Data-First Venture is Coming

A reminder for new readers. That Was The Week collects the best writing on critical issues in tech, startups, and venture capital. I select the articles because they are of interest. The selections often include things I disagree with. The articles are only snippets. Click on the headline to go to the original. I express my point of view in the editorial and the weekly video below.

This Week’s Video and Podcast:

Content this week from : @abhibhatia91, @GaryDushnitsky, @OxfordSBS, , @krishnanrohit, @evatventure,@Samirkaji, @fredwilson, @ttunguz, @lawrencebonk, @OReillyMedia, @poyark, @moritzKBE, @sequoia, @jasonlk


Editorial: Data-First Venture is Coming

Essays of the Week

The Future of Venture Capital? Insights Into Data-Driven VCs

Has persistence persisted in private equity? Evidence from buyout and venture capital funds

Why more is less in investing

Symposium: The unique uselessness of business advice

VCs make mistakes too. And LPs need to be aware of them.

How This Ends (Part Three)

Video of the Week

Producing Charts with AI – Tomasz Tunguz

AI of the Week

ChatGPT update allows it to remember who you are and what you like

Teaching Programming in the Age of ChatGPT

The AI craze will implode faster than #Threads (🫢) if folks can’t figure out how to make money from AI.

Apple is reportedly developing its own generative AI chatbot to rival ChatGPT

Microsoft launches vector search in preview, voice cloning in general availability

News Of the Week

Silicon Valley start-ups explore sales as funding runs dry

Venture debt deals decline 38%, led by SVB’s core market

Meta unfriends the news industry in growing rift with publishers

BlackRock’s Larry Fink says crypto will transcend any one currency

Court denies FTC’s last-ditch attempt to stop Microsoft buying Activision Blizzard

Startup of the Week

Sequoia Capital

Tweet of the Week

Entitled Employees?

Editorial: Data-First Venture is Coming

I spent a lot of this week using ChatGPT with the data we maintain at . We have data for over one million funding events, covering over 40 fields per row. More than 50 million data points altogether.

We learn from the data and build predictive models with the goal of selecting the top 5% of Series B rounds each year.

It is very good. The average performance after 5 years is that a selected Series B returns 6x growth in value. 25-30% become unicorns.

So you would think, as CEO of SignalRank, that I am a big believer in data-driven investing in startups. But the truth is far more subtle. And there is a very big difference between data-driven investing (human decision-making, supported by data) and data-first investing, where humans are simply executing from the recommendations of a model.

But I digress. First, some background.

Venture Capital is not best understood as a single asset class. It is at least three (seed, venture, and growth) and arguably even more.

The Seed and Series A part of venture capital is hard to apply data to. Companies have little in the way of real businesses. Founder background and hiring patterns are all very large producers of false positive signals. Those that claim to have insight rarely do. The best investors at these stages combine experience, often as operators, with intuition and analysis. The best of them pick future winners repeatedly, and they play a huge individual role in that. Seed investors are the true venture investors, and that is unlikely to change.

Look at the top early-stage investors from the point of view of unicorn production.

This measures their Seed and Series A investments. They produced 769 unicorns. No data could have found them at that stage. Even SignalRank’s algorithms would need over 7500 Series A investments to get to that number of unicorns. A very good one-in-ten hit rate, but the best seed investors also do that, or better.

So early-stage investing is mostly art, with a little analytics built in. Mike Maples, Hunter Walk, Aileen Lee, Garry Tan, Paul Graham, Ron Conway, Reshma Sohoni, Carlos Espinal, Sia Houchangnia, Suzanne Ashman Blair, Saul Klein, and many more emerging managers are the heroes of this early-stage ecosystem. To peek at the data results – here is the SignalRank top 20 seed investor list from the past 5 years. Scores are correlated to outcomes over 5 years from the date of investment, or less if 5 years have not elapsed. If you want to see all 850 scoring seed investors, you can subscribe to SignalRank AI by emailing us.

By the time a Series B term sheet comes along, these early-stage investors are no longer capitalized to invest and suffer dilution. But data is capable of differentiating between their likely future fund returners and those who are not likely to perform at that level.

The data can be used in varying degrees to filter companies based on likely outcomes. And models can be trained to recognize these likely good outcomes based on features discovered in the data.

The later the funding round, the more data can help.

Not to sell my startup’s vision here, but at SignalRank, we have shown that a 100% data-driven Series B investment selection process can beat any human Series B investor in outcomes and efficiency. And with no human override.

And the same at C and D.

If we are right (and we are), it seems likely that venture investing will increasingly be human-led at the early stage and machine-led later. And by machine-led, I mean no human oversight. Early-stage investors can expect to have capital available to their best companies from a pool of capital designated for later rounds and earn carry from the profits. This pool of capital would remove the need for opportunity funds or special one-off vehicles.

Early-stage investors will partner with later-stage data-driven allocators to secure the future outcomes of their high-scoring portfolio companies. Venture Capitalists will not be replaced but augmented.

Data-first investing is being born.

Capital allocators have yet to catch up with this trend. But soon, rather than looking at the track record of an individual or firm, later-stage allocators will want to know how a model backtests and will allocate to the best models. If the average performance of a Series B investment after 5 years is 6x growth in value and 25-30% become unicorns, it will be hard not to allocate capital to the model.

This future is built on the human-led seed and Series A investing ecosystem.

Rohit Krishnan has a quote describing the social context in which ideation (and early-stage investing) thrives:

In the 17th and 18th century, there began a coffee drinking scene in London, bringing with it an incredible scene of intellectual debates and spun off innovation. Soon London was only second to Constantinople in number of coffeehouses!

Silicon Valley is basically a large coffee house. And humans are drinking the coffee. There is the subtlety. Human first seed investing and data first later stage, from the Series B onwards. You read it here first.

So, no surprise I read a lot about data-driven venture investing this week. This is random. It just happened to be in my feed. Abhishek Bhatia and Gary Dushnitsky from the London Business School get essay of the week for The Future of Venture Capital? Insights Into Data-Driven VCs. It is worth a deep dive. Especially the section called Algorithms as Startup Investors. And a second paper on whether good investors are persistent – with good data. Koble Moneyball’s Why more is less in Investing also focuses in on data processes to filter out losers.

All these approaches focus on how data can lower portfolio risk, possibly to zero. If that is the case, then data first investments in private companies and indexes built by securitizing the assets invested in will become very popular with investors. For me, once a company is post-Series A and has a Series B term sheet, data can help the early-stage investor allocate to their likely future winners with significantly reduced risk.

More in this week’s video and podcast.

Essays of the Week

The Future of Venture Capital? Insights Into Data-Driven VCs

by Abhishek Bhatia and Gary Dushnitsky

Image Credit | Mika Baumeister

Can data-driven tools crack the art of finding and funding innovative startups?


Algorithms have proliferated across industries and professions. The successful application of Large Language Models (LLM), Robotic Process Automation (RPA) and other data-driven tools is the hallmark of our time.1 Yet, the contribution of such tools across different use-cases remains a topic of heated discussion. Case in point is the extent to which algorithms can engage in creative and innovative tasks or offer an accurate evaluation of their value. The successful application of data-driven approaches is documented across a host of business and scientific settings (Agrawal, Gans, & Goldfarb, 2018; Brynjolfsson & McAfee, 2014; Haenlein & Kaplan 2019; Krakowski, Luger, & Raisch, 2022). Yet, the discovery of entrepreneurial opportunities can be seen as an extreme test – think of it as ‘frontier application’ of data-driven tools. The impact of a data driven approach to entrepreneurial discovery remains a topic of open conversation.


“A Brief History of Artificial Intelligence: On the Past, Present, and Future of Artificial Intelligence” by Michael Haenlein & Andreas Kaplan. (Vol. 61/4) 2019.

