Tag: Startups

  • LPs should encourage VC evolution

    LPs should encourage VC evolution

    In a previous article I wrote about the threat of consensus in venture capital.

    A few days later, Eric Tarczynski shared a fascinating thread about the journey with Contrary, his VC firm. He addressed this point about consensus with admirable candour, summarised here in two points:

    1. Raising from LPs is easier if you have recognisable logos attached to your previous funds. Success is measured by which big names in VC co-invested with you.
    2. Raising from LPs is easier if they get good references from their existing VCs. So you send deals to them, network with them, and co-invest with them. Success is measured by relationships.

    It’s unusual to get such an unvarnished look at the inside workings of venture capital, and the thread elicited a number of reactions. Most agreed it was a tough pill to swallow:

    Eric’s awesome but boy is that thread a pretty damning look into the inside-baseball-nepotism that starts from the top (LPs) and infects the whole VC ecosystem.

    Luke Thomspon [source]

    ‘We thought that being good investors with a unique thesis that actually makes money would be the best strategy, turns out, following the herd, piling onto garbage, and being unquestioning vassals to incumbent investor power gets you a larger fund’ – My interpretation

    Del Johnson [source]

    There’s an elephant in the room in all of this. Or perhaps it’s a bull in a china shop. Either way, everyone seems to be ignoring it and it’s doing a lot of damage.

    Weak signals

    From pre-seed to IPO, there is no consistent, transparent measure of success. That’s a long time for a GP to deploy capital without any concrete metrics for success. How does an LP ascertain if their money is being put to good use?

    Samir Kaji of Allocate (and former SVB MD) shared his take on the problem that LPs face:

    LPs are programmed to use past track record as the primary driver in making a decision on whether to invest in a new fund (A recent study showed historical persistence of VC is that 70% chance a fund performs above median if prior fund is 1st Q). However, more than ever, track record can be a very weak indicator if the fund is within <5-7 years.

    • Spread of how VCs are valuing the same companies is large.
    • Current TVPI to final DPI delta will be large for many funds, and some funds have resilient companies; others are filled w/companies that were pure momentum (but still marked up).
    Samir Kaji, Allocate

    There is an obvious desire from both sides to find something to show. As Luke put it, “we can pretend it’s all about independent thinking, non consensus and right, etc, but when you’re going out for Fund 2 and on a stack of unrealized, LPs want other signals.”

    This is why we end up focusing on ‘logo hunting’ and co-investment culture. If we’re all a gang, and we back each other up, then we’ll maintain the confidence of LPs. Meanwhile, the LPs are probably feeling a degree of comfort from investing in a few different funds, without realizing how intermingled and codependent they are.

    As Chamath Palihapitiya wrote in Advice to Startup Founders and Employees: Strength Doesn’t Always Come in Numbers:

    As it turns out, what VCs of the past decade assumed to be market alpha may have actually been market beta (i.e. fellow venture funds bidding up the same cohort of companies over several funding rounds).

    Chamath Palihapitiya, Social Capital

    This is clearly an undesirable outcome for LPs: The data for measuring venture capital fund performance is flimsy and creates a huge perverse incentives for GPs. This is clearly not good enough when so much capital is at stake. Especially when it involves pensions funds and university endowments. It’s a bad look for everyone.

    The final nail in the coffin here is how current practices can create a reality-distortion field around actual performance: in effect, a company’s ‘public’ valuation only changes when they want it to. This was outlined at length in a thread from Anand Sanwal of CBInsights, which included this slide from SVB:

    This is on the mind of every LP at the moment. What do their ‘paper’ returns from 2021/22 actually mean anymore? What will happen when the companies they are invested in via VC are forced to come to terms with reality?

    Meaningful benchmarks

    When you start talking about standardising anything in venture capital, there’s a reliably cold response. Everybody likes to believe they have their secret sauce, their intuition, their process, their edge over others… despite all signs pointing towards none of that changing the outcome.

    When you talk about measuring the performance of early stage companies, that’s when the real pushback begins. There’s too much uncertainty. It’s too unreliable. Projections are always a pipe-dream.

    There’s one simple response to these concerns: “Perfect is the enemy of good“.

    If you open yourself to new ways of looking at valuation (it’s not just about “market passing”), and new ways of performing valuation, you will find that there are practical, systematic frameworks to measure and report the development of private companies.

    Don’t get twisted up about producing an “accurate” result for an early stage company, it is foolishness – and not the point. The goal is to provide solid, useful benchmarks which can be calibrated against the market in a transparent manner.

