20VC: Ash Fontana on The 5 Core Characteristics That Make Data Valuable, What VCs Can Learn From Italian Craftsmen and Howard Marks & The Importance of Vertical Integration In Scaling Today

Ash Fontana is a Managing Director @ Zetta Venture Partners, the fund that invests in AI-first companies with B2B business models. As for Ash, prior to Zetta he started the money side of AngelList, where, he launched online investing, created the first startup ‘index fund’. He also ran special projects like AngelList’s expansion into Europe and the UK. Simultaneously, Ash led syndicates and made investments in Canva, Mixmax and others. Before AngelList, Ash co-founded Topguest, a Founders Fund-backed company that built customer analytics technology and was ultimately sold in an 8 figure transaction 18 months after the company was founded. In Today’s Episode You Will Learn: 1.) How Ash made his way into the world of venture with AngelList and how that led to his joining Zetta today, investing exclusively in AI? What did Ash's time working on his family farm teach him about vertically integrated businesses? What were his biggest takeaways from AngelList and working alongside Naval? 2.) What does AI-first really mean to Ash? How crucial is it for companies to have proprietary datasets today? Are data moats truly defensible and real? What are the 5 characteristics that determine the level of defensibility of a dataset? How does Ash analyse the quality of a dataset? What does Ash do to determine if they are predictive of value? 3.) We often hear the term, "system of record", why is Ash so much more excited by the "system of intelligence"? Why is the basis of competitive advantage shifting from SaaS today as a model? How do the margin structure vastly differ when comparing AI-first companies to SaaS companies? How does that mean one should view capital efficiency? 4.) What does Ash believe drives business model quality? What are the commonalities in the business models of those that have made it big? Why does Ash believe it is difficult for incumbent companies to become AI-first? How difficult is it for incumbents to acquire smaller AI-first firms and integrate their policies and technology? 5.) Why does Ash love Howard marks and what has been his biggest learnings from studying him? How has Ash applied these learnings to his investing today? What has Ash also learned from the Italian masters of design? How has this study helped Ash as a VC? What has Ash optimised lately? What is Ash's favourite optimisation? Items Mentioned In Today’s Show: Ash’s Fave Book: The Strategy of Life: Teleology and Mechanics in Nineteenth-Century German Biology As always you can follow Harry, The Twenty Minute VC and Ash on Twitter here!

Speaker names marked ? are inferred from speaking time and the show’s known hosts — they are not voice-verified.

• 0:00Harry Stebbings?

  So people often ask, how do we go so deep with our guests in such a short space of time? And the truth is, research. We research a lot for every guest and speak to many, many people around them. This episode is a result of eight conversations with eight different people close to this guest. And so with that, I'm super excited to welcome our guest today. It sounds super strange, but I've been a long time fan of how this guest really thinks and invests. And so with that, I'm thrilled to welcome Ash Fontana, Managing Director at Zetta Venture Partners, the fund that invests in AI-first companies with B2B business models. As for Ash, prior to Zeta, he started the money side of Angel List, where he launched online investing, created the first startup index fund, and also ran special projects like Angel List's expansion into Europe and the UK.

• 0:41Harry Stebbings?

  At the same time, Ash led syndicates and made investments in the likes of Canva, Mixmax and others. And before Angel List, Ash co-founded TopGuest, a founders-funded back company that built customer analytics technology and was ultimately sold in an eight-figure transaction just 18 months after the company was founded. And I do also want to say a huge thank you to those eight incredible people for their help with the schedule today. So a huge thanks to Annie McLaughlin, Zavain Dar, Leo Polovets, Jocelyn Goldfine, Kyle Doherty, Greg Sands, Alex McCourt and Eric Worshing for some fantastic question suggestions today. But before we dive into the show today, I'm sure you've heard about it, but my word, this is a product I love, Carter. Carter simplifies how startups and investors manage equity, track cap tables and get valuations.

• 1:23Harry Stebbings?

  Go to Carter.com/20vc to get 10% off and more than 800,000 employees and shareholders. Use Carter to manage hundreds of billions of dollars of equity. And Carter now offers fund admin, so you can see real-time data in the Carter platform and work with Carter's team of experienced fund accountants also in the background. Go to Carter.com/20vc to get 10% off. And speaking of companies that have taken our ecosystem by storm, I have to talk about Brex, the company which built the corporate card for startups and who is the fastest company to reach unicorn status in history. Brex founders Henry Fian Pedro built a payments business in Brazil, but kept getting rejected for a corporate card in the States. So they built Brex with no personal liability, up to 20 times higher card limits

• 2:06Harry Stebbings?

  and huge rewards like seven X points on Uber and Lyft, four X on Brex Travel and 2X on SaaS software. If you're a VC backed startup based in the US, see if you qualify for a Brex card today at brex.com and get card fees waived for life by entering the code HARRY during signup. And finally, you have to check out Rank Science. Rank Science is the easiest way to grow organic traffic and get your content ranking higher in Google. Their AI-powered platform uses real-time search data and natural language processing to help you craft content that is highly relevant to search intent and maps directly to what users are searching for. For customers with lots of landing pages, you can leverage Rank Science's SEO/AB testing platform to easily test which HTML changes cause an uplift in search traffic. Also, as an added bonus,

• 2:48Harry Stebbings?

  their platform also allows marketers to make SEO enhancements to websites without eating up precious engineering resources. Rank Science is used by hundreds of companies to grow organic traffic, from startups to publicly traded companies, including the likes of BuildZoom, Goldbelly, Sweetness and Career Karma. Go to rankscience.com/20vc for your first month free. But that is quite enough for me, so I'm very excited to hand over to Ash Fontana, Managing Director at Zeta Venture Partners.