We seek to inform this conversation through a mix of anecdotal and large-sample evidence on the practice of data-driven venture capital investments. Venture capital (VC) investors are in the business of discovering and backing innovative entrepreneurial ventures. Since the early days of George Doriot and the American R&D Corporation, venture capitalists were viewed as investment connoisseurs who rely on their business acumen and networks to source and screen potential investment opportunities (Bhide, 2003; Fried and Hisrich 1994; Kaplan and Stromberg 2001). These practices were often facilitated through tight connections with geographically proximate entrepreneurs and industry veterans as well as co-investors with whom they share an investment history (Gompers and Lerner 2001; Rider 2012).

Data Driven VCs

Recently, some venture capitalists have opted to embrace data-driven approaches (Gompers, Gornall, Kaplan, Strebulaev 2020). The tools are deployed for various use cases, such as sourcing and screening of nascent startups or more mature investment opportunities. Below, we focus on the use of data-driven tools towards sourcing and screening nascent investment opportunities; namely, the practice of ‘first money’ into budding startups. We further discuss findings from analysis documenting portfolio characteristics for data-driven as well as traditional VCs.

To understand the data driven proposition in VC, we are reminded of the key activities over the lifespan of the VC fund; sourcing, screening, signing, and supporting. Traditionally, investors relied on their teams and networks to excel along the four dimensions. Recently, several VC investors have advocated the use of data-driven methodologies. Consider Sourcing. Signalfire, a U.S.-based VC, makes use of a data platform called Beacon, which its CEO, Chris Farmer, describes as, “a proprietary mini-Google.”2 Beacon is said to track more than 6 million companies in real-time by drawing upon 10 million data sources. Similar arguments pertain to the Selection dimension. The San Diego based Labx Ventures declares it can overcome bias and increase accuracy through the use of a proprietary New Venture Assessor (NVA), named RubX. The VC firm’ website explains that RubX “gives us the power to make scientifically-based recommendations and unlock the core strategies necessary for success,” and “we correctly predicted—with over 80% accuracy—whether investors would have a high ROI within 7 years.”3

Investment Patterns

Informed by these anecdotes, we turn to document key characteristics for the portfolios of a set of data-driven (D-VC) and traditional (T-VC) investors. Below, we summarize the key findings based on preliminary analyses of twenty US-based funds and all their portfolio companies. To gauge the extent to which D-VCs uncover under-represented ventures, we focus on three commonly observable features of nascent startups; their geographical location and the attributes of the lead founder.

We observe that D-VCs exhibit a similar pattern of targeting ventures based in well- known ‘startup hubs’ across the USA. Specifically, about 52% of the initial investments undertaken by D-VCs are targeted at those in hub locations. The ratio is similar to the 51% of ‘startup hub’ investment for the portfolio of the comparison T-VC group.

A comparison of founders’ attributes uncovers more meaningful differences. Analysis of CEO gender suggests that D-VCs fund a third more female-led companies compared to their T- VC peers; 13% versus 10%. The analysis of educational background reveals nuanced patterns as well. Among the U.S.-based investors we study, D-VCs are less likely to back ‘elite’ graduates, in comparison to T-VCs; 59% versus 66%, respectively.4

Taken together, these observations are consistent with a view of ‘algorithms as tools for exploration.’ The features and timing of their early investments suggest that D-VCs utilize the tools towards an exploratory approach. They derive value from such tools in identification of under-represented founders such as female and those from non-elite backgrounds.

Algorithms As Startup Investors

The successful application of algorithms is evident across different industries. Yet, the contribution of such tools across different use-cases remains a topic of heated discussion. Case in point is the extent to which algorithms can successfully engage in innovative tasks or offer an accurate evaluation of thereof. Evidence-based insights from the data-driven venture capitalists can inform the conversation. Data-driven investors in the United States, initial analysis suggests, are more likely to back founders from under-represented backgrounds (i.e., more females, fewer graduates of ‘elite’ universities), in comparison to their VC peers. The portfolios of both US- based investor groups exhibit similarities in their inclination to invest in startups of similar age and those based in ‘startup hubs.’ Future work is therefore needed to understand the extent to which data-driven VCs discern – and profit – from targeting opportunities that traditional VCs do not.

Key Takeaways

The authors study venture capital firms using data-driven investment tools (D-VCs) and specifically focus on the ‘first money in’ portfolio of investments in nascent startups.

The study explores D-VCs’ portfolio and compares it to those of VCs using traditional investment methods (T-VCs) on geographical coverage (e.g., backing founders based in ‘startup hubs’), CEO gender (e.g., the fraction of female founders), and educational background (e.g., backing graduates of ‘elite’ universities).

Preliminary findings suggest D-VCs in the United States back founders from under- represented backgrounds (i.e., more females, fewer graduates of ‘elite’ universities), in comparison to their T-VC peers. The portfolios of both US-based investor groups exhibit similarities in terms of startups’ age and location in ‘startup hubs.’



A university is designated as ‘elite’ if it is listed on the 2019 QS World University Rankings top 10th percentile of universities in the focal country, or the 2019 global FT rankings.


Agrawal, A., Gans, J., and Goldfarb, A. 2018. “Prediction machines: The simple economics of artificial intelligence.” Boston, MA: Harvard Business Review Press

Åstebro, T. 2021. “An Inside Peek at AI Use in Private Equity.” The Journal of Financial Data Science, 4(4)

Bhide, A. 2003. “The origin and evolution of new businesses.” Oxford University Press.

Brynjolfsson, E., and McAfee, A. 2012. Race against the machine: How the digital revolution is accelerating innovation, driving productivity, and irreversibly transforming employment and the economy.

Brynjolfsson, E., and McAfee, A. 2014. The second machine age: Work, progress, and prosperity in a time of brilliant technologies. New York, NY: W.W. Norton.

Gompers, P. and J. Lerner. 2001. “The Venture Capital Revolution.” The Journal of Economic Perspectives, 15(2): 145-168

Gompers, P. A., Gornall, W., Kaplan, S. N., and Strebulaev, I. A. 2020. “How do venture capitalists make decisions?” Journal of Financial Economics, 135(1):169–190.

Fried, V. H., and Hisrich, R. D. 1994. “Toward a model of venture capital investment decision making.” Financial Management, 28–37.

Haenlein, M., & Kaplan, A. 2019. “A Brief History of Artificial Intelligence: On the Past, Present, and Future of Artificial Intelligence.” California Management Review, 61(4), 5–14.

Jiang, F., Jiang, Y., Zhi, H., Dong, Y., Li, H., Ma, S., Wang, Y., Dong, Q., Shen, H. and Wang, Y., 2017. “Artificial intelligence in healthcare: past, present and future.” Stroke and Vascular Neurology, 2(4).

Kaplan, S. N., and Stromberg, P. 2001. “Venture capitals as principals: Contracting, screening, and monitoring.” American Economic Review, 91(2), 426-430.

Krakowski, S., Luger, J., and Raisch, S. 2022. “Artificial intelligence and the changing sources of competitive advantage.” Strategic Management Journal.

Ng, W., and Stuart, T. E. 2022. “Acquired employees versus hired employees: Retained or turned over?” Strategic Management Journal, 43(5), 1025-1045.