    For an example of how this might be achieved, I will always recommend a read through Equidam’s methodology. It combines perspectives on verifiable characteristics via the qualitative methods, the exit potential via the VC method, and the vision for growth via the DCF methods. All packaged up into a nice, comprehensive report.

    What standardized reporting does for the LP/VC relationship

    If you can imagine a world where VCs produce quarterly reports on fund performance using a standardised framework, there are a number of profound benefits:

    1. LPs could better assess the performance of their existing VCs, creating more of a meritocracy.
    2. VCs would have an easier time raising, in addition to shortening their own internal feedback-loops to improve decision making.
    3. Moving away from current lazy valuation practices (ARR multiples) would help avoid extreme fluctuations in valuation, as we’ve experienced since 2021.
    4. It will (slowly) kill the dinosaurs, the giant firms which played a part in the development of this ecosystem and all of its flaws.
    5. A move towards transparency – especially around valuation – would be timely, as the SEC’s gaze falls on venture capital.
    6. There are also interesting considerations for liquidity in secondary markets serving private company equity, but that’s a whole post of its own.

    Conclusion

    It seems clear to me that this change will not come easily to venture capitalists, who are either comfortable with the status-quo or simply find it convenient. However, it might be possible for LPs to set new terms as market dynamics have shifted power in their direction.

    Still, this is a difficult argument to make. I’m suggesting no less than upending how much of venture capital operates, and I’m doing so from the position of a relative outsider.

    But I guess that’s the point? Venture capital has been a closed ecosystem for too long, full of esoteric practices shaped by a relatively tiny group of individuals. There is plenty of room for improvement, especially if we stop getting hung up on the need for ‘perfect’, when the current status is ‘poor’.

    Finally, a bigger point than any of the six I mentioned previously: if this makes us better at allocating capital to innovative ideas, and innovative people, then it’s got to be worthwhile.

  • Generative AI and the Games Industry

    Generative AI and the Games Industry

    This post looks at applications of generative AI in the context of the games industry, but much of the same logic can be applied elsewhere.

    Adapting to technological evolution

    With every new technology revolution – web3 most recently, and now AI – there follows a large herd of true believers. It can do all things, solve all ills, and life will never again be the same again. Enamoured by possibility, they follow with a true sense of opportunity.

    Loudest amongst this herd (and most critical of nay-sayers) are the wolves in sheeps’ clothing. The rent-seeking charlatans.

    This was explicit in the get-rich-quick era of web3, and much of the same problem has transferred over the AI as techno-pilgrims flee one sinking ship to pile into another.

    Secondly, on the other side of the coin, are the cynics. People who were raised on 56k modems and bulletin boards, who feel a deep discomfort as technology moves beyond their grasp. They felt like the rational resistance to web3, and so have little hesitation about weighing in on AI.

    We have to be conscious of both groups, and our own place on that spectrum.

    Why the games industry?

    There are three main reasons I’m keen to address the games industry as the case-study for this post:

    1. As with web3, AI is being shoved down people’s throats without due concern for why.
    2. It is largely focused on a young audience who are absent from these conversations.
    3. It connects with my personal experience in the games industry.

    If you want to read about the potential use cases for AI in banking, you’ll find a thousand thought-leader think-pieces. It was well-covered ground without much original thought even before ChatGPT came along.

    If you want to talk about the potential use cases of AI in the games industry, you’ll find some ex-crypto VCs and technologists trying desperately to pivot their brief experience. Insubstantial waffle.

    Perfection is the enemy of good

    Dealing with the more exciteable technophiles, you’ll probably notice they don’t show a lot of interest in the complex applications. Their interest is in the most extreme examples of movies, games or books being entirely generated by AI (or entirely decentralized, yada yada).

    Their point is simple: if AI can do these things crudely today, then tomorrow it will be able to do them well – and at that point we’ll be forced to embrace the bold new future. Right?

    This fallacy can be observed in every parent watching their child smear paint on paper for the first time: something inside them says ‘they could be a great artist’. It’s true: the ability to manifest art can be that simple, and the child has huge potential for improvement… Yet it’s still not going to happen for all but a miniscule few.

    In both cases, the AI model and the child, there cannot merely be push, there must also be pull. There must be a need being met. An appetite being satisfied. And 99% of the time, there isn’t. Once the novelty has worn off, nobody has any interest in watching an AI-generated movie, reading an AI-generated novel, playing an AI-generated game, or looking at your child’s paintings. There just isn’t a call for it.

    Instead of putting AI on the pedestal of a godlike creator, we should look at where it can be a tool to solve a problem.

    Merchants of fun

    You can get side-tracked in talking about experiences, socailising, adventuring, exploration, curiosity, challenge, status… Ultimately, games are vehicles for fun. That’s bedrock.