• 3:20Speaker 01

  You have now arrived at your destination.

• 3:23Harry Stebbings?

  Ash, my word, it is such a pleasure to have you on the show today. I mean, I feel like I know you so well, having heard so much from Tyler at GC, Alex at Clearbit and so many more. So thank you so much for joining me today, Ash.

• 3:34Speaker 02

  Thank you very much, Harry. I'm just happy to be in your credible orbit of positivity.

• 3:38Harry Stebbings?

  Oh, you are too kind. There's a huge amount of espressos and espresso martinis that drive it, that I would like to kick off today with a little bit on you. So tell me, my friend, how did you make your way from the wonderful island of Australia to the world of venture in the valley?

• 3:52Speaker 02

  Yeah, I guess as a kid, I liked doing two things. I liked pulling apart computers and pulling apart companies. And my bedroom was strewn with all sorts of computer parts and dinner table conversations at home were littered with accounting terms, weirdly enough. And I never really knew what I wanted to do with these two interests because they seemed so unrelated. I just picked up a magazine one day and read about a thing called venture capital and really had no contacts in the industry or anything like that. I just set my sights on the US and... Found my way here at every opportunity, first for travel and then for study and then for love and then I eventually settled in New York and took it from there.

• 4:28Harry Stebbings?

  Now, I did hear from Carl at GC that you spent time in your family business in Australia. So it's a super interesting one because it's a different field. How do you think that time and experience impacted? How do you think as an operator and investor today in terms of mentality?

• 4:41Speaker 02

  Well, I was indeed a farmhand in another life and had macadamia in that breakfast this morning. I think the big thing I learned from how we run that business is the importance of vertical integration and that is controlling every single part of the production process. And we do everything from genetic engineering to build our own satellite imaging software, right through to consumer branding. Controlling every part of that process allows you to capture all the value you create by improving each sub-process that's part of any of those sort of activities. And I think that has certainly paid off for us over time and I've learned how to apply that to other businesses too.

• 5:16Harry Stebbings?

  Totally with you, Dan. It reminds me of a conversation I had with Keisha Boy where he said exactly the same. The second experience though that I do want to touch on is your time with Angel List. And Andy McLaughlin asked this one specifically. He said, what did you learn from your time at Angel List? And how have you applied that to your venture investing with Zeta?

• 5:33Speaker 02

  Well, I mean firstly, I just learned a lot about investing in startups from the Val. He has hundreds of hard earned heuristics about what it takes to make a successful startup. Secondly, I guess I learned the importance of finding signals before anyone else through semi automated, high volume screening, which is obviously something we had to do at Angel List because there are millions of companies on the platform and you've got to sort of sort through them to help the investors find great opportunities. But that's something we also do at Zetta today in our own way. And at Angel List, we found companies like Uber, Zymogen and Open Door before anyone else or before many others. So I think finally, I just learned the nuts and bolts of how to run a venture fund as a business at scale. And to enable syndicates, we set up vehicles there that now manage billions of dollars

• 6:18Speaker 02

  of investors' capital, but importantly, at a very low cost to those investors with a really high degree of legal compliance. So all that structuring was really important too and obviously important today as a setup.

• 6:30Harry Stebbings?

  And I didn't expect you off schedule this early, but you said there about kind of signal discovery. And I hear a lot of funds today talk about algorithmic sourcing and how they intentionally use technology and AI to discover stuff. Anyone else largely to me is really using LinkedIn Sales Navigator effectively. What do you think about this next gen of AI driven sourcing venture?

• 6:50Speaker 02

  I think like a lot of what people call AI, it's more about augmentation than automation. And it's a little bit of automation and that is, I think you can use a lot of data sources to help you figure out where to look and who to talk to. The programmatic analysis of early stage companies, I think, is a full zeron. And I have seen some of the best data in the world at Angel List. I've talked to all sorts of people experimenting with that data when I was there and since then. And it doesn't really work. There's nothing to really systematize there, I guess, to use a word that's not a word. But I think it can be very, very helpful if you have many, many different data sets to sort of point you in the right direction. And yeah, using LinkedIn Sales Navigator is fine, but at Zeta, we have about fifty something data streams that we bring in.

• 7:34Speaker 02

  And then we use humans to annotate them. And then we have some human screens, then we have some automatic screens. And anyways, it's a multi-step process that goes all the way through to finding companies. And, you know, we're finding a thousand companies a month through this system. And then taking a hundred of those through diligence and bringing one term sheet stage every month or two.

• 7:53Harry Stebbings?

  Holy shit.

• 7:54Speaker 02

  We're a tiny firm, right? We're only five investors. But so we have to find ways to find companies at a decent rate without using too many people.

• 8:02Harry Stebbings?

  That is incredible. No, absolutely. An incredible discovery process there. But you mentioned the kind of AI and augmentation element. And I want to start today on the field that you invest largely in being the world of artificial intelligence and data. Now, I'm sure a term we both see everywhere is AI first. Now bluntly, Ash, what does AI first really mean today?