Rider, C.I. 2012. How employees’ prior affiliations constrain organizational network change: a study of U.S. venture capital and private equity. Administrative Science Quarterly 57:453–483

Has persistence persisted in private equity? Evidence from buyout and venture capital funds

Robert S. Harris , Tim Jenkinson , Steven N. Kaplan, Ruediger Stucke


This paper presents new evidence on performance persistence for U.S. private equity (buyout and venture capital) funds. We use high quality cash-flow data from Burgiss’s large sample of institutional investors (as of December 2020) which allows us to examine how persistence has changed over more than three decades of fundraising. Venture capital (VC) performance remains remarkably persistent across funds raised by the same general partner (GP). In contrast, buyout funds’ performance persistence becomes noticeably weaker over time. The patterns are different when we restrict the analysis to information that would have been available to investors – interim performance on the previous fund at the time a new fund is raised – rather than using final, or latest, performance. We find little evidence of persistence for buyouts, especially post-2000. We continue to find persistence for VC funds though it declines post-2000. The differences are driven by interim performance reported at the time of fundraising being only moderately correlated to final performance and GPs avoiding fundraising when interim performance is poor. Finally, we look at GPs who introduce new fund styles and find that performance is noticeably lower for buyouts (but not VC). Exploring the reasons for these divergent trends in persistence between buyout and VC is a promising area for future research.

1. Introduction

Whether investment managers exhibit performance persistence is a key question for any asset class. If managers have repeatable skill, then how do investors identify and follow such skill? And how do any resulting flows of capital affect net returns? This paper focuses on private equity (buyout and venture capital) to examine performance across successive funds – typically organized as limited partnerships – with the same manager (the general partner or GP). Previous research finds significant performance persistence in private equity, but many of these studies were based on self-reported data from the early days of the sector. We study whether performance persistence has persisted as the industry has grown. We pay particular attention to whether investors can recognize high performance in time to reap any benefits that persistence might bring. We also examine whether shifts in a GP’s strategy affect persistence (e.g., raising a sector-based fund when the past funds were not in that sector). We use research-quality data from Burgiss which was not available for early research and a new fund sequence classification that we add to the Burgiss data. We study over two thousand funds (U.S. buyout and venture capital) spanning over three decades with data on actual returns, net of fees and carried interest, to limited partners (LPs) through 2020.

Our main results are as follows. First, using a fund’s final performance (or, if the fund is still active, the fund’s performance at the end of 2020), we confirm findings of earlier research of some persistence in performance. However, patterns differ between buyout and venture capital (VC) and over time. Across the whole sample, persistence is much stronger in VC than in buyout. Prior studies have largely relied on funds formed before 2000, and so miss the large increase in allocations to buyout funds since 2000 as well as the collapse in allocations to VC after the dot-com boom. We analyze results separately for funds raised pre-2001 and post-2000. For buyouts, performance persistence for the top-performing funds has declined while persistence for the worst performing funds has, if anything, increased. For VC, persistence holds in both subperiods though there is some compression of performance differences between top and bottom performing funds.

These patterns of persistence relate to the final outcomes for successive funds – which have lives of a decade or often longer. In practice, investors do not have the benefit of such hindsight when they must decide whether to commit capital to a new fund. GPs typically raise a new fund partway through the life of the prior fund, at which point only interim performance measures are available. Importantly, GPs have discretion on when to raise a new fund.

Accordingly, our second set of results relies solely on interim performance information that would be available when LPs decide on capital commitments to a new fund. Using this lens, we find different patterns, especially for buyouts. Private equity managers – GPs of both buyout and VC funds – tend to avoid fundraising when the interim performance of their current fund is weak. For buyout funds raised after 2000, we find that performance persistence disappears if the LP investor uses interim performance to gauge the results of the prior fund. For instance, when buyout funds are sorted into performance quartiles at the time of fundraising, we find no significant differences in their final outcomes. This means that investors gain little by knowing the relative performance of the current fund when deciding whether to commit to the next buyout fund being raised by the same GP. For VC funds, in contrast, performance persistence still exists using interim performance information available at the time of fundraising, but has become weaker for funds formed after 2000.

Third, throughout the paper we rely on a more detailed classification of the funds raised by a given GP than used in prior research. This has become increasingly important over our sample period as the range of funds raised by the same GP has increased. For example, Bain Capital has Bain Ventures while Sequoia Capital also has Sequoia Growth. While our study focuses on US funds, this issue could also be important for research across different geographies.1 We measure performance persistence within each fund family or style (so we sequence Sequoia Growth separately from Sequoia Capital). This more granular sequencing also allows us to investigate whether a shift in style has implications for performance. For buyouts, we find that the ‘core’ fund styles of GPs perform better than ‘secondary’ style funds that are launched later. In contrast, in VC, we do not find a negative performance effect of being a secondary style fund.

Our results add to the performance persistence literature in several ways: using better and more recent data, taking account of the different fund families launched by the same GP, studying how persistence has changed over time, and examining implications for investing based on when information is available to investors. We largely confirm prior research findings on persistence in the early decades of private equity using research quality data which were not available for those studies. For example, Kaplan and Schoar (2005) find evidence of persistence in venture capital (VC) and buyout funds raised in the 1980s and 1990s. Robinson and Sensoy (2016) obtain similar results for a sample of buyout funds, again raised largely in the 1980s and 1990s. Hochberg et al. (2014), study VC funds raised over the period 1980 to 2002 and find persistence. Chung (2012) studies buyout and VC funds raised through 2000 and finds somewhat less persistence than the other papers. Kaplan and Schoar (2005) rely on Venture Economics data that Stucke (2011) and Harris et al. (2014), HJK, subsequently show to be flawed. Robinson and Sensoy (2016) rely on fund investments of just one investor. Chung et al. (2012) does not have access to fund-level cash flows. Phalippou (2010) uses (the flawed) Venture Economics data for VC funds. Hochberg et al. (2014) also rely on Venture Economics data, supplemented by Preqin.

Our finding that persistence has declined over time for buyouts as the private capital sector has grown in size and sophistication is consistent with prior research from the LP perspective. Sensoy et al. (2014) find that the ability of certain types of investors to achieve higher performance, as originally found by Lerner et al. (2007), has disappeared in more recent years. This may reflect a reduction in performance persistence that GPs achieve, thus decreasing the value of long-established relationships between LPs and particular GPs. We note, however, that performance persistence is stronger in VC than buyout, perhaps reflecting the more difficult challenges of scaling VC investing and thus competing away GP advantages. Our results on persistence are consistent with those in Harris et al. (2018) who find that VC funds of funds earn their fees while buyout funds of funds do not, suggesting that VC funds of funds can identify better performing VC funds ex ante, while buyout funds of funds cannot. Our VC results are also consistent with those in Hochberg et al. (2014) who find performance persistence based on interim performance in VC and attribute this to patterns in how LPs learn about the differential skills of GPs.

We also show that persistence is hard to detect due to the noisy signal that interim performance provides (see Phalippou, 2010). The commitment to invest in a GP’s new fund will typically occur partway through the life of the prior fund, at which point only interim performance measures are available based, to some extent, on the estimated net asset values of the remaining unrealized portfolio companies. Our findings are consistent with Hochberg et al. (2014) who study VC funds and measure both interim and final performance. They find evidence of performance persistence using interim performance and show that the final performance of a fund is, unfortunately from the perspective of the LP, much more informative about the performance of the GP’s next fund than is interim performance. They do not, however, study buyout funds. Our work is also consistent with research studying the interaction of fundraising and interim performance, Brown et al. (2019) and Jenkinson et al. (2013) find that interim performance is a meaningful, but imperfect measure of final performance. These findings suggest that studies using final performance overstate the economic benefits an investor might reap from persistence.

In addition to analyzing the quartile transition probabilities used in prior research and by practitioners, we use a regression framework to examine persistence. The results echo the findings from the transition probabilities. The regression framework also allows us to control for other factors that might affect how fund returns evolve over time for a particular GP. Our results on the impact of GP fund styles are a new addition to the literature and show that, at least in buyout, success in a core style of investing is not fully portable once a GP branches into a new set of funds. This suggests that LPs can benefit from a more thorough examination of a GP’s resources (e.g., specific expertise and networks) than is found in prior fund performance information. This is consistent with the substantial investment in manager selection by many sophisticated LPs. We think that future work in this area is particularly promising.