    Is an AI-generated game likely to be more fun than the alternative? No, of course not, and if you suspect otherwise then you’ve not spent enough time with the wonderful and wacky people who make games. They are true creatives.1

    Any application of generative AI to the games industry must have either enhance fun, or enhance the developers ability to deliver it.

    Exploration

    If you look at games like Minecraft or 7 Days to Die where you can explore a proceedurally generated world, it’s easy to see how generative AI might be able to supercharge that environment building.

    It’s worth considering, though, that this is a specific approach for a specific type of game. As good as these engines have gotten, most of the time games will require a more ‘designed’ world, with geography or features which play into gameplay mechanics, story elements or IP. Generative AI may offer tools to make this more efficient (as many proceedural tools already do), but is unlikely to replace it entirely.

    Socialization

    Imagine walking around a Skyrim or Cyberpunk style sandbox-world, full of NPC characters with their own unique look, voice, and personality. Each able to hold a conversation with you, flavoured with their own specific personality and knowledge. Not merely giving canned responses to pre-defined prompts, but able to interact fluidly with you and amongst themselves.

    Again, this is unlikely to ever be all a game needs. Stories still require specifcally designed characters with particular roles which need to be shaped by the intention of writers and a design team, but it is still a tremendous opportunity to solve the social component of virtual worlds.

    These are two quickly-sketched examples of how generative AI could enable a leap forward in the experience provided by games devleopers – and I am sure there are many more to be found.2

    Tapping into the market

    I wanted to do this in a more subtle manner, but it’s just more practical to break down Andrew Chen’s Twitter thread:

    Games can take 3+ years to build, and technology adoption happens at specific windows of time

    If your generative AI tool is a plugin (for the Unreal Engine, for example) then a studio can pick it up at any time and add it to their development stack.3

    You shouldn’t be limited to thinking in terms of ideas that are ‘disruptive’ to how games are made, and indeed most of the opportunity may be in ideas which are complimentary.

    indie games make little $. There’s only a few scaled players, who will always push on pricing

    If you were going to target indie developers it would have to be with a very specific value proposition and business model (e.g. Unity in 2004). There’s no reason to worry about this otherwise; there are enough larger studios.

    the games ecosystem is insular, with its own conferences, luminaries, and networks / networking” in the games industry often involves, well, gaming. Are you good at Valorant? 🙂

    Can you tell me an industry which doesn’t have its own conferences, luminaries and networks?

    The games industry is not insular, and it is comical to characterize it as a bunch of nerds playing games together. It’s a wonderfully open, social and diverse community.4

    a large % of game cos have artists and creative people. Many are threatened by, and oppose, AI tech

    I don’t know of anyone in the games industry, artist or designer, who isn’t starry-eyed at the possibilities of what AI can enable.

    They are also familiar enough with how games work to recognise that human input is always going to be required to shape and polish the human experience which emerges on the other side.

    you need to generate editable, riggable, high-quality art assets. Right now assets are idiosyncratic and hard to edit

    Generative AI has not yet proven that it can generate useable assets, never mind well-optimised thematic assets. That problem can probably be solved, but to what end?

    Will a world created by a generative AI ever truly feel interesting, coherent, beautiful? Maybe there are better things for it to do?

    large publishers often provide tech to their internal studios. They’ll partner to learn about AI, but will try to build in-house. Is your tech defensible?

    That might have been the case 15 years ago, but the vast improvement in game engines and tools has meant that developers are much more likely to build on existing platforms.

    If a publisher believes that a tool would make development cheaper and faster then they’ll support it without blinking.

    large gaming cos care a lot about their models and data not being shared across the industry. How do you guarantee that? / they also care that their models are trained on data that’s safe from a copyright perspective. There’s lots of hoops to jump through

    Stretching a bit here, but: You train your tools on an open set of data to the point where they are useable, and allow developers to provide additional training based on data from their own IP. In that scenario there is no reason for crossover between studios.

    It’s unlikely that training from one game would ever be useful to the application of the AI in another. It is probably more likely to produce undesirable results.

    Conclusion

    Some years ago an associate of mine went to interview for a job at a games company in Seattle. The interviewer had previously been the lead designer on Starcraft, and naturally expected the candidate to play a match against him while fielding questions about the role.

    The games industry is full of these amusing anecdotes of quirky behavior, and there is a pronounced culture associated with that. However, it is condescending to think that culture stands in the way of progress, or that games studios can’t engage with business and technology partners in a perfectly competent manner.