• 8:20Speaker 02

  Well, I'll take full blame and responsibility for introducing this piece of jargon. I put it in a, in a little blog post, I think in like twenty sixteen or something, and it seems to have caught on. I guess, look, if you think about traditional software companies, they focus on building product features, getting qualitative feedback from human customers and then manually iterating on those. Product features and you have great product managers and very high empathy engineers to do that really well. AI first companies, on the other hand, they focus on building predictive models, not product features, getting quantitative feedback, not qualitative feedback, and from machine agents, from how the model's just running itself rather than human customers, and then automatically iterating on those predictive features.

• 9:02Speaker 02

  So I think that's sort of the difference if you think about a traditional software company versus an AI first company. And an AI first company, you know, every strategic decision, every product, sales team, marketing, capital allocation decision, relates to improving the predictive model rather than just adding another feature. And AI first companies, I think, can beat incumbent software companies because the incumbents started in an era where customers were afraid of putting their data in the cloud or afraid of giving up their data. And so if you're a company that's not an AI first company that started sort of before this AI era, you can't really build any predictive models of consequence because you don't have any of that customer data and you have to sort of go back to your customers, renegotiate the contracts, change the way you build your product, collect more data,

• 9:45Speaker 02

  and hire a whole new teams of machine learning engineers to sort of rejig yourself for this AI era. And our first company, on the other hand, just starts that way.

• 9:52Harry Stebbings?

  Can I ask, how difficult is it for the incumbents to acquire AI first companies and then integrate? And then they kind of skip that challenging transition?

• 10:00Speaker 02

  That's a really good question. I think often it's actually relatively straightforward. Because if you think about it, an AI first company has to build the models, the predictive models. They also have to build the product and then they have to build a way to deliver that product through services or other sort of integration methods. And incumbent companies usually have the product part down pat and maybe the delivery or the services part down pat, but they don't have the model part. So it's all part of the same chain in the end. At the end of the day, you have to give the customer the full solution. And incumbent companies have at least part of that solution. They just don't have the really valuable part, which is the predictive part, the part that sort of helps you make a decision, not just calculate something for you. So we've seen incumbent companies very successfully acquire AI first companies

• 10:43Speaker 02

  if they spot the right one that fits nicely into their product.

• 10:46Harry Stebbings?

  Can I ask, in terms of the data side, if we go back to the early stages of? Evaluating AI-first companies or open banking in the UK has meant that we've seen real democratization of financial data. My question to you is, for me analyzing fintech companies and broadly AI-first companies, how crucial is it for the company to have proprietary data sets?

• 11:04Speaker 02

  Yeah, you know, I don't think it's crucial to have a proprietary data set. It's just mostly the case that you need it. You need proprietary data to build a model that generates results that are far better than what someone else can generate by throwing commodity data into, you know, an openly available model. Lots of research published on archive every day, lots of open source tools and whatever else. So, you know, if the playing field is level in that regard and then the data is the same, then you'll probably end up with a very similar result. But if the playing field is level on the model side, but the data is proprietary, you might get a different and sometimes better result. So I think data modes are indeed real. And we evaluate those modes very carefully at the diligence stage. We look at things like accessibility, fungibility, dimensionality, breadth and perishability.

• 11:49Speaker 02

  What we mean by that is basically how hard is it to get? Can it be substituted by similar data that's easier to get? Can you use it to feed a model and generate some sort of real outcome? Do you have enough of it to represent reality? Is it broad enough? And then is it fresh enough to be close to reality? And so we sort of look at that on a qualitative basis, but also have different ways to quantify that. And that's how we evaluate the data mode.

• 12:11Harry Stebbings?

  Can I ask, in terms of the perishability, in some ways it's brilliant because it creates real defensibility. But then in other ways, it's also difficult because you lose that defensibility with every kind of refresh. How do you think about whether it's good to have high perishability or low perishability? What's the answer there?

• 12:27Speaker 02

  Oh yeah, very good question. It really depends on what you're trying to model, right? Like if you're trying to model financial markets, which are changing every microsecond, you have to work with perishable data. You don't really have a choice because the reality is very worked every microsecond. However, if you're trying to model something different, like a complex system, like you're trying to develop a fundamental understanding of like how a disease state develops, you know, that's not going to completely change over a very short period of time because those diseases have evolved to develop in a certain way over a long period of time, over hundreds of years or something like that. So it's okay to have sort of older patient data if you're trying to do that. So it really depends on what you're trying to model. And sometimes perishability is really important and sometimes it's just not important at all.

• 13:10Harry Stebbings?

  Can I ask also, you said fungibility. What does fungibility of data mean? And what does that look like in the good case?

• 13:16Speaker 02

  Ah, yeah. So it means like, can it be replaced by something very similar and achieve essentially the same result? So, you know, you could spend all of your time trying to get this data set that's really hard to get on customer preferences, like proprietary survey data or something like that, and spend a lot of money on that. And then someone could come along and just go, well, we can also just derive those preferences based on like, um, what people are clicking on, or there's actually a free data set that's not exactly the same, but tells us the same thing about what these customers want. So fungibility is just very similar data going to get you to the same outcome as you feed it through a machine learning model.

• 13:50Harry Stebbings?

  Totally. That must be a real oh shit moment when you realize someone else.