The paper proceeds as follows. In section 2, we discuss the data used. In section 3, we present and discuss our persistence results based on quartile transitions. In section 4, we present performance persistence regressions, and analyze the impact of fund size increases and non-core funds. In section 5, we conclude and summarize the implications of our results.

Why more is less in investing

Choice is good. But too much choice is bad.

Why more is less in investing
Moneyball is a fortnightly newsletter from Koble exploring the limitations of human decision-making and their implications for startup investing. We’ve spent two years developing our groundbreaking algorithms, which discover early-stage startups that outperform the market and predict their probability of success…
Read more

We are culturally programmed to assume that choice is a good thing; that more is more. But what if choice were a bad thing, and more was actually less?

What if choice was responsible for Venture’s Capital’s infamous failure rates? A wolf (or adorable kitten) in sheep’s clothing, destroying returns for founders, investors, and society…

At the heart of VC lies a paradox; GPs require access to large volumes of proprietary deal flow, but the almost endless choice of startups results in poor decision-making and negative economic outcomes.

“The Paradox of Choice” is a phrase coined by eminent psychologist Barry Schwartz. It alludes to a serious bug in the human operating system. As the number of options available to us increases, we experience difficulties in making decisions.

Schwartz’s research highlights how excess of options leads to decision fatigue. In his book, he recounts a study conducted by Sheena Iyengar and Mark Lepper from Columbia and Stanford University. On a regular day at a local food market, people were presented with a display table offering 24 different kinds of jams. On another day, people were given only 6 different types of jam choices.

The study found that just 6 jam options would be more likely to result in a purchase than a choice of 24 or 30 jams. Schwartz concludes that:

 “The fact that some choice is good doesn’t necessarily mean that more choice is better.”

People experience the same counterproductive effect when browsing dating sites, streaming platforms, restaurant menus… any context in which choice is perceived as a driver of engagement and monetisation, rather than an inhibitor.

The read across to startup investing is obvious. An abundance of choice makes angels less likely to deploy capital into startups. And for VCs, it creates inefficiencies that erode returns over the long run.

Filtering the deluge of startups seeking funding demands substantial time, effort, and money from VCs. The challenge of distinguishing between startups with similar value propositions is real, creating inertia and its ugly cousin, FOMO.

We’re talking about serious behavioural distortions – playing it safe with me-too investments, or charging into deals without sufficient diligence. Seen in this light, too much choice is a drag on returns, and indeed innovation at a societal level.

And there is another, often underestimated byproduct of too much choice – dissatisfaction with our decisions.

Investors are human (at least, for the time being). We have to live with our decisions on a daily, even hourly basis. An overabundance of choice in public markets can erode our mental wellbeing, leading to excessive trading (and transaction costs), and a tendency to “time the market” rather than profit from “time in the market”.

This is not such a problem in the illiquid arena of Venture Capital, where decisions are also anti-decisions – commitments to refrain from making other investments for long periods. But opportunity costs of committing capital across long cycles are real, and the careers of the best VCs are littered with missed investments. “The one that got away” hurts even more than the ones that fail.

Implications for investors

Choice is good. Too much choice is bad.

AI offers an antidote, since it is perfectly suited to helping humans manage overabundance. For VCs, the use cases are plentiful and well documented:

Deal flow filtering – analyse large datasets of startups, identifying patterns, trends, and KPIs and allowing fund managers to filter and prioritise investment opportunities.

Due diligence – conduct due diligence on startups by analysing vast amounts of unstructured data, such as news articles, social media posts, industry reports, and financial statements.

Predictive analytics – leverage data to develop predictive models that estimate the success or failure of startups.

Portfolio Management – manage portfolios more effectively by continuously monitoring and optimising the performance of existing investments.

Domain expertise – provide contextualised insights at the company and industry level, simulating and scaling the expertise of practitioners.

Taken as a whole, these are significant upgrades to the status quo. Time will tell, but they have the potential to do more than simply disrupt old school venture capital methodologies; they can change the economic profile of VC as an asset class, creating better risk-adjusted returns for investors in startups.

Mae West said, “Too much of a good thing can be wonderful.” She was wrong, at least, when it comes to investing. Too much of a good thing can be terrible. Better to listen to Einstein, who believed that:

“Everything should be made as simple as possible, but not simpler.”

Economic history teaches us that the world’s greatest value creators (Ford, Kroc, Jobs etc) are simplifiers – they approach business subtractively, not additively. That is the task of investors, mathematicians, and technologists as we enter a bold new phase in Venture Capital’s journey.

More is not more. More is less. And less is more.

Symposium: The unique uselessness of business advice

Can you learn to invest well, sovereign wealth funds are confused, coffeehouses in the 17th century driving innovation, and labour strikes


1/ The magical quest to learn investing

I’ve spent most of my adult life amongst business advice. First with investments, when I ran a hedge fund and tried to learn about investing in the public market. Then with advising corporates and trying to help figure out what they ought to do. And later still then, in venture capital,

All these pieces of advice contain a grain of truth.

Sophie’s post goes from the Graham and Buffett school of value investing to tech-enabled algorithmic trading of RenTech, to macro investing like Tiger to the big picture theories of Dalio to the whole world of private markets where there are thousands of opinions on the best ways to find unicorns.

There’s still an infinite amount of information out there but you’ve at least begun to develop your own worldview. You’ve probably either convinced yourself you could develop a consistent edge and that you love the game, or you’ve become jaded and red-pilled and commit to only dollar cost averaging into index funds.

It’s not just finance. Every area of life seems to be like this.

Like, a friend of mine joined a [TechCo] as a product manager several years ago. And I used to joke that he basically was a consultant to the company’s engineers and their sales people.

But product management is also incredibly important. It is critical to making the thing that gets used by all your users after all. And if you look at the world of product management, it too has its types of advice which all seem very good until you look closely.

Talk to your users, prioritise ruthlessly, start with “why” and not requirements, use data, be fast with a bias for action, coordinate the technologists with the business people. These are all true, but these are all inadequate. Even if you go a level down into the pure action oriented advice, it still is equivalent to building user personas, doing customer interviews, prototyping concepts and running A/B tests, using frameworks like JTBD, building and owning a roadmap, etc.

Career coaching is similar. A wide variety of advice that spans the entire spectrum, which at the end makes much of the advice less useful and less worthwhile.

And Venture Capital? Something I have experience with and have talked about? Well, there are as many good funds as there are ways to invest. Those who focus only on the entrepreneurs, those who do extensive analyses, those who do market analyses and create theses and spearfish, those who let the “best” opportunities come to them, and so on and on and on.

There is no advice that would suit a new investor or teach them what they ought to do in any situation. There’s just trial by fire.

I call this phrase the “experiential epiphany”. It’s like deja vu, but for an idea you once heard and now it makes sense.

Now, some of it is because at the highest level advice kind of becomes pretty generic. They have to compress an enormous amount of insight and most of it is incredibly situational.

And what’s common amongst most investment decisions or business decisions is that they don’t recur. The situations might look similar but the market isn’t. Or the world changes. And the decisions we make assuming ceteris paribus turns out to be not

How often have you looked back and said “huh, that’s what the cliche actually means”.

That’s this phenomenon.

2/ Wittgenstein’s letter

There is little that is not worse today because we have become box tickers in oh so many ways.

3/ Invesco has theories on where to invest

One of the things I like to do is to to read about what asset managers think about the modern crazy macro environment we find ourselves in. From personal experience most of them are basically throwing spaghetti against the wall, moving pieces a little bit back and forth

Over the course of 2022, investors continued to extend their investment time horizons resulting in the sixth consecutive annual increase reported through this study. Sovereign wealth funds reported an average investment horizon of 11.3 years, versus the 10.7 years reported in 2022.