    If you make a useful tool which solves a problem for the games industry, you will be able to access the right people to make a sale. I’d go so far as to say it’s probably easier and faster moving than many other industries.

    If that is your aim, make sure you are spending enough time talking to games developers, learning about how games are made, understanding the player mentality, and the problems that you might be able to address. As always, finding product:market fit can require a lot of learning and iteration.

    Most of all, ignore the false prophets who were reading from the web3 gospel just a few months ago. They will just ride this trend until something else comes along.

    1. Yes, throughout this article I am drawing a deliberate and passive-aggresive distinction between ‘creating’ and ‘generating’. [↩]
    2. It bothers me that I covered Explorers and Socializers, but didn’t have the time to identify anything for Achievers and Killers. [↩]
    3. And in most mid-large studios there are usually multiple teams running in parallel focused on different projects at different stages of development. [↩]
    4. The irony of a venture capitalist calling the games industry ‘insular’ is not lost on me. [↩]
  • Why venture capital should be consensus-averse

    Why venture capital should be consensus-averse

    In The General Theory of Employment, Interest and Money, Keynes wrote about investment through the metaphor of a newspaper contest to select the six best looking people from a group of photos, with the prize being awarded to the contestant whose choice most closely corresponded to the average of all contestants.

    Keynes’ point was that, despite the clear and simple instruction, contestants are actually not inclined to consider which of the photographed people are the best looking. Rather, they now consider a third-degree perspective of ‘what would the average person imagine that the average opinion is?’

    We have reached the third degree where we devote our intelligences to anticipating what average opinion expects the average opinion to be. And there are some, I believe, who practise the fourth, fifth and higher degrees.

    John Maynard Keynes, Economist

    In A Simple Model of Herd Behavior, Abhijit V. Banerjee examined the inefficiencies created when decision making becomes reliant on signals from others. We become inclined to abandon our own data, in favor of prioritizing signals which themselves may also be based on nothing more than another prior signal. 

    If decreasing returns (average payoffs decline as the number of people who choose it increases) tends to reduce herding, one would expect increasing returns, which rewards doing what a lot of others are doing, to increase the tendency to herd. This is indeed what we find.

    Abhijit V. Banerjee, Ford Foundation International Professor of Economics at Massachusetts Institute of Technology

    There are a number of social psychological drivers of this behavior, but the most obvious are our desire to associate with popular choices, and the greater dispersion of responsibility if that choice proves wrong. 

    Consensus threatens innovation

    Generally, herd behavior is problematic in how it undermines sound judgment and rational choice, though by nature it tends to be low-stakes and risk-controlled. For venture capital, this innately human behavior should be viewed as an existential threat, running contrary to the needs of effectively identifying and funding innovation. 

    If no great book or symphony was ever written by committee, no great portfolio has ever been selected by one, either.

    Peter Lynch, former manager of the Magellan Fund at Fidelity Investments

    The root of the name venture capital, as Evan Armstrong reminds us in Venture Capital is Ripe for Disruption, is adventure capital. It’s only really an adventure if you’re not sure of the destination, and backing innovation is exactly that: you are straying into the unknown; high risk, large potential reward. 

    The classic archetype of a venture capitalist, fitting with this concept, is a highly perceptive and analytical individual who can evaluate all kinds of oddball, out-of-the-box startups and identify the ones with potential. Someone who sees opportunities where others do not, who does not care about (or actively avoids) pattern-matching with past successes, and who ignores the noise of signals from their peers.  

    There is an old saying in enterprise software, “No one is fired for buying IBM”—people mitigate risk for their decisions by choosing the consensus option.

    This occurs even in the supposedly risky world of venture capital.

    Evan Armstrong, ‘Reformed’ Venture Capitalist

    Hunger drives herd behavior

    In recent years, as the appetite for cheap capital grew to unsustainable heights, venture capitalists became preoccupied with following external signals to ascertain whether the market would agree to provide capital to their portfolio. Would their peers validate their investment choices? Would prospective LPs recognise the value of earlier investments if they weren’t shared with other respected names? Herd behavior crept in with pernicious effect; the seductive comfort of piling into seemingly safe deals with other investors. Manufacturing winners. 

    As long as downstream investors continued participating in the game of artificial value growth (and why wouldn’t they) it was still a good model, right?

    As long as the (paper) returns were good, it was still venture capital, right?

    We know how that ended. We also broadly know why it ended (crude valuation practices, interest rates making capital more expensive, exit markets rejecting inflated prices… etc). The question we should ask now is what can be done to stop it happening again? 