• 13:54Speaker 02

  You've got to get ahead of it.

• 13:55Harry Stebbings?

  Tell me, that is, if that's the buzzword of kind of AI first, the other buzzword of the day that I love when investors say is, you know, it's a system of record because everything's a system of record today. But you've spoken before about systems of intelligence. What does that mean to you?

• 14:08Speaker 02

  I remember writing about this in, um, twenty sixteen or something. So I guess the distinction is a system of record gets data in a structured form as it's manually entered by a human sort of after the fact in like instruction field. So like a CRM or an ERP or something like that. A system of intelligence gets data usually in unstructured form and it's automatically collected by machines as data exhausts as the product is used. So the distinction being structured versus unstructured, manually entered versus automatically entered, and then post facto or after the fact as, as opposed to in real time as it's just exhausted or comes out of the product. David Wilkinson, he's an entrepreneur, recently back started a company called Hash.

• 14:50Speaker 02

  He said to me, the system of record owns your assets and records what your assets are doing. But the system of intelligence owns your strategy, as in it actually is part of your decision making process. So a system of record just sort of really is just a database, um, with a UI on it. A system of intelligence is something that is, you know, a core part of your decision making process. And make suggestions to you as you're trying to make a decision.

• 15:12Harry Stebbings?

  I love that differentiation from David. I do have to ask, when we think about this kind of parlay or entrance of system of intelligence and we apply that in business models to it. And do we apply the same SaaS model that we have before? Because I know that, and I've heard you say before about the basis of competitive advantage shifting from SaaS today as a model. How do you think about the business model aligned to system of intelligence and why the SaaS model today may be deteriorating?

• 15:36Speaker 02

  So business models between SaaS companies and AI first companies are very different. Because you know, with a SaaS company, you're sort of pitching based on a bit of a workflow improvement. We can make this process a little bit faster because you know, we're calculating stuff a little bit faster for you. And these features sort of automate something really nicely. And so the customer thinks of it in terms of a bit more of a cost saving. An AI first company tends to pitch based on earning someone revenue or helping you make a decision you couldn't otherwise make. So if you're able to make this prediction, you'll be able to increase yield on your production line quite a lot. And AI first companies tend to have a little bit more of a shared benefit model or a percentage of the increase in revenue model. So I think that's how things differ a bit on the business model side.

• 16:18Speaker 02

  On the competitive advantage side, I mean, if you think about a customer choosing between two SaaS products, they tend to choose one over the other based on features or integrations. And look, these days in the age where you know, a seven year old can build an app, features are relatively easy to copy. And having more integrations than your competitor is just a bit of a rat race. Intelligence systems, on the other hand, they predict something, they gather more data as they get feedback on those predictions from customers. That data then improves the model, which makes it more attractive to customers, which gets you even more data and so on. And we call this the virtuous loop in actually our very first blog post under the name of Zeta. And this loop runs faster and faster every time and it creates sort of an unassailable advantage over competitors.

• 16:58Speaker 02

  So yes, in summary, in SaaS, you're competing on features and integrations and they're easy to copy and it's just a rat race. And in AI first companies, you can get this runaway advantage very quickly because the system compounds in value itself. Like without it, without any effort required beyond the initial effort to build the model.

• 17:16Harry Stebbings?

  Can I ask, in terms of that kind of runaway in value, you guys are going. Just pre that kind of value inflation.

• 17:22Speaker 02

  Yeah.

• 17:22Harry Stebbings?

  How do you think about deciphering from signals? When is the moment just pre value inflation point? What does that look like in terms of model activity, flexibility?

• 17:31Speaker 02

  Yeah, absolutely. So, look, no startup is perfect from day one. So what we try to do is help them figure out how to get from where they are today to the point where they're generating a lot of value for customers. And so, you know, this starts with looking at the minimum accuracy required to exceed human performance, because once you've exceeded human performance or something like that, you're going to be able to sort of save something for a customer. And then we look at, okay, once you get to that minimum accuracy, how stable is the model as you throw different data at it? And then what's the fundamental payoff that this prediction will yield for the customer? Like, is it less errors on their production line? Is it tighter inventory management? Is it something like that? And, you know, along the way, we're looking at the usual metrics for machine learning models, such as the F1 score.

• 18:15Speaker 02

  And then, once we've sort of figured out where the startup is today, by looking in detail at their machine learning experiments that they've run. And look, they could be running these experiments and not even have a product to deliver this yet. They could be running the machine learning model and delivering the results to customers in like an Excel sheet or a PowerPoint presentation. But once we've looked at these machine learning experiments, we try to help them figure out what needs to be done to get them from where they are today to the point where they're generating results that are really meaningful to the customers. So if it's a low accuracy model, they probably need to do a bit more feature engineering and figure out different predictive features that can get them to a higher accuracy. Or if the model's not performing very well, then maybe they just need to throw more data at it. Or if the model's not very stable, maybe they just need to run it over a longer period of time.

• 18:58Speaker 02

  And if the model's just not producing a prediction that's not very valuable to customers, then they just need to spend more time with customers figuring out exactly what prediction they want. So that's how we sort of work through it with startups and figure out where are you today? And how close are you to delivering that customer value? And there's no other way to do it than to dig into the experiments they're actually running.

• 19:15Harry Stebbings?