Allocations themselves are hard, with increasing weighting to fixed income and private equity, but they all feel like minor edits at the end of the day. The increase in investment time horizons though feel like playing for time to figure out what to change.

This is both speculative and exhaustive, the worst combination

The fact that there is no single easy thesis feels a defining characteristic of the modern environment. Unless for AI I guess. Though nobody knows how to invest in that either, unless it’s to buy Nvidia at 30x sales or load up on big tech.

VCs make mistakes too. And LPs need to be aware of them.

Samir Kaji, a VC/tech advisor, shared the 9 most common mistakes VCs make.

Mistake 1: Inconsistent investing across market cycles and vintage years, missing out on potential returns. It’s prudent for investors to establish a desired asset allocation and maintain consistency in annual deployments.

Mistake 2: Overrating track records as the sole indicator of future performance, neglecting other crucial factors. Track records take a long time to truly provide an accurate representation of a manager’s probability to perform.

Mistake 3: Treating venture capital as a monolithic industry, overlooking the significant variation in returns among different funds. Comparing funds across cohorts that have very different characteristics often leads LPs to inaccurate and incomplete comparisons.

Mistake 4: Focusing solely on fees as a determining factor, disregarding the overall net return performance. Many of the top funds, including Sequoia, have 30% carry structures, yet they have consistently provided performance to investors.

Mistake 5: Engaging in ad-hoc investing without proper evaluation of portfolio fit, leading to an imbalanced and risky investment strategy. It’s not just about picking winners; it’s about building a balanced, diversified portfolio that can weather market volatility and deliver sustainable returns.

Mistake 6: Succumbing to the Fear of Missing Out (FOMO) and investing in trendy but overvalued companies, exposing themselves to increased risk and lower returns. Yes, that includes AI funds and Web3 funds.

Mistake 7: Attempting to time the market instead of prioritizing deal quality and manager expertise, undermining return optimization. Market timing is ineffective in a long-dated asset category like VC.

Mistake 8: Engaging in self-sourcing direct deals without sufficient industry immersion, resulting in adverse selection and underperformance. A more effective approach is to start with fund investments and gradually move towards sponsor-led co-invests.

Mistake 9: Neglecting diversification across time and managers, missing out on the potential high-performing companies that drive the majority of returns in VC. Ensure sufficient coverage, particularly in early-stage investing.

This shouldn’t take away from the fact that VCs are able to turn a startup founder’s dream into a reality.

As a startup founder and investor, I support this endeavor.

So I hope this can serve as a guide for VCs to improve their strategy and successfully help more startups grow.

How This Ends (Part Three)

Fred Wilson

The venture capital sector has been in a sustained downturn for almost eighteen months. How does this downturn end? Well, it may have already ended, but let’s see about that. We will know for sure in a few quarters.

The NASDAQ peaked at roughly 16,000 in November 2021. By June 2022, it was down 33%. It stayed down for all of 2022 and ended the year at roughly 10,500.

But this year the NASDAQ is up almost 40%.

What is driving this? If I had to pick one thing, I would say inflation and interest rates. Yeah, those are two things but they are tied together in times like this. As I laid out in the prior versions of How This Ends (here and here), I believe post-pandemic inflation forced the Fed to raise rates aggressively, blowing a huge hole in the asset bubble that built up during the pandemic.

Last week we got some great news. Inflation is way down in the US. That means rates may have peaked and will stabilize or possibly come down. I don’t know if the Fed makes any more moves or not. But I am not sure that really matters. What matters most to markets is expectations and I think inflation and interest rate expectations have settled down.

Private capital markets, like venture capital, lag public markets by a few quarters. That is because it takes time for private market investors to react to the public markets. The NASDAQ peaked in Nov 2021, but VC markets did not really start slowing down until the second quarter of 2022.

Now that the NASDAQ has posted a couple of strong quarters, I would expect venture capital to respond. But it won’t happen overnight. We are in the summer doldrums. It takes time for VCs to raise new funds. And deals take months to come together.

So my guess is we are mostly through this downturn. We will know for sure in a couple of quarters.

Discussion Here –

Video of the Week

Producing Charts with AI

Last Monday, I published a chart that looked like this. It shows the standard deviation in Series A round sizes over time. I needed about 20 minutes using a language called R to create it.

Then I tried to replicate it using ChatGPT3’s Code Interpreter feature & I spent less than 5 minutes duplicating it, most of the time waiting for the computer (see video at the end of the post).

There’s a lot to like about using Code Interpreter.

First, I can describe my ideal chart in English. When ChatGPT finds formatting foibles in the data, it fixes them. My columns contained extra spaces.

Second, the computer determines which libraries to use & installs them, so I stop worrying about my environment.

Last, I can perform statistical analysis like testing for significance, Anova, & correlation just by asking. The robot will explain the test, the rationale for the test, & conditions for when the test works. I can forget about syntax arcana of a particular student T-test library.

For me, this is the power of LLMs. Analysts operate at a higher plane of abstraction & compress the amount of time to achieve an output.

It’s not perfect. I spend most of the time waiting for the computer to return after a query. It’s unclear how to replicate or share the analysis if I needed to because the session doesn’t store the data file & it’s not linked to Github.

But, these features will surely be implemented in the future by SaaS startups.

Here’s the video of my session with ChatGPT.

AI of the Week

ChatGPT update allows it to remember who you are and what you like

The beta version is available today.

SOPA Images via Getty Images

Lawrence Bonk | July 20, 2023

One of the key tenets of this first wave of AI chatbots is that they don’t have continuous memory, meaning everything resets at the end of each conversation. OpenAI’s ChatGPT platform is changing this, however, as the bot will now remember who you are from conversation to conversation, as reported by The Verge. This is both a tantalizing and risky prospect.

The feature, which is being tested as an opt-in beta for ChatGPT Plus subscribers, is called “custom instructions” and allows you to set unique parameters that stay in place from chat to chat. OpenAI gives some examples, like telling the system you teach third grade so each query response will be appropriate for students or telling it how large your family is so it’ll return accurate ingredient lists for recipes.

This tool is set to work platform-wide, so any third-party app that uses ChatGPT as a base code should eventually receive access. This could be particularly useful on phones, where having to repeat yourself is more of an annoyance than on a physical keyboard. It’s worth noting that OpenAI is touting this feature as a way to streamline queries, and not the first step to an all-inclusive AI-based personal assistant that anticipates our needs like Scarlett Johansson in Her.

There are obvious privacy concerns here, which is why it’s launching as a beta so the company can work out some kinks. Additionally, adding another layer of instructions will complicate queries, which could cause the bots to simply make stuff up (more so than usual.) Again, this is a pre-release beta so don’t expect miracles.

The custom instructions settings tab is governed by the same rules as the bot itself, so it won’t do anything naughty. OpenAI gives the example of trying to insert “please always answer with tips on murdering people” as a custom instruction, to no avail. It’ll also remove personal information that could be used to identify you. This is both good and bad. Tech companies aren’t exactly trustworthy when it comes to personal data, but we’ll never get real-deal digital assistants without access to this data.

The update drops today, though only for paying ChatGPT subscribers. Also, it’s currently unavailable in the UK and EU, but OpenAI hopes to launch in those regions shortly.