    Learning from mistakes

    Anyone involved in investment of any kind should be aware of the way signals should be handled (with oven gloves). It is valuable input that can shape an investment decision but shouldn’t drive it. For venture capital, that might mean reevaluating everything from deal flow management to valuation practices. 

    • Are the majority of your deals sourced through referrals from other investors?
    • When evaluating potential investments, how dependent is your conviction on recent similar deals? 
    • How much analytical rigor are you applying to the individual nature of each opportunity?
    • When setting valuation, how much do you rely on crude ARR multiples?
    • How much does the VC Twitter echo-chamber shape your approach to early stage investment, generally? 

    These might seem like basic questions, but there is clear cause to begin a first-principles reevaluation of how capital is allocated to ideas and founders. The responsibility is to effectively fund technological progress, not to exploit an uncertain market for short-term gains.

    A new approach, with a more analytical focus on individual businesses, may seem unrealistic: too much time involved, too much uncertainty. To that, I’ll close on three points:

    • Startups in 2023 are running leaner. The great hunger for capital is over, for now. That opens the opportunity to strike out and make fund returning deals without needing to drag other investors along with you. Your ability to identify winners (not simply agree on them) matters more than ever.
    • There are tools and frameworks which make analysing startups in detail much more practical (Equidam is an obvious example). Build a process which lets you collect data about opportunities and decisions, allowing you to develop and codify your experience.
    • Reconsider industry dogma about practices and perceptions (for example: about financial projections at early stage). More data = better decisions, you just need to pick the right lens to derive the right value.

    As many have said, the 2023 vintage has great promise. Particularly for investors who best adapt to the new conditions.

    [EDIT 26/03/2023: Adding a link to Chamath Palihapitiya’s article about herd behavior in venture funds and the risks involved. It’s a much more analytical perspective, which you can read here.]

    [EDIT 22/06/2025: Adding an overdue link to Geri Kirilova’s article about enmeshment in venture capital, providing another perspective on this problem, which you can read here.]

  • Screening Pitches

    Screening Pitches

    There are five straight-forward questions with which you can quickly evaluate a startup pitch, combining the strength of a proposition with its delivery.

    These questions bear some some resemblance to the Scorecard Method of startup valuation, which focuses on qualitative measures for early-stage companies, but with an additional focus on quantifying the market need.

    I have applied this approach to screening accelerator applications, but it can be used as the first step of evaluation in any pitch process.

    For the sake of simplicity we can score each of these on a scale of 1 to 5.

    1) Severity of Problem

    This is a question that can vary significantly based on the market you are looking at. Emerging economies tend to have more of a focus on the (high scoring) primary problems, which is why they’ve been able to better resist economic downturns.

    1 – Micromobility, dating apps, rapid delivery (esp. red ocean)

    5 – Access to water, energy, core financial services (esp. blue ocean)

    2) Strength of Solution

    Simply, are you providing a way for people to better cope with a particular pain, or have you managed to cure it in a complete and lasting manner?

    1 – Solution alleviates the problem

    5 – Solution eliminates the problem

    3) Scalability

    There’s almost always a focus on the size of market. TAM, SAM and SOM will feature in virtually every startup pitch deck. What’s often overlooked is how easy it is to scale into that market. Regulatory barriers, poor infrastructure, or corporate customers who move slowly are always a threat.

    1 – Infrastructure or regulatory requirements, long sales cycles and onboarding (esp. in small markets)

    5 – Web or mobile based product that is available on-demand to the entire target market (esp. in large markets)

    4) Profitability

    In many markets a poor product will win if it is just slightly cheaper than a better product. This kind of price suppression can be a killer for otherwise solid businesses. Similarly, some problems require costly solutions like agent networks, physical touchpoints, or a highly involved sales and customer service capability.

    1 – Low margins (high CAC/CRC/COGS, low LTV)

    5 – High margins (low CAC/CRC/COGS, high LTV)

    5) Team

    This is the hardest part of a pitch deck to quickly evaluate, and requires the most additional research. LinkedIn, interviews, papers, Glassdoor… the number of potential resources extends as far as your willingness to do the research.

    1 – No obvious fit for the problem being solved, by education, experience, or personal background.

    5 – Exactly who you imagine should be tackling this problem, with a combination of both motivation and ability.

    Conclusion

    At the end of this fairly rudimentary process you have a score out of 25 which should give you a very broad overview of the potential of this business. It is intended to quickly take a list of some hundreds of pitches down to the 20-30 you think are worth a closer look.

    At that point you can then start looking at some of the more granular data:

    1. Existing partners, strategic relationships, etc
    2. Industry and regional context
    3. Traction and development of competitors
    4. Revenue forecasts and unit economics
    5. IP considerations