  So if that's kind of perfecting the accuracy around the AI model itself, in terms of the business model itself, I do want to touch on this because before we dive into the really nitty gritty on business models, I spoke to Carla General before, and he said that you have a unique sense of business model quality. And can articulate frame out why companies get big and word for words, that's a great quote there. What do you think he meant here? And can you unpack your thinking on what drives business model quality?

• 19:40Speaker 02

  Yeah, I think Kyle's talking about my obsession with both compounding competitive advantage through these virtuous loops. And then also my obsession with vertical integration. So I sort of talked about that virtuous loop concept before, but I'll talk about vertical integration now. And I just believe the best products in the world are made by vertically integrated businesses. Apple vertically integrates from hardware to software, Amazon vertically integrates from warehouses to websites. And going all the way back to like Andrew Carnegie vertically integrated from mines to his own steel mills. And for an AI first company, this means doing everything from collecting your own proprietary data, building your own models,

• 20:18Speaker 02

  delivering those predictions to customers through your own products, then offering your own services to integrate with a customer's business process. And this is the hard path, you know, it's easier to just buy a bit of data or use an open source model or deliver it through some other third party product or platform or ecosystem. But doing this all yourself vertically integrating is the only path that lets you capture all the value created. And you know, this is what I've seen on the farming business that we run, but it's also what I see every day with AI first companies as well.

• 20:47Harry Stebbings?

  Speaking of value capture there, it kind of aligns to one of my biggest fashion points in the topics to discuss, which is market size. I always say honesty is kind of BS in the very, very early days. Do you think TAM is total BS in the early days?

• 21:00Speaker 02

  Yeah, I go back to something a partner at Benchmark said once, which is what can go right? And that is what is a company doing differently that could be valuable to a lot of people? And I start actually from bottom up. So I don't do these sort of top down TAM models. When we're making an investment, I have this little tool I've built. And what it does is it looks at a company's product roadmap. It values the functionality of each feature on that roadmap according to customer ROI. So what we hear from customers like this feature will deliver this ROI to me. And then it generates an annual revenue per user based on that ROI. So if you're saving a customer a hundred dollars, then you can probably charge ten dollars. It multiplies by that by the number of potential customers.

• 21:40Speaker 02

  In the world, which we get from various market research papers. So then I've got that average revenue per user times number of potential users in the world. And then I apply a margin. So what's the business's margin to figure out how the revenue flows to the bottom line? And then I apply a valuation multiple from comparable public companies. And so then you've got sort of like an estimate of market cap. Now, of course, it's a long way between where the startup is today and getting to that point. So I apply a very extreme discount rate that gets us all the way back to today. And like takes into account all the execution risk there. And then it spits out a number. And you know, if that number, which is the potential market cap of the company discounted to today is bigger than the size of our fund, then that's the starting point for thinking about making investments.

• 22:20Speaker 02

  So building this little tool that goes right from the product roadmap through to the potential market cap of the company, of course, with a lot of assumptions along the way, is how I do market size, very much bottom up. I don't do this sort of top down stuff and I'm with you. I just don't find it that useful.

• 22:34Harry Stebbings?

  I'm totally interested there because you said that about comparables to public market companies. In terms of comparables for AI first companies, there's not that many in terms of in the public sphere. So how's valuing an AI company different from valuing a pure play SaaS company?

• 22:48Speaker 02

  So with valuing a SaaS company, you're sort of valuing it based on how close they are to getting customers to use certain features. And you sort of look at usage metrics and cohorts and stuff like that. With an AI company, you're valuing it based on how close they are to offering a valuable prediction to customers. And I sort of went through before how we figure out at the very early stage, like pre-product or pre-traction, if the model's working or any of that sort of thing. Now, you know, how will they ultimately be valued? Will they ultimately be valued very different to SaaS companies? Yes, because they have pricing power. And that is, and this is the ultimate measure of a moat, right? Like, can you increase prices? As in, is your product just so much better than competitors that they'll, that you're able to just charge way more than those competitors?

• 23:33Speaker 02

  And for an AI first company, you know, they should be able to do that because they're offering a prediction that just no one else can offer at that degree of accuracy. No one else can tell you with 99% accuracy where all your products are in the store or exactly what your customers want at this point in time, or whether this car. Should be repaired or replaced if you're an insurance company, no one can get to that. And so if you have that quality of prediction, then you should be able to charge more, then you should be able to be valued more. And also, AI-first companies tend to initially have much lower margins than SaaS companies, but eventually much higher margins than SaaS companies. And that's another dynamic I can talk about.

• 24:09Harry Stebbings?

  Yeah, can we unpack that? How long can you stomach low-margin AI companies until you get to that inflation point of high? And does that not just mean it's so much more capital intensive and for you as the investor, so much more diluted on your shareholding?

• 24:22Speaker 02

  Yeah, look, AI-first companies are really expensive to build, as I sort of said before, which is you have to build the model and do the machine learning research. You have to build the product to deliver that, then you have to offer and build services to label data and to integrate with customer data and that sort of thing. So they take a lot of money to build, and I think a lot of people don't really realise this, they're a little bit more expensive. However, the point to make is a lot of these initial costs to build a model, so the fundamental machine learning research, the data labelling and whatever else, are relatively fixed, as in there's a certain critical mass of data you need to get the model to superhuman performance. And once you're there, it starts generating its own data, so you don't have to label data.

