and
32:28classic SaaS, and it's not, it's not, it's not that, that easy, but I think there is also some
32:32other component that play a big role. For example, you know, the duration of the contract that you
32:36have, the average size of the contract, you know, if your clients are loyal or not, and all those kind
32:41of different metrics that will adapt to be devaluation, even with AI, they will, they will be the same
32:46across, you know, your churn rate, even if you've got a AI native company, if you've got a high
32:51chain rate, uh, it's still, it's still bad. So, so yeah, so I think AI is still valid.
32:55And you mentioned on the prep call, how you can sort of prototype a new product feature in a couple
33:01of days, but then it might actually take months to ship because you're sort of sorting out compliance,
33:08data security issues. Um, what would you like to change about that? What, like, what value do you see
33:15in sort of the second stage of that? And what, what would you like to see change?
33:19Well, I think the value is very clear. Um, it's, I think it's, I think it's honestly very,
33:24um, it's absolutely amazing. And sometimes, you know, you can have some clients that are coming
33:27from, uh, you know, with, uh, you know, use case. And so it's not necessarily, you know,
33:32something that you've heard about it before, or even if, you know, we got right now over 200 clients,
33:37uh, into 40 different countries, but we still have sometimes some people coming with some new use
33:42cases and, and we're like, Oh, that could be very interesting. And now we are able to show to the
33:46clients how it's going to look like exactly in like a couple of hours. And so for most of the
33:51clients,
33:52it's, it's so amazing that you, you ask something and, you know, and a couple of hours is, is available.
33:57So I think it's great. Um, but, but I think honestly, um, you know, we, yeah, we, we also, of
34:06course,
34:06using cloud code and everything to do so, but the difference between this and being able to
34:11really use the platform and making sure that the data is going correctly into the right place.
34:16And, and there is no bugs and it's fitting with the rest of the tech stack. I don't think this
34:20is something we can compress a lot. Uh, of course, if you make AI much smarter, you know, you will
34:24have less problems and less things, but for example, on the last prototype that we did, some part was
34:28very nice and working perfectly, but some part were actually completely broken and, and we never work
34:33with lab coding. So then you need to get your senior engineers that will rewrite a part of the code.
34:38Um, yeah, so we definitely moving sometimes, you know, from six months to develop a new feature
34:43to three or four months, which is already great. Um, but being able to do everything, uh, by a safe,
34:49secure way and, and everything in a couple of days, I'm not sure it's going to happen, uh, anytime soon.
34:55When reporting, I've heard from some investors who I'm sure aren't yours, but as sort of how their job
35:04is changing as a result of portfolio companies is kind of so excited by the ability to vibe code new
35:10features and, you know, create something completely new and add onto their business,
35:17you know, every few days. And they're sort of saying that the guidance that they're giving to
35:22founders has become more about prioritization and kind of keeping track of the main thing. Okay,
35:30and focus just, just a, on a funny note, no later than yesterday, one of the CEO of our portfolio
35:37company at the board meeting say, Oh, AJ, the company is AI native, uh, operating in, in climate
35:44tech. And he said, I have too many things to deal with right now. So I built an agent and
35:49then another
35:50agent, uh, to do this and then another one. And now I have agents to manage my agents, but then
35:55I need
35:56another agent that is a chief of staff of the other agents. It's about distraction at some point.
36:03Have you had similar experiences, Christian?
36:06Yeah, well, that's indeed a very good point. I think focus matters a lot in the end.
36:10It's good for every business. I mean, this is, this is also in AI times, I would say there's no,
36:15no big difference in that, that respect.
36:19Perfect. Uh, how, how does your business use AI internally? Where have you seen the biggest wins?
36:25Well, I think this is, uh, what I already explained is actually quite clear, you know,
36:30the, the ability to, to build something very, very quickly. Um, then on a personal level,
36:35I try to avoid using it as much as possible. Um, cause I, I think sometimes we just delegated
36:40too many things to, to, to AI and I try to keep my brain lively, you know? Um, but I
36:47mean,
36:47it's one more time. I think, I think if you need to demonstrate some capabilities or if you,
36:51if you want to, you know, showcase something, I think it's, I think it's an amazing tool and I
36:55think we should keep continuing using it. Uh, then of course we also, I mean, AI, I mean,
37:00then if, if I dig into more specific use case into the engineering team, uh, for example, for QA,
37:05it's amazing. So if you want to test, uh, to run some tests before to launch a feature or before
37:11to,
37:11to push in production, I think, I think AI is super strong for that. Uh, we, we use automation, uh,
37:17since very long time, but AI is able to go much more into the details and, and track some,
37:22you know, possibly, possible bugs, uh, way in advance or before or something that are not necessarily
37:28clear from the developers. So I think that's the, that's the main use cases that we have right now
37:33internally. Amazing. And we're nearly out of time. So I'll just ask you all one quick fire question to
37:39end on. What is the most underappreciated quality in a founder stroke investor? Um,
37:47so starting with you, Hiroshima, what's the most underappreciated quality in a founder?
37:58Uh, just what's the best quality in a founder?
38:01Uh, not about the passion, but you know, when I talk to, uh, founders like, you know,
38:08many, most of them recently use AI and try to come back with you, you know, just, uh, more comfortable
38:14answer or pitch deck, but it's almost the same. So when I talk to the founder, if those founder has
38:20a
38:20clear view of like, what is the industrial pain points are and what the, uh, the, uh, you know, the,
38:26uh,
38:26great answers at the solution that the industry may accept, if the, those guys have a clear vision to it.
38:33I mean, I, I, I believe it. I believe if not, if not.
38:37Perfect. Christian. Um, well, for me, it's the ability to, to sell because you need to sell your company all
38:44the
38:44time, not just the product, but the company. You do this for the C round, you do this for the
38:48A round, B round,
38:49whatever. You do this in front of public investors when you go IPO. So you always have to basically be
38:54able
38:54to sell this sales, sales quality. It's not everybody has this, but it has to be trained.
38:59I think that's the most important or very, very important feature what we're looking for.
39:03Pauline. Um, so this one has been mentioned. Another one would be how much they are hungry,
39:09um, hungry to solve a problem really day, day and night, hungry to make money, hungry, but hungry to
39:18something. We, we, we see, I think right now a bit too many entrepreneurs being entrepreneurs
39:23by lifestyle, um, more than, you know, the true reasons, I think so. Yeah. Ambition and
39:31angry. Um, maybe if I can add something, um, I think it's the ability to manage a, um, emotional
39:41rollercoaster. Um, so for example, if I can just give you a very quick example is, uh, you know, we,
39:47we raise our, our first seed round and that was such a big thing. You know, I spend a lot
39:51of time,
39:52you know, working every day, doing the speech extra. It was a lot of, a lot of work and we
39:57raise this amount of money and then everything was on Silicon Valley bank. And then during one
40:01weekend, all the funding disappeared. And that was the worst weekend of my life. And then on Monday,
40:06everything comes back and then you got your funding again, but it's just one example. But you know,
40:10during the past year, we went through the COVID, we went through like the bank crashings and so many
40:14things. And I think the ability to stay constant and to keep continue believing your visions and keep
40:20continue trying and trying to prove that everyone is wrong. And you're right. I think,
40:23I think this is something that, uh, no, no matter what happened outside,
40:25I think it's something that is really important. Keep calm and carry on. Okay.
40:29That is a great point to end on. Thank you everyone. Thank you very much. Thank you.
Comments