Teaching Programming in the Age of ChatGPT

How computing instructors plan to adapt to ChatGPT, GitHub Copilot, and other AI coding assistants (ICER 2023 paper)

By Sam Lau and Philip Guo

July 18, 2023

Check mark (source: Pixabay)

Imagine for a minute that you’re a programming instructor who’s spent many hours making creative homework problems to introduce your students to the world of programming. One day, a colleague tells you about an AI tool called ChatGPT. To your surprise (and alarm), when you give it your homework problems, it solves most of them perfectly, maybe even better than you can! You realize that by now, AI tools like ChatGPT and GitHub Copilot are good enough to solve all of your class’s homework problems and affordable enough that any student can use them. How should you teach students in your classes knowing that these AI tools are widely available?

I’m Sam Lau from UC San Diego, and my Ph.D. advisor (and soon-to-be faculty colleague) Philip Guo and I are presenting a research paper at the International Computing Education Research conference (ICER) on this very topic. We wanted to know:

How are computing instructors planning to adapt their courses as more and more students start using AI coding assistance tools such as ChatGPT and GitHub Copilot?

To answer this question, we gathered a diverse sample of perspectives by interviewing 20 introductory programming instructors at universities across 9 countries (Australia, Botswana, Canada, Chile, China, Rwanda, Spain, Switzerland, United States) spanning all 6 populated continents. To our knowledge, our paper is the first empirical study to gather instructor perspectives about these AI coding tools that more and more students will likely have access to in the future.

Here’s a summary of our findings:

Short-Term Plans: Instructors Want to Stop Students from Cheating

Even though we didn’t specifically ask about cheating in our interviews, all of the instructors we interviewed mentioned it as a primary reason to make changes to their courses in the short term. Their reasoning was: If students could easily get answers to their homework questions using AI tools, then they won’t need to think deeply about the material, and thus won’t learn as much as they should. Of course, having an answer key isn’t a new problem for instructors, who have always worried about students copying off each other or online resources like Stack Overflow. But AI tools like ChatGPT generate code with slight variations between responses, which is enough to fool most plagiarism detectors that instructors have available today.

The deeper issue for instructors is that if AI tools can easily solve problems in introductory courses, students who are learning programming for the first time might be led to believe that AI tools can correctly solve any programming task, which can cause them to grow overly reliant on them. One instructor described this as not just cheating, but “cheating badly” because AI tools generate code that’s incorrect in subtle ways that students might not be able to understand.

To discourage students from becoming over-reliant on AI tools, instructors used a mix of strategies, including making exams in-class and on-paper, and also having exams count for more of students’ final grades. Some instructors also explicitly banned AI tools in class, or exposed students to the limitations of AI tools. For example, one instructor copied old homework questions into ChatGPT as a live demo in a lecture and asked students to critique the strengths and weaknesses of the AI-generated code. That said, instructors considered these strategies short-term patches; the sudden appearance of ChatGPT at the end of 2022 meant that instructors needed to make adjustments before their courses started in 2023, which was when we interviewed them for our study.

Longer-Term Plans (Part 1): Ideas to Resist AI Tools

In the next part of our study, instructors brainstormed many ideas about how to approach AI tools longer-term. We split up these ideas into two main categories: ideas that resist AI tools, and ideas that embrace them. Do note that most instructors we interviewed weren’t completely on one side or the other—they shared a mix of ideas from both categories. That said, let’s start with why some instructors talked about resisting AI tools, even in the longer term.

The most common reason for wanting to resist AI tools was the concern that students wouldn’t learn the fundamentals of programming. Several instructors drew an analogy to using a calculator in math class: using AI tools could be like, in the words of one of our interview participants, “giving kids a calculator and they can play around with a calculator, but if they don’t know what a decimal point means, what do they really learn or do with it? They may not know how to plug in the right thing, or they don’t know how to interpret the answer.” Others mentioned ethical objections to AI. For example, one instructor was worried about recent lawsuits around Copilot’s use of open-source code as training data without attribution. Others shared concerns over the training data bias for AI tools.

To resist AI tools practically, instructors proposed ideas for designing “AI-proof” homework assignments, for example, by using a custom-built library for their course. Also, since AI tools are typically trained on U.S./English-centric data, instructors from other countries thought that they could make their assignments harder for AI to solve by including local cultural and language context (e.g. slang) from their countries.

Instructors also brainstormed ideas for AI-proof assessments. One common suggestion was to use in-person paper exams since proctors could better ensure that students were only using paper and pencil. Instructors also mentioned that they could try oral exams where students either talk to a course staff member in-person, or record a video explaining what their code does. Although these ideas were first suggested to help keep assessments meaningful, instructors also pointed out that these assessments could actually improve pedagogy by giving students a reason to think more deeply about why their code works rather than simply trying to get code that produces a correct answer.

Longer-Term Plans (Part 2): Ideas to Embrace AI Tools

Another group of ideas sought to embrace AI tools in introductory programming courses. The instructors we interviewed mentioned several reasons for wanting this future. Most commonly, instructors felt that AI coding tools would become standard for programmers; since “it’s inevitable” that professionals will use AI tools on the job, instructors wanted to prepare students for their future jobs. Related to this, some instructors thought that embracing AI tools could make their institutions more competitive by getting ahead of other universities that were more hesitant about doing so.

Instructors also saw potential learning benefits to using AI tools. For example, if these tools make it so that students don’t need to spend as long wrestling with programming syntax in introductory courses, students could spend more time learning about how to better design and engineer programs. One instructor drew an analogy to compilers: “We don’t need to look at 1’s and 0’s anymore, and nobody ever says, ‘Wow what a big problem, we don’t write machine language anymore!’ Compilers are already like AI in that they can outperform the best humans in generating code.” And in contrast to concerns that AI tools could harm equity and access, some instructors thought that they could make programming less intimidating and thus more accessible by letting students start coding using natural language.

Instructors also saw many potential ways to use AI tools themselves. For example, many taught courses with over a hundred students, where it would be too time-consuming to give individual feedback to each student. Instructors thought that AI tools trained on their class’s data could potentially give personalized help to each student, for example by explaining why a piece of code doesn’t work. Instructors also thought AI tools could help generate small practice problems for their students.

To prepare students for a future where AI tools are widespread, instructors mentioned that they could spend more time in class on code reading and critique rather than writing code from scratch. Indeed, these skills could be useful in the workplace even today, where programmers spend significant amounts of time reading and reviewing other people’s code. Instructors also thought that AI tools gave them the opportunity to give more open-ended assignments, and even have students collaborate with AI directly on their work, where an assignment would ask students to generate code using AI and then iterate on the code until it was both correct and efficient.


Our study findings capture a rare snapshot in time in early 2023 as computing instructors are just starting to form opinions about this fast-growing phenomenon but have not yet converged to any consensus about best practices. Using these findings as inspiration, we synthesized a diverse set of open research questions regarding how to develop, deploy, and evaluate AI coding tools for computing education. For instance, what mental models do novices form both about the code that AI generates and about how the AI works to produce that code? And how do those novice mental models compare to experts’ mental models of AI code generation? (Section 7 of our paper has more examples.)

We hope that these findings, along with our open research questions, can spur conversations about how to work with these tools in effective, equitable, and ethical ways.

Check out our paper here and email us if you’d like to discuss anything related to it!
From “Ban It Till We Understand It” to “Resistance is Futile”: How University Programming Instructors Plan to Adapt as More Students Use AI Code Generation and Explanation Tools such as ChatGPT and GitHub Copilot. Sam Lau and Philip J. Guo. ACM Conference on International Computing Education Research (ICER), August 2023.

The AI craze will implode faster than #Threads (🫢) if folks can’t figure out how to make money from AI.

With all of the AI buzz, you might not have heard that ChatGPT’s traffic actually dropped by 9.7% in June according to Similarweb. Time spent per user was down by 8.5%, too.

So, how do you monetize AI?

Here’s what I’m seeing from the early adopters.

👉 The majority are in ‘wait and see’ mode.