• 25:04Speaker 02

  And it's already added a degree of accuracy where it's providing a huge amount of value to a customer. So you don't need to spend that money anymore. So you tend to start at forty or fifty percent gross margin, but then you get to ninety-five, ninety-eight percent, we're seeing in some companies we work for, gross margin, and then you just stay there. As opposed to SaaS companies where constantly having to build another feature or deliver a service differently or whatever else. And so you sort of see those margins level out at about seventy percent.

• 25:30Harry Stebbings?

  I mean, my word, ninety-eight percent gross margin makes me incredibly excited, I have to say. I do want to move into, if I'm kind of a segment before we move into the quickfire though, and I spoke to, as we know, many of our mutual friends, and it was Kyle at G C that said, you're a student of venture, and something that we actually have in common, we both love Howard Marks, and you're what a god.

• 25:49Speaker 02

  Indeed.

• 25:49Harry Stebbings?

  Question from Kyle, why the love of Howard Marks and what do you like specifically about him?

• 25:55Speaker 02

  Look, I guess Howard Marks is just constantly focused on what he calls the most important thing. And that is pricing a risk that others can't price. And it was sort of very intuitive or like my liking of his work is really just because it resonates with my pathological allergy to competition. And I just think it's really important to focus on always pricing a risk that other people just seem to find too hard to price or haven't developed the tools to price or whatever else. And that's what he focuses on. So that's why it resonated with me. I think more broadly, it's really important to study investing in all forms because it's fundamentally about that. It's about pricing risk, whether you're pricing risk in real estate, in debt, in whatever else. And there's so much to learn from pricing all sorts of assets that you can apply to pricing startups.

• 26:40Speaker 02

  Like, for example, yesterday, I used something I learned pricing debt ten years ago to understand the startup I was meeting that's offering instant payouts to doctors. And today I've got a board meeting and I'll be using some terms, like very sort of esoteric terms I found useful doing these billion dollar M and A deals fifteen years ago to help with a much smaller acquisition. So I think you just have a lot to learn from all these different fields of investing that do come in handy in the very early days of a startup, but of course as they grow and become big companies themselves.

• 27:08Harry Stebbings?

  I agree with you. The trouble is when I read them and study them, especially when it's kind of non-equity and non-tech, is the element it makes me realise that we're overpaying for everything in tech. And I'm pretty concerned by the pricing that we're seeing. Say, are you as concerned as I am or is it, hey, Harry, this is the new norm. Get used to it, my friend.

• 27:24Speaker 02

  I guess yes and no. I guess yes in pockets for sure. I think there are a lot of people that are starting companies and have certain expectations of what that company is going to end up looking like that aren't really interrogated. And they haven't really done much fundamental analysis of the market and whatever else. And they get themselves into a bit of trouble because they price their first round very high. And, you know, that prices out the follow-on investors, so they can't get follow-on capital. And look, in the early days, you just need capital to make to even have a shot at building a company. So I'm concerned in as much as people sort of price themselves out of ever even having a shot. I'm not concerned in that, you know, these companies can eventually generate so much value that the initial valuation at the seed stage where we invest doesn't really matter all that much.

• 28:10Speaker 02

  And I guess another reason I'm not concerned is I think in many parts of the world, people are a bit more reasonable and have a different approach to interrogating the risks inherent in an opportunity and what that opportunity might eventually look like. And they're not sort of clouded by a lot of lazy thinking that happens when a lot of people around you are just doing a thing and you just assume you can do the same thing. So, yeah, I'm sort of, I am and I'm not concerned in various ways.

• 28:34Harry Stebbings?

  Listen, I think if VC ever doesn't want you, you could always be a politician with that answer. But I do want to talk about another group of people that I know you studied. Tough to draw a correlation to Howard Masters this one, but Italian craftsmen, according to Eric July, he asked, what have you learned from the Italian masters of design and how's it helped you as a VC?

• 28:53Speaker 02

  Well, I guess, look, Italians are just completely obsessed with craftsmanship and being incredible craftsmen. And investing is a craft. I think a lot of people sort of forget that in the technology investing world. They sort of come at it as technologists and then some of them eventually sort of fall in love with the craft of investing. But it is a craft like any other. And I think what I've learned is firstly, if you really want to be a master of a craft, you've got to commit your whole life to it because all good things come from compounding knowledge and network and skills. And, you know, an investor with 15 years' experience isn't 50% better than an investor with 10 years' experience. They're twice or three times as good. So I think the first thing I learned from these master craftsmen in Italy is commit your whole life to one thing. The second

• 29:35Speaker 02

  is investing tools. I mean, the best craftsmen don't just use their tools better than anyone else. They actually end up building their own tools to do certain things, building their own jigs and whatever else. And you can apply that to investing too. Constantly working on tools like checklists and valuation models and whatever else and sourcing tools can help you make better decisions on a very reliable, repeatable basis. Thirdly, invest in apprenticeship as a craftsman yourself. You don't really know something until you try to teach it to others. That's sort of the common trope, the teacher learns from the student. And I really believe that and see that in master craftsmen as well. And then I think finally, it's, and this is sort of all enveloping, pay attention to the details. Pay attention to that dovetail joint or the way a certain corner is rounded or the angle.

• 30:20Speaker 02

  Or approach on a handle or something like that. And in investing, this means pay attention to the algorithm in the footnote of the research paper that the founder wrote, or a formula in a model or a certain clause in a legal contract, because this is where you learn how to analyze things that others just can't analyze or aren't analyzing. And then that leads you to be able to, going back to what I was saying about how it marks, price a risk that others aren't pricing. So attention to detail is certainly something I've learned from them.

• 30:47Harry Stebbings?

  I mean, my God, who knew you could take so much of a telling craftsman and apply it so clearly to venture? I love that all. And the final element, though, that I do have to ask, every one of our mutual friends said that I've got to talk about it, and it's kind of your desire for self-optimization in every area. So first, what have you optimized lately?

• 31:04Speaker 02

  I guess I've moved to a sort of global optimization rather like than local optimization or self-optimization. And, you know, over the last couple of years, I've just been going systematically reducing my impact on the environment. And so this includes everything from cutting out meat to not eating almonds and boycotting that industry. And actually, every week I go through my house and I try to find one product and replace it with a version of that product that has less impact on the environment. So that's sort of like a weekly optimization I'm going to right now.

• 31:33Harry Stebbings?

  I love that weekly optimization. And then if that's the most recent, what's your favorite optimization or life hack that's maybe have the most profound impact?

• 31:41Speaker 02

  This is a tough one because I get asked this question a lot. I sort of reject this term life hack because it's sort of like an oxymoron, right? Like your life is long, but a hack is short. And really, the things that work over time are those that are consistently done. If you just eat less, move more and sleep well every day, you'll be, you're doing great. So I sort of have a real problem with like picking life hack. I mean, that said, if anyone's interested, my friend Oz and I wrote this post, a hundred ways to get more done. So there are 100 hacks in that post, if anyone cares to look them up. And some of them are pretty quirky, then maybe a little bit different to what you've heard. So I encourage people to go there if they want a little hack or two.

• 32:19Harry Stebbings?

  I absolutely love that. I haven't seen it, but I will definitely check that out and probably pick up absolutely none of them. I do want to finish on a question from one of our mutual favorites, Alex McCoura, clear bit.

• 32:28Speaker 02

  Oh, yeah.

• 32:28Harry Stebbings?

  That you have an incredible Six pack question from Andy at Uncork with that in mind, how do you build relationships with other investors? Wait for it, it is correlated.

• 32:37Speaker 02

  Okay.

• 32:38Harry Stebbings?

  Activities, what works in that physical activity bonding environment?

• 32:42Speaker 02

  Yeah, this is something I struggle with. Like, I enjoy being outdoors and physically pushing myself to no end. That is for sure. Andy's right in that regard. But I'm really hesitant to build work relationships this way because it can discriminate against people in all these little ways. So I think the only thing to try to answer Andy's question, the only thing I've found is, you know, just go on long walks with people because that doesn't really sort of leave anyone out. All of that said, and given I've got your audience right now, if anyone's got a physical challenge for me, I'm up for it and contact me and we'll go and do it together.

• 33:12Harry Stebbings?

  Yeah, no, totally. At Ash Fontana, if you are an iron man, I will pay you a huge amount of money to beat Ash.

• 33:19Speaker 02

  I'm scared now.

• 33:20Harry Stebbings?

  That is too funny. I do want to move into the quick fire round there, Ash. So I say a short statement, you hit me with your immediate thoughts. Are you ready to rock and roll?

• 33:28Speaker 02

  I'm ready.

• 33:28Harry Stebbings?

  So I'm terrified by the amount of books that you've read, but tell me, what's the favourite book and why?

• 33:33Speaker 02

  Oh, I'm terrified of answering this question really quickly. I'll tell you what I'm currently reading. I'm reading a book, and this is a curve ball, called The Strategy of Life, and it's an epistemology book, and it's specifically about teleology as manifested in early 19th century German biologists. So this was this really fertile period of biology. And I think the point of it, or what I get out of it is, it reminds me of the need to develop an understanding through reductionism of something, rather than just sort of derive it from experiments. And this is always hard work in this day and age. It's just easier to derive understanding by running a hundred A B tests. But this is fairly incremental. Often we can just make leaps by just thinking more deeply, rather than running just another experiment. So this is a very strange book from this movement of German biologists 150 years ago.

• 34:17Speaker 02

  It's actually very relevant, I think, to thinking about epistemology and machine learning.

• 34:22Harry Stebbings?

  Incredibly interesting. Definitely will not be picking that one up. Sorry, Ash. Tell me, what do you know now that you wish you'd known at the beginning of your time in venture?

• 34:31Speaker 02

  Oh, this is a vulnerability moment. My partner Jocelyn actually gave me great advice that I pass on to a lot of younger founders in particular. Look, people go fast early in their career because of their strengths, but they slow down if they don't work on their weaknesses. And I thought that was really good advice, and I think that's very relevant to someone early in their career.

• 34:50Harry Stebbings?

  That's super interesting and totally screwing up my quickfire round because I often ask the question to my mentors, hey, I'm great at this and I suck at all these things. What should I do? And they always say, Harry, can up tables, anyone can do them, valuation models, basic ones, anyone could do them, commodity skills, only you can do X, focus on that.

• 35:10Speaker 02

  I think that's really important early on, like you've got to make your mark and learn how to bring some value into the world that others can't bring, and that'll get you far very quickly. I think once you've done that and thinking about it more in an emotional sense rather than a functional sense, then start thinking about your weaknesses because they're the things that are going to slow you down once you've built up the momentum. So I think it's just a sequencing thing. I think that advice that you're getting from your mentors is great, but then it's thinking about, well, once I've got that and established that and can deliver this value into the world, what else do I have to do to make sure I can broaden the audience for that value that I can deliver?

• 35:47Harry Stebbings?

  Oh God, I don't particularly want to think about my weaknesses, Ash, it'll take too long. Tell me, how do you get comfortable saying no to great deals, but they're not on thesis for you and Zessa?

• 35:56Speaker 02

  Oh, this is a really hard one. I always just tell the team that we don't have to do everything, we just have to do a small number of incredibly good things, and that's what gives us comfort, to be honest.

• 36:06Harry Stebbings?

  Why have you been investing in Europe since the beginning, since before the US has made that march in recent six months?

• 36:12Speaker 02

  Oh yeah, fundamental AI research in Europe, in certain areas at least, is the best in the world. There's more research papers, AI research papers that is coming out of Europe than anywhere else, but European AI entrepreneurs get one tenth the funding of US entrepreneurs. So my very first investment for Zeta was Trackable, a company in London that I came across from doing some research into active learning in 2014, and our largest investment is in Cambridge as well, a company called Ingenia that is sort of using some Bayesian research out of Cambridge to make predictions to make the energy grid more efficient. So yeah, there's this better fundamental research in certain fields and shit money.

• 36:47Harry Stebbings?

  From Andy McLaughlin, pros and cons of being a foreigner in the valley. Clearly he's feeling it.

• 36:53Speaker 02

  I think there are only pros. Having an accent is like playing life on easy mode. People at least remember who you are.

• 36:59Harry Stebbings?

  This is totally true, although I get the common question, are you Benedict Evans?

• 37:03Speaker 02

  Oh, yeah.

• 37:04Harry Stebbings?

  And then the next one from Greg Sands, he says, is WeWork full of shit? Sorry to ask.

• 37:09Speaker 02

  Well, all I'll say is not totally. Creating something from nothing is very, very hard. And they've definitely done that.

• 37:16Harry Stebbings?

  Totally agreed. Good answer. What's the biggest challenge of your role with Zetta today? Penultimate one?

• 37:20Speaker 02

  I mentioned before my pathological allergy to competition and there's an influx of seed funds, all sorts of AI experts, so to speak, and think that is very difficult for me to deal with on a daily basis, given I'm so allergic to competition. We just try to stay the course.

• 37:36Harry Stebbings?

  Well, Ash, you know, I'm a huge fan of yours. I've wanted to do this for a long, long time. A huge thank you for joining me today and I can't wait to see the exciting times I have for you and Zessa.

• 37:44Speaker 02

  Thank you very much, Harry, and likewise on a 20 minute VC and with the fund.

• 37:50Harry Stebbings?

  Now, if you couldn't hear from my tone, I absolutely loved having Ash on the show there. If you'd like to see more from Ash, which is a must, then you can find him on Twitter at Ash Fontana. I do want to say a huge thank you both to the amazing people that suggested such fantastic questions and to Ash for being such a good sport on the episode today. I really do appreciate that. If you'd like to see more from us, you can do so on Instagram at H Stebbings 1996 with two Bs. It would be great to see you there. But before we leave you today, I'm sure you've heard about it, but my word, this is a product I love, Carter. Carter simplifies how startups and investors manage equity, track cap tables and get valuations. Go to Carter.com/20vc to get 10% off and more than 800,000 employees and shareholders use Carter to manage hundreds of billions of dollars of equity.

• 38:33Harry Stebbings?

  And Carter now offers fund admin, so you can see real time data in the Carter platform and work with Carter's team of experienced fund accountants also in the background. Go to Carter.com/20vc to get 10% off. And speaking of companies that have taken our ecosystem by storm, I have to talk about Brex, the company which built the corporate card for startups and who is the fastest company to reach unicorn status in history. Brex founders Henry Can Pedro built a payments business in Brazil, but kept getting rejected for a corporate card in the States. So they built Brex with. No personal liability, up to 20 times higher card limits and huge rewards like 7X points on Uber and Lyft, 4X on Brex Travel and 2X on SaaS software. If you're a VC backed startup based in the US, see if you qualify for a Brex card today at Brex.com

• 39:18Harry Stebbings?

  and get Cardfees waived for life by entering the code HARRY during signup. And finally, you have to check out Rank Science. Rank Science is the easiest way to grow organic traffic and get your content ranking higher in Google. Their AI-powered platform uses real-time search data and natural language processing to help you craft content that is highly relevant to search intent and maps directly to what users are searching for. For customers with lots of landing pages, you can leverage Rank Science's SEO/AB testing platform to easily test which HTML changes cause an uplift in search traffic. Also, as an added bonus, their platform also allows marketers to make SEO enhancements to websites without eating up precious engineering resources. Rank Science is used by hundreds of companies to grow organic traffic,

• 39:58Harry Stebbings?

  from startups to publicly traded companies, including the likes of Build, Zoom, Goldbelly, Sweetness and Career Karma. Go to rankscience.com/20vc for your first month free. As always, I cannot thank you enough for your support and I can't wait to bring you a fantastic episode on Friday.