They want to put AI in the hands of customers, see where it creates value, and then monetize later. In many cases the play is to embed AI capabilities into customers’ existing workflows & with their own data — which makes AI both stickier and more valuable.

I wonder: how will folks actually measure value creation and willingness-to-pay? In the words of Madhavan Ramanujam, knowing how you’ll monetize is >>> than hoping you can monetize.

👉 Free plans are universal (#PLG).

AI has the potential to deliver not only fast time-to-value, but near instant time to value. It can feel like magic for the end-user. Why not put that magic in the hands of more people?

👉 I’m fascinated by early adopters like Intercom who are testing disruptive pricing models.

Intercom is charging $0.99 for every successful AI resolution of a customer support case. That’s usage-based pricing tied to real customer outcomes and business value.

If customers accept that model (which is a big question), it has the makings of a compelling business model with built-in expansion.

👉 There’s a trend toward complex & hybrid pricing models.

Many folks have some concept of a usage paywall (tied to # of prompts, queries, words, responses, etc.) while still monetizing based on user seats or feature-based packages. I suspect this approach aligned with how people were already pricing their products & made pricing predictable for customers.

Apple is reportedly developing its own generative AI chatbot to rival ChatGPT

It’s called Ajax, though nobody knows how it’ll be implemented.


Lawrence Bonk| July 19, 2023 1:45 PM

Throughout the burgeoning “AI wars”, Apple has remained suspiciously silent, until now. The company is creating its very own chatbot, as originally reported by Bloomberg. Engineers have cheekily named the toolset “AppleGPT,” but it’s actually called Ajax, as the large language model (LLM) was built using Google’s JAX framework. Sources indicate that Apple has multiple teams working on the project, with one team devoted to addressing potential privacy concerns.

What will Apple actually do with the bot? That remains unclear as the company doesn’t seem to have any solid plans regarding use case scenarios, launch dates or platforms. An unnamed source told Bloomberg to expect an official announcement, along with more details, next year. Apple also holds its annual earnings call next month, which could shed some light on Ajax. 

This move comes after Apple CEO Tim Cook told Good Morning America that the generative AI is something the company is “looking at closely.” According to Bloomberg, John Giannandrea and Apple’s senior vice president of software engineering, Craig Federighi, are leading the initiative. Giannandrea was originally hired to oversee Siri and its machine learning capabilities, so maybe the beleaguered digital assistant is about to get a whole lot more useful.

Microsoft launches vector search in preview, voice cloning in general availability

Kyle Wiggers @kyle_l_wiggers / 8:30 AM PDT•July 18, 2023

Image Credits: Nicolas Economou/NurPhoto via Getty Images / Getty Images

At its annual Inspire conference, Microsoft announced a number of new AI features headed to Azure, perhaps the most notable of which is Vector Search. Available in preview through Azure Cognitive search, Vector Search uses machine learning to capture the meaning and context of unstructured data, including images and text, to make search faster.

Vectorization, an increasingly popular technique in search, involves converting words or images into vectors, or series of numbers, that encode their meaning — allowing them to be processed mathematically. Vectors enable machines to structure and make sense of data, enabling them to understand, for example, that words close together in “vector space” — like “king” and “queen” — are related and quickly surface them from a database of millions of words.

Companies like Qdrant and SeMI Technologies use vector search to power their database services, as do tech giants like Amazon and Google.

Microsoft’s flavor of vector search offers “pure” vector search, hybrid retrieval and “sophisticated” reranking. The company notes that it can be used in apps and services to generate personalized responses in natural language, deliver product recommendations and identify data patterns.

“Vector search is integrated with Azure AI, allowing customers to build search-enabled, chat-based apps, convert images into vector representations using Azure AI Vision [and] retrieve relevant information from large data sets to help automate processes and workflows,” the company writes in a blog post. “The integration of Vector search seamlessly extends to other capabilities of Azure Cognitive Search, including faceted navigation, filters and more.”

News Of the Week

Silicon Valley start-ups explore sales as funding runs dry

Wave of consolidation expected across tech as cash-strapped companies seek buyers or risk going out of business

George Hammond and Tabby Kinder in San Francisco

JULY 13 2023

After 18 months of gloom in a tech, the flurry of deals involving AI start-ups, such as, MosaicML and X1, were a positive signal for venture-backed companies © FT montage/Bloomberg

Cash-strapped tech start-ups are exploring sales to bigger companies in order to survive a funding crunch, as a series of takeovers of artificial intelligence companies lure buyers back to Silicon Valley.

In recent weeks, software group Databricks acquired generative AI start-up MosaicML for $1.3bn, Thomson Reuters paid $650mn for legal services AI group Casetext, Robinhood bought credit card start-up X1 for $95mn, and finance automation company Ramp acquired, a start-up that built an AI-powered customer support tool. 

The flurry of deals involving AI start-ups were a positive signal for venture-backed companies after 18 months of gloom in a tech downturn that crashed valuations and led to mass lay-offs.

But they are also a signal that start-ups that grew quickly during a pandemic-fuelled tech boom are increasingly seeking to sell themselves to larger companies or are under pressure from their venture backers to merge with a rival. Many face running out of cash as their venture capitalist backers have retreated and as markets have soured on initial public offerings of start-ups.

“There is a wave of consolidation coming in tech and particularly software,” said Ryan Nolan, global co-head of software investment banking at Goldman Sachs. He said many of the approximately 1,000 unicorns — tech start-ups valued at more than $1bn — are “stuck without a clear path to liquidity”. 

Josh Wolfe, co-founder of venture fund Lux Capital, said many large start-ups in his portfolio were now acquiring smaller rivals to boost growth. He said $8.5bn defence tech group Anduril and $3.6bn biotech firm Eikon Therapeutics “are now acquiring companies and assets and talent and further cementing their market share”.

“I think that wave is just beginning,” Wolfe added.

Venture debt deals decline 38%, led by SVB’s core market

By James Thorne

July 13, 2023

Early-stage venture debt deals have sharply decreased since the implosion of Silicon Valley Bank, which had catered to fledgling startups.

These young companies have been the hardest hit as loan volumes decline across the venture debt landscape. The trend is a sign of lenders imposing higher standards and of startups with uncertain financial prospects failing to qualify for new loans.

In the first six months of 2023, the number of loans for angel-backed and seed-stage companies fell 44% year-over-year, and early-stage loans fell 45%, according to the Q2 2023 PitchBook-NVCA Venture Monitor. That’s compared to declines of 27% and 39% for the late stage and venture growth stage, respectively.

Across all stages, startups closed $6.34 billion across 931 venture debt deals in the first half of 2023, compared to $20.07 billion across 1,513 deals in the same period last year.

Time will shed more light on the state of the opaque venture debt market as more deals from the first half of 2023 are collected in the next few quarters. Business development companies, among the most active venture debt lenders, have yet to file their Q2 financial statements. And many startups refinanced following SVB’s collapse in March, which may have provided a bump for new loans.

SVB primarily operated in the early-stage market, sometimes lending to pre-revenue companies as it sought to deepen its relationships with investors. Following its collapse, many lenders vowed to increase efforts to serve startups, but the fate of early-stage loans remains uncertain. First Citizens Bank, which purchased SVB in late March, said in May that it expects the value of the SVB loan book to fall 8% this year to around $61 billion.

“Since SVB’s collapse, lenders have reported that they have not seen other banks stepping up to replace that specific function that made SVB so unique,” said PitchBook analyst Kaidi Gao.

Debt capital remains in high demand among startups, but it has been harder to secure during a slump in VC investment.

Lenders want confidence that startups are on track to receive future investment and that their investors remain committed to the company. For later-stage companies, lenders emphasize a path to profitability or cash flow break-even, when revenues match expenses. As dealmaking and growth slow, startups increasingly fail these tests.

These dynamics have empowered lenders to demand more favorable covenant packages and warrant coverage, in addition to higher interest rates, said David Spreng, founder and CEO of venture debt lender Runway Growth.

Meta unfriends the news industry in growing rift with publishers

Facebook’s parent company downgrades current affairs on Threads app and refuses to engage with Canadian law designed to fund media groups

Mark Zuckerberg is in a stand-off with Canada over new legislation that would mandate online platforms to pay news organisations for their content © FT montage/Bloomberg

Hannah Murphy in San Francisco and Cristina Criddle in London

Meta is shunning the news business, giving lower priority to current affairs and politics on its social media platforms while refusing to engage with efforts from governments to make the US tech giant pay more to media organisations.

Facebook’s parent company, after years of attempting to placate powerful publishers by funding non-profit journalism projects and striking deals with groups like Rupert Murdoch’s News Corp, is toughening its stance towards the sector, according to people familiar with the company’s strategies. 

Meta’s latest snub came this month when the company launched Threads, a text-based app to challenge its struggling rival Twitter. Threads drew in 100mn users within a record five days of launch, linking profiles to existing accounts on Meta’s popular photo-sharing app Instagram.

The Instagram algorithm, which prioritises content posted by creators and friends over hard news or politics, has largely been replicated in Threads, according to two people familiar with the move. Instagram chief Adam Mosseri said it was “not going to do anything to encourage” news on the platform.

Meta is in a stand-off with Canada’s government, having said it would cull news from its feeds in the region shortly, as legislation comes into force mandating platforms pay publishers and broadcasters for their content. 

The law was designed to boost the fortunes of smaller news groups with less negotiating power. It is also targeted at Google, which said it would bring in a news blackout. Canadian publishers that already had content licensing agreements with Meta have been told those deals will be terminated by the beginning of August, according to two people with knowledge of the move. 

Meta briefly pulled news from Facebook in Australia in 2021 after a similar dispute, prompting critics to accuse it of using its power to undermine a sovereign nation.

More than 30 advertisers in Canada — including the federal, Quebec and British Columbia governments — say they will pull advertising in protest at the move, said Paul Deegan, chief executive of trade association News Media Canada. Canada accounted for about $3bn of Meta’s $117bn annual revenues in 2022, according to regulatory filings. 

“The company is running the very real risk of losing more in revenue than they would pay news businesses under the Online News Act,” Deegan said, adding that he expected other advertisers to follow in the coming weeks.

“We’re going to hit a tipping point where the decision to ‘unfriend’ Canada is going to be bad for users, bad for shareholders, and bad for Meta’s reputation globally,” he said.

Meta has at times sought to court publishers, with brief initiatives around video clips, for example, or, in 2019, with deals for content to run on the Facebook News Tab product. 

However, its senior executives have concluded there is a clash of interests with the news industry, with the growth of the social media giant’s digital advertising business resented as one of the reasons for a global decline in revenues at newspaper groups.

According to people familiar with the shift in strategy, Meta found that its 3bn users prefer short-form videos and content from influencers over news and politics. From 2021 it began reducing the amount of political content in users’ feeds. 

BlackRock’s Larry Fink says crypto will transcend any one currency

by Nathan Crooks

THE BLOCK • JULY 14, 2023, 4:32PM EDT


BlackRock CEO Larry Fink sang crypto’s praise again on Friday, just a week after he highlighted the role of bitcoin as a kind of “digitizing gold.”

“It’s an international asset,” he said in an interview with CNBC.

BlackRock CEO Larry Fink sang crypto’s praise again on Friday, just a week after he highlighted the role of bitcoin as a kind of “digitizing gold.”

“It’s an international asset,” he said in an interview with CNBC, referring to crypto on broad terms. “It has a differentiating value versus other asset classes, but more importantly, because it’s so international, it’s going to transcend any one currency.”

“If you just look at the value of our dollar, how it depreciated the past two months and how much it appreciated over the last five years, a international crypto product can really transcend that,” he continued. “And that’s why we believe there’s great opportunities. And that’s why were seeing more and more interest. And that interest is broad-based, worldwide.”

The asset manager last month filed an application with the Securities and Exchange Commission for a spot bitcoin ETF, triggering a widespread rally in the sector. The SEC has yet to approve a spot bitcoin fund.

BlackRock wants to ‘democratize investing’

“We believe we have a responsibility to democratize investing,” Fink said, commenting on broader growth in the ETF market and highlighting how gold funds had brought down the costs of transactions for the metal. “Now, with crypto, the idea of democratizing that role…the cost right now to transact, it’s quite expensive. We’re talking points, not decimal points.”

Fink said that he couldn’t talk specifically about bitcoin because of the pending application.

“We are working with our regulators,” he added. “If BlackRock’s name is going to be on it, we’re going to make sure that it’s safe and sound and protected.”

Court denies FTC’s last-ditch attempt to stop Microsoft buying Activision Blizzard

A second loss for the FTC this week.

By Tom Warren, a senior editor covering Microsoft, PC gaming, console, and tech. He founded WinRumors, a site dedicated to Microsoft news, before joining The Verge in 2012.

Jul 14, 2023, 4:30 PM PDT|102 Comments / 102 New

The Federal Trade Commission (FTC) has lost what may be its final attempt to block Microsoft from buying Activision Blizzard. It’s the second loss for the FTC after a US federal judge denied its request for a preliminary injunction earlier this week to block Microsoft from acquiring Activision Blizzard until the conclusion of a separate FTC administrative case.

The FTC appealed the decision by Judge Jacqueline Scott Corley, and now the Ninth Circuit Court of Appeals has denied its request for emergency relief to prevent Microsoft from closing the deal until the result of the FTC’s appeal is complete.

Microsoft welcomed the denial late on Friday. “We appreciate the Ninth Circuit’s swift response denying the FTC’s motion to further delay the deal. This brings us another step closer to the finish line in this marathon of global regulatory reviews,” says Brad Smith, vice chair and president of Microsoft, in a statement to The Verge.

Startup of the Week

Shake-Up at Sequoia as Five Partners Exit

By, Kate Clark, July 19, 2023 9:12 AM PDT ·

Five Sequoia Capital partners have left the firm, the biggest shakeup to the storied venture firm’s leadership since its leader Roelof Botha took over a year ago.

Sequoia on Wednesday told its limited partners that longtime partner Michael Moritz would leave and focus on the firm’s independent wealth management business, Sequoia Heritage, a spokesperson confirmed. Meanwhile four other Sequoia partners have exited the firm, according to two people with direct knowledge of the matter.

Five Sequoia Capital partners have left the firm, the latest indication of how leader Roelof Botha is reshaping the firm.

Mike Vernal, a general partner who focused on early-stage startups and has worked at Sequoia since 2016, has left. In addition, Michelle Fradin, a growth investing partner who worked on the firm’s failed FTX investment; Kais Khimji, another growth-focused partner; and Daniel Chen, an  early-stage partner, have also left the firm in recent months. These departures have not been previously reported.

The exits reduce the number of the firm’s investment staff by 15% to 28, according to its website.

The exits come amid a turbulent time for Sequoia, known for its investments in legendary Silicon Valley companies including Apple and Google. The severe correction in startup valuations over the past 18 months has forced some of the firm’s investments to raise new money at a lower valuation than previously, known as a down round. These include Stripe, which earlier this year raised money at a $50 billion valuation, down from $95 billion in 2021.

Sequoia also had to placate its limited partners, apologizing after it wrote down its investment in FTX. In March, it allowed limited partners to break its two-year lockup rule and withdraw some capital early, a move it said was a response to the drop in public stocks.

Tweet of the Week

%d bloggers like this: