On my mind - by Michael Dempsey

This email spawns from this thread.

These thoughts are often poorly edited and researched and usually span areas of venture capital, emerging technology, and other random thoughts.

I’d love for this newsletter to be more of a discussion vs. me shouting to the internet, so please email me back and I promise to respond.

Thoughts

1) Minding The Inequality Gap

A ton of my research recently has been sitting between exciting technological development and a future that pushes the gap further between have/have nots and/or further distances the upper classes of society from middle and lower classes.

We can look at this in its earliest stages today with family planning (the mere cost of IVF boxes out many families) and will continue within that category while extending through traditional healthcare and longevity.

Why I’m writing about this today though is an interesting paper surfaced to me by Gwern (follow them). This paper analyzes predictors of educational success based off of both DNA differences between children identified via polygenic scores, paired with their parents’ socioeconomic status.

The paper concludes that there is a meaningful gap between “high-high” (high socioeconomic status and polygenic scores) and “low-low” equal to A- to C- grades, or 77% to 21% college attendance rate.

What was most relevant to me however was the middle results which show mixed variance. In the conclusion the researchers frame a suggestion that very much fits in line with this have/have nots future:

We suggested, for example, that high GPS partially compensates for the disadvantages of children from low‐SES families, increasing their chances of going to university from 21% to 47%. This raises the possibility of doing more to help this group reach its full potential.

Ultimately the researchers caveat that the accuracy of these predictors at an individual student level are incredibly difficult to truly understand, however the sentence above very much painted a picture in my mind where we figure out what groups of age 6-16 humans are worth investing in, and which aren’t, based off of some predictive factors.

As with all points in this space, somebody could easily say that we are already are doing this (whether in other countries that are more explicit about doing so, or in the US where such a narrative exists, and may be factual, across various demographics of people).

Nothing to tie this up but an area that is definitely top of mind these days, and over the next decade as we continually try to quantify the world around us.

2) Illiquidity Discounts & CIO Job Security

There was a great piece written by AQR surrounding the concept of an illiquidity discount. A year ago, I answered a question surrounding how the venture landscape could change versus a potential pullback in broader macroeconomic conditions.

The idea that investors will want to flee to potentially illiquid assets makes a lot of sense for the psychological fortitude/regret components that AQR mentions (basically, if you see a stock falling, but you can’t sell, or you don’t see it falling at all because illiquid markets mark less often, then you may enjoy that in downside volatile times). While I generally believe that the net dominant strategy from a returns/diversification standpoint in a market pullback would be to invest in things like illiquid seed funds (they have longer time diversity, though likely pricing will be impacted less vs. growth) I actually think the vector that could be most compelling is for the investment committee members themselves and how they can keep their jobs.

When a sports team struggles and starts to underperform the historical past few seasons, often the manager is the first person replaced. Similarly, when the world begins to burn, often everyone looks at the CIOs and other heads of these groups and the decade-long bull run we saw was all luck, while the market downturn should have had some semblance of skill to be detected. To not detect that downturn is to not have skill, and public markets and shorter-term investments show you incredibly quickly how little skill one can have.

If you’re a CIO or core investment group searching anywhere for yield, it’s likely that you can actually save a few more years of your job by backing venture funds that are expected to lose money over their first few years (the J Curve), have little meaningful data that is thrown off for the first 5-7 years, and are considered potentially low correlation to public markets. I think there are a large batch of people that the dominant career strategy could be to park it in long-term, illiquid assets with little opacity to returns, and venture is an increasingly institution-grade market for that.

3) “Founder is all the matters at seed” is wrong.

There’s a view within early stage venture that I believe has gotten simplified and bastardized over the past 5-7 years surrounding the idea that “founder is all that matters”. It sounds nice as we think about humans being core atomic units of a startup, however it both doesn’t take into account founder-market fit (which I believe is 100% real) as well as how financing dynamics have changed at the early stage.

I don’t have vast data on this but I believe that due to an abundance of pre-seed, seed, and seed extension capital, as well as a higher bar for Series A, founders are able to either make small iterations from pre-seed through seed extension (over the course of 2-4 years) as they try to iterate towards product market fit (showing small inflection points at each raise before failing again) or have a massive, quick failure, which some of the best founders can engineer towards an acqui-hire or soft landing before restarting.

I anecdotally am seeing and hearing far less hard pivots surrounding companies starting in one market and pivoting entirely into another. Because of this, the idea that all that matters is the founder (built on the premise that great founders will pivot around until they hit PMF) feels wrong and overly simplistic.

4) Escapism vs. Reality in Animated Storytelling

There’s a passage in the book Marvel Comics: The Untold Story surrounding the differences between the Marvel and DC universes and the creative decisions of how and where those universes were set. The key view of Marvel was that telling stories that tied to real places gave them an outsized advantage vs. DC which was based in fictional places. As the author says:

…but every kid knew that they were tethered to their respective Metropolis and Gotham City and that never the twain would meet. Who cared if the Acme Skyscraper fell, or the First National Bank had to give up its cash? Timely's NYC on the other hand, was rife with real stuff to destroy.

When looking at this through the lens of virtual influencers/digital celebrities, it’s become quite clear that people feel the need to blend these characters into our real world both as a storytelling mechanism and as a potential monetization strategy for selling real world goods. I think that a large number of the creators of this IP have misjudged the value of escapism and extending their universe though, as the real world tether only more closely ties its IP to potential holes that can be poked in stories, and narrows the ultimate reasoning for why these characters should build a following.

One of the core components of the animation industry is asking the question as to why a given character or universe should be told via animation vs. live action. Where many have gone wrong is thinking of animation as a pure medium vs. a storytelling tool. While in the mid-1900s, we wanted to see our real world in a more dynamic way because our view of what was happening in the world was so narrow. I’m not sure that over the past 50-80 years, this principal has held true for non-live action IP. Today we see every part of the world in a form of hyper-reality, at our fingertips. In 2020 we not only crave escapism, but we also crave creativity and novel ideas/stories.

It in some ways reminds me of this quote from The Mission of a Lifetime.

The immensity of earth used to inspire awe. Then we went to space.

5) Venture firms either evolve with or strongly mold new employees

I’ve now been in the investing world broadly long enough to see a bit of a generation turn (and mostly, expansion) in the venture capital space, as well as to know people just as they join their firm and to spend time with them over the years as they spend more time in that culture.

I increasingly believe that venture firms either evolve with their humans, or the humans bend to their culture in an outsized way. This feels obvious to say but in reality venture firms, which are not scalable, and where team members operate towards a single goal in a day to day decentralized way, feel this far more than any startup.

For the latter, some firms have shown an ability to take a person and integrate them, showing the power of long term brand network effects proliferating all the way to the behavior, analysis, and hiring process. The person often trends towards a lowest common denominator and/or hyper-specific style that is associated with the firm and its partners.

Alternatively other firms have decided to let the humans play an independent game, functioning on a single vector while having a different narrative structure surrounding how they invest and why. This leads to some fragmentation and probably less strong brand network effects in venture at the fund level, but probably a more defensible long-term moat in terms of recruiting and aggregate partnership ability to win a wide range deals.

As with many things in venture, I don’t have a view on which is better just yet.

Research / Links

• Domain Stylization: A Strong, Simple Baseline for Synthetic to Real Image Domain Adaptation - Similar to last newsletter’s thoughts on the desire use ML to do text generation of 3D scenes, I’ve also continued to see tons of research surrounding ability to transfer lower quality environments to higher quality standards. This idea was compelling when transferring various synthetic datasets into real world datasets to be used on largely computer vision tasks (most often in autonomous vehicles). There are various arguments surrounding the necessity for high quality 3D environments in ML training, but either way teams are starting to really crack this idea of building synthetic datasets in scalable manner with potentially real diversity of data.

• An algorithm that learns through rewards may show how our brain does too Cool article + paper related to mapping dopamine release in the brain and it’s strong resemblance to how reinforcement learning reward mechanisms work. I’m not going to try to take a stance here but if you’ve ever been stuck working through a problem (this mostly resonates with research or being a bad software engineer for me) where you are trying to accomplish a task and then it goes better than you thought, this premise is quite interesting and intuitive vs. prior views around more binary dopamine release.

• Write-A-Video: Computational Video Montage from Themed Text - I previously have written about the continual task fo trying to utilize ML to go text to scene. This paper (with accompanying video) is a bit more realistic and interesting as it focuses on using paired video repositories in order to curate the creation process based off fo text in what they call visual-semantic matching.

• Illegible Text to Readable Text: An Image-to-Image Transformation using Conditional Sliced Wasserstein Adversarial Networks...Using ML to make handwriting into printed text. Just a cool use-case.

If you have any thoughts, questions, or feedback feel free to DM me on twitter. All of my other writing can be found here.

On my mind - by Michael Dempsey

This email spawns from this thread.

These thoughts are often poorly edited and researched and usually span areas of venture capital, emerging technology, and other random thoughts.

I’d love for this newsletter to be more of a discussion vs. me shouting to the internet, so please email me back and I promise to respond.

Ask - We’re hiring at Compound. If you’d like to work with me or know someone who would be a good fit, please send them my way!

Thoughts

1) Narratives & Pseudosecrets

I wrote a post discussing the importance of narrative building as well as how companies build their narrative around both TAM expansion and sequencing to an ultimate future via what I call Pseudosecrets. I’ve spent a lot of time explaining this phenomenon or principle to founders over the past few years and finally decided to write out my thoughts in long-form.

2) Serendipity in Venture Capital is BS

I also wrote some thoughts on the history of networks within VC, the current state of venture capital, and ideal investing models at the seed stage.

3) Fortnite x Star Wars + VR UIs

There will be a lot of hot takes surrounding the Fortnite experience, as always, however while many people were *mindblown* around the in-game experience, if you’ve spent time in VR, it’s likely it felt quite familiar. Watching experiences in a 3D, free-roaming world on a flat screen is sub-par, but has been a well-known UI/UX choice for VR developers and is in some ways considered a killer app today. Unlike Fortnite however, in VR you are immersed via head tracking, hand tracking, and a first person view. Ultimately for me, it felt a bit odd watching my character jump around in front of the flat digital screen, in this large world.

Where Fortnite did innovate on the UI was an ability to “focus” with the right click, which allowed the user to be forced into watching/following what Fortnite deemed most important. This is something VR should adapt more often, especially in storytelling centric experiences.

Related to that, the Fortnite x Star Wars experience was cool because of the IP, live nature. and mechanics surrounding user voting, but it also was the first meaningful in-game experience that didn’t progress the Fortnite story. While the metaverse story (read more about this in my Narratives post above) seems to be the goal, and a profitable one at that, I hope that after a reset of a season in terms of vaulting multiple game mechanics, Fortnite continues to innovate on gameplay, and not just on becoming one of the world’s largest native advertising platforms.

4) On the “f**king mint” meme

Disclaimer: I’m a millennial snipering into Gen Z trends here so take it with a grain of salt.

I’ve been pretty fascinated by the “f**king mint” meme (shout out to trying to avoid spam filters) that has grown on TikTok. Basically the point of the meme is to go around and say self-deprecating/embarrassing/not-so-great things about you, your life, and/or all of your friends and be somewhat OK with it.

I think this meme speaks to something a little more specific going on within gen Z which is self-comfort, open expression, and re-assurance (notice I didn’t say confidence) that millennials perhaps adopted about a decade later in age than this cohort of teenagers has. It’s a small, potentially overfit signal I’ve noticed as we’ve continued to look at other related thesis areas, but one that I think the meme perfectly embodies and could bleed through to other behaviors and purchasing decisions.

5) Mobile compute ML research is underserved. But does it matter?

Michael Dempsey@mhdempsey

Machine learning is cool until you try to re-implement a new paper you're excited about and end up realizing you're about $30k short (and weeks away) on training compute cost.

March 10th 2019

1 Retweet39 Likes

There are countless research papers that are continually published popping up pushing the limits of what machine learning can do across a myriad of use-cases. The underlying issue with much of this research however is the required compute to make something possible, let alone reproducible.

One could argue that the job of most research labs is to figure out if something is possible, and compute will eventually catch up to make things production-ready at the commercial layer, however I’m not sure we’ve seen this happen as quickly as we’d like as an industry, largely due to the clout coming not from efficiency but power of algorithms.

One area I’d love to see increased publishing and experimentation on is ML at the edge, specifically within mobile phones. If we eventually believe that these mobile supercomputers will turn from bundled interface + compute to possibly compute-centric devices (think post-AR hardware) then we should also be pushing the boundaries on what type of algorithms we can efficiently run.

Examples of compelling mobile-centric ML research I’ve seen recently include this paper which tackles real-time monitoring of drivers via mobile phones, as well as one on mobile action recognition. I hope we see more in the future.

6) AI generated art novelty decay happened even faster than I thought

Two months ago I wrote about novelty decay AI generated art and more. Specifically I said:

The novelty of “this was made by AI” or “this is digital” will continue to exist, but at an increasingly decaying premium as time goes on. The novelty premium of our favorite artist’s lives may only compound as we see them grow, change, and we build deeper personal connections to them, their tribes, and their ups and downs.

It looks like that decay could have happened significantly faster than even I anticipated with the recent mediocre performance in AI generated art. I’d wager that this decay in value comes partially from the tiring of the group (Obvious) as well as the way in which this art is generated. Once we see a meaningfully different technical approach to art generation, perhaps we then will see another pop in prices. With that said, I don’t believe we’ll see lasting value unless a larger artist is able to manage both the story surrounding the process and their own personal journey, alongside their pieces.

Research / Links

• A Mobile Manipulation System for One-Shot Teaching of Complex Tasks in Homes - This is a really interesting paper (and video) that walks through a team using VR to train a mobile, in-home robot on specific tasks. The success rate ends up being around 85% across the board, which is also not good enough (despite a 99.6% success rate due to ability to correct errors) however the other problem is the time to complete which is 20x slower on average (and some tasks were 100x slower than a human!). I imagine the eventual future of an in-home robot is going to draw on some form of imitation learning that maps humans more closely to the robot (or vice versa) so that learned tasks can be scaled by more human teaching. Not sure when that future will come though.

• When equity factors drop their shorts (article here): On the lack of value short trading positions create.

• Generating High-Resolution Fashion Model Images Wearing Custom Outfits - Generating fashion images from scratch in high resolution. I've spoken to a few different people about this task of generating stock fashion images with GANs. Over the past 6 months we've seen incredible pace as full-sized human body synthesis has improved drastically and thus we can now combine models of pose understanding, and GANs in order to tie together new types of synthesis. What has been funny is people doubting the ability that pace is going to happen in this field specifically. While in some industries I'm a hardcore skeptic on tech people automating or innovating from the outside, when it comes to fashion and stock imagery (and other non-scalable, image related practices within e-commerce) after spending a little time with fashion industry founders, I'm beginning to believe stronger that this innovation may be one that comes from outside the industry vs. an operator from within.

• Neural Voice Puppetry: Audio-driven Facial Reenactment - Take audio, push it to deepfake. Pretty cool.

• Generating Animations from Screenplays - Multiple researchers continue to try to crack this utopian vision of inputting text and outputting realistic 3D scenes. Very few have been able to do it at high -degrees of freedom, high quality of art, and by enabling emotion. The argument to be made is that we can speed up the initial work and go from creation -> tuning, or have a more granular storyboarding pipeline if it’s automatic, but many artists just get annoyed by poor implementations of animation that they are then forced to re-do vs. create.

If you have any thoughts, questions, or feedback feel free to DM me on twitter. All of my other writing can be found here.

On my mind - by Michael Dempsey

Michael Dempsey@mhdempsey

I just cleared out my inbox and so figured I'd start the end of the "work week" with 5 things that are on my mind, largely inspired by my now-archived inbox.

September 28th 2018

7 Likes

This email spawns from this thread.

These thoughts are poorly edited and researched and usually span areas of venture capital, emerging technology, and other random thoughts.

I’d love for this newsletter to be more of a discussion vs. me shouting to the internet, so please email me back and I promise to respond.

5 Thoughts

1) I’m seeking inspiration. Let’s find it together.

Here are a few areas I’ve spent a lot of time in this year:

• Adversarial attacks on machine learning models

• AI friends

• Animation

• Avatar-first products

• Creative ML

• Gender fluidity and its impact on consumer

• Human pose estimation research

• Fashion + ML

• Full-stack robotics companies

• Future of family planning

• Modified humans, animals, and plants.

• Psychedelics and their impact on healthcare

• Using ML to decode non-verbal communication across animals and humans

Here are other areas I want to continue to talk to people about that I’ve spent some time in this year:

• Science-based CPG products

• The past, present, and future of game engines and large scale simulation engines

• Social VR

• Projection mapping + holograms

• Enabling more angel investors

• Space (non-earth observation related)

If you’d like to chat about any of these or other post-science project ideas, feel free to respond to this email or directly email mike@compound.vc

2) On CTRL-Labs & Fund Dynamics

Michael Dempsey@mhdempsey

It's kind of crazy that CTRL-Labs can get acquired for $500M-$1B on $72M publicly raised in 4 years...& it will return the fund for (i think) none of the investors involved (maybe Fuel Capital). Lots of thoughts here but fund sizing & ownership for VC funds in 2019 is tough.

September 24th 2019

16 Retweets164 Likes

The tweet above triggered some conversations over the past 24 hours. Friends at Lux, Spark, Founders Fund, GV, and others were all right in a big way, and right on a difficult space to build conviction on. To his credit, Josh Wolfe was a giddy public cheerleader for this company since before he invested (I remember how much he was excited the first time he told me about them). It is easy to be a cheerleader as a VC when the company is at series C+ and has nailed commercialization, it's a lot harder when it’s a Series A neural compute interface startup, so props to Josh.

The point I was making with my tweet was that creeping fund size and round dynamics in 2019 make it really difficult to nail venture returns and change investing dynamics...and this company is a great example of that.

Large “full-stack” firms that view their entry point as majority series A, now must deploy so much capital into a company over its lifespan, gathering more ownership, and thus trading deal multiple for cash-on-cash returns, as they target their 2.5-3x net fund. The dynamics of "will this return the fund" may no longer necessarily apply to these firms. Instead, with larger funds (and opportunity funds), the question weighs capital deployment over initial ownership in a much stronger way than vintages past.

The question becomes: "Can we get enough capital into this company so that if it has the outlier company that we hope to have 1-2x/fund, we'll return the fund (or maybe 75% of it)?"

But well-regarded founders can command substantial capital ahead of traction, and those founders working on massive upside, possible platform companies can further bend round dynamics in a capital abundant world. Specifically, for opportunities that are as "moonshot", capital intensive, and high upside as something like CTRL-Labs is/was, it then makes sense to bend but not break, get into the early rounds with the hopes that you can deploy not only your venture fund into the company, but also your opportunity fund capital. This is something that could have led to five $400M+ fund size firms splitting (to varying degrees) 3 announced rounds ($11M, and two $28M rounds).

Regardless of these dynamics, this is a great investment and as noted, the IRR has got to be incredible.

Two interesting notes in terms of the investors:

1. You can make the argument that Spark has now had this happen to them 3 times, as early backers of Oculus (alongside Matrix, who also invested in CTRL-Labs) and Cruise as well. So impressive.

2. Lux on a different vector (massive moonshot, capital intensive) saw this play out successfully (though over more time) with Auris Surgical (sold for $3.4B-$5B+, raised $733M).

Okay, enough armchair VCing, congrats to my friends who invested, and maybe to NYC which hopefully will get a few more deep tech angel investors in the mix.

3) Will digital goods be what kids today, buy as adults tomorrow?

I recently read This Is Not a T-Shirt by Bobby Hundreds. In the book he makes a point about Japan's role in sneaker collecting culture. Retro AF1s, Dunks, etc. were really sought after in the 2000s because the kids who wanted these shoes in the 1980s were now young professionals with money and were coming to market to "fulfill their fantasies". What is the version of those goods today? What will today’s children lust after 10-20 years from now once they have some discretionary income? Is it goods that already fetch high prices on thee aftermarket that are modeled from the beginning as scarce (a la Yeezys or Supreme), is it digital goods that they grew up with but couldn’t convince their parents to purchase (Roblox goods/Fortnite skins?) And if it IS digital goods, then how will aftermarkets really exist outside of NFTs (maybe NFTs are the answer) when the platforms these digital goods live on become irrelevant or stop functioning?

4) Expectations vs. reality, VC & parenting edition

In the early days of your venture career you often think about all the amazing things you’re going to do to find people, help companies, invest better, etc. Some of them truly do give alpha, while many of them fail for various reasons. As your portfolio grows you start to see where things break in this business and have to be very mindful about the experiments you run to achieve better returns, founder NPS, scale, and more.

Some of these lessons are ones that others who have come before you can teach you, but you really need to learn for yourself.

I wonder if the same thing happens with parenting. You think you’re going to do all of these amazing things for/with your children, but eventually you realize at each age the diminishing (or negative) returns that these idealistic thoughts have, and/or life continues to get in the way of some of these ambitions, and you end up not being able to actually do all of the things that in your childless state you thought you would be able to do one day.

5) Novelty Decay in AI-generated/synthetic content is taking hold

I wrote a quick note on how the novelty of both digital celebrities and AI-generated music is decaying and capping the upside of new entrants. I only expect this to accelerate.

Research / Links

• Game character generation from a single photo - This paper has really interesting implications surrounding 2D photo → 3D asset generation. In general this is an area of research we’ve seen a bunch of people take aim at over the years (most famously, and early, Hao Li’s lab, and more recently everyone from universities to Facebook). There are novel bits and pieces technically, but the main thing is that this is a future that many people have thought about or wanted for some time (Loic at Morphin has often talked about how his origin story of his company started by wanting to put himself in FIFA and other games). Just always cool to see clear science fiction start to push forward towards reality on fairly “mainstream” ideas.

• How much real data do we actually need: Analyzing object detection performance using synthetic and real data - This paper takes a look at the value that synthetic data can bring to training models. Specifically it realizes something that I don't feel is intuitive to most which is that the open-sourced synthetic data training sets don't generalize very well, nor do they have great diversity of data (these problems go hand in hand). What I've seen spending time in this space is that diversity of data is incredibly misunderstood and underemphasized. Even when looking at something as seemingly "gold standard" years ago as GTA V, researchers and engineers I spoke to at the time realized that the diversity of textures and more were incredibly underwhelming and wouldn't transfer learnings well to a real world environment. What we've started to see now is data expansion via style transfer or just synthesis, essentially utilizing GANs to increase diversity. My feeling on this paper is that it is quite negative for how many perceive synthetic data, especially in areas where highly generalizable perception models are needed (think self-driving), however it also tells me that companies focusing on specific types of perception/understanding are likely far ahead of where current open-source datasets are. I previously wrote about AV simulation and am an investor in AI Reverie.

• Making the Invisible Visible: Action Recognition Through Walls and Occlusions - Using a combination of visual and non-visual signals to recognize actions through walls and more. Pretty incredible.

• Adblockradio - Using ML to remove ads from radio feeds. Pretty funny.

If you have any thoughts, questions, or feedback feel free to DM me on twitter. All of my other writing can be found here.

On my mind - by Michael Dempsey

Michael Dempsey@mhdempsey

I just cleared out my inbox and so figured I'd start the end of the "work week" with 5 things that are on my mind, largely inspired by my now-archived inbox.

September 28th 2018

7 Likes

This email spawns from this thread. The process for this will evolve but as you'll see, some thoughts are random, and most are unfiltered or poorly edited. Either way, let me know what you like, don't like, or want to talk about more.

1 Thought

Michael Dempsey@mhdempsey

As a sidenote - Is there any interest among founders to see VC investment memo templates? At the very least, I'm always curious to see how different firms approach this (and am always interested in learning from those that don't do memos at all)

Michael Dempsey@mhdempsey

@micahjay1 @fcollective Do you all use the same template for deal memos, and are they laid out in the same way as Joseph's post? If not, would you publicly share your deal memo template?

August 12th 2019

1 Retweet100 Likes

Spoiler alert: There was interest so see below for my memo template and high-level thoughts.

1) Thoughts on Investment Memos

Michael Dempsey@mhdempsey

Re-reading old investment memos is the best. It's not always practical but honestly I think all firms should write memos, even if they are done a few days after signing/close in an expedited process.

November 12th 2018

27 Likes

I’ve talked a lot about investment memos in the past and how I believe we should be more open/transparent about the overall structure. At times, I’ve even shared the exact internal memo with other friendly investors who were doing diligence for a round I was leading. 

The summary of my thoughts are:

• Memos are great internal snapshots of thinking. You can go back and see what you got right/wrong about each investment, and how you were thinking about a market and team.

• For some firms (not ours) memos help convince the rest of the partnership to get comfortable with an investment and okay it.

• Memos for follow-on investments are just as important as new investments (though probably impact decision making less realistically). I structure these differently than new investment memos.

• Investors should share memo structure more openly. There is very little that is proprietary about a memo template, however there aren’t many that are public. Examples include the famous Youtube memo by Roloef Botha, Tim Devane’s post, and USV which uses their blog posts as memos. Steve Schlafman also once mentioned that he shows his memo to the founders before taking to partnership to see if they’d change anything. I’d imagine this is firm/partnership dynamic dependent on its efficacy/value.

• There’s a lot of vulnerability to sharing a proper investment memo, especially after the fact. VC is an industry that is built on constant paranoia driven by massive information asymmetry. The last thing many people want to do is expose themselves.

  Michael Dempsey@mhdempsey

  @maxbulger @daveambrose @fredwilson 100%. IMO VCs on average don't share (and you know I've tried) because it's a lot easier/safer to sound smart in 3 talking points than it is in a well reasoned memo that recognizes downside possibility and leaves you vulnerable.

  January 2nd 2019

  5 Likes

How I Use Memos

I’d say ~70% of the time when I write a memo, I end up offering to make an investment in the company, so sometimes a memo goes nowhere but into my archives.

I start my memo usually after meeting #2 with a company and I’ve found it’s very helpful in figuring out what do I have left to understand about the business, what areas should I push on, and separates the company from the investment (i.e. a good company doesn’t always mean a good investment, often due to price).

At times where I have expertise overlap with my partners on an investment, I also will use the memo to bring them up to speed.

My memos traditionally run long for seed (8–10 pages on average) and have a separate attachment of all of the notes I’ve taken as I’ve been meeting with the company. If we’ve had phone calls/video calls, they are multiple pages and have multiple quotes that the founder said. If our meetings have been in person, I usually distill my notes after the fact, as I don’t love taking handwritten notes, and I don’t want to have electronics out in person.

For deals that move quickly, I almost always still try to write a deeper memo within weeks after we invest, to have that record of thinking.

The Actual Memo - CLICK HERE

Here is my investment memo template in google doc form (note: we don’t use a standardized memo at Compound, each partner has their own version/structure). 

In a future post, if there is interest, I’ll walk through various parts of the memo. Please email or tweet at me all specific questions and I’ll do my best to answer them all in this post.

For now, here are the opening parts broken down:

Summary:

This is straightforward. For information about the “required exit for RTF” section read my post on this equation.

Risk/Mitigations:

I try my best to understand what are they key risks within a given business on both from going Seed to A, but also from becoming a fund-returning investment. The biggest learning as I’ve done more investing is that for many risks, there are no mitigations.

Sometimes founders are first-time founders who have never managed a budget before, others have been at big companies and you worry about their ability to move quickly and efficiently, sometimes they have to build really expensive and high-quality engineering teams. Often for these more amorphous risks, the *honest* mitigation is just a leap of faith, or maybe a small proof point in their past or that came through a reference we did on the founders. In most of these cases I will just write “N/A”.

Other common risks are (but not limited to):

• Concern on the founding team’s ability to fundraise and the market dynamics within the industry — As an investor often investing in highly-technical businesses that won’t have major proof points at Series A, it’s unfortunately important for us to think about follow-on capital dynamics both in the sense of Have any other VCs made investments in competing companies (essentially boxing them out for follow-on)?, Is there a deep pool of possible strategic investors (these can be helpful for early commercialization or less-valuation sensitive follow-on)? Or even; is there a specific love for this type of technology from international, non-VC investors?

• Raising enough capital to hit meaningful inflection point (this can be both technical and execution risk roped into one)— I think about investing as giving a company enough money and help to reach an inflection point of value so that they can raise more money (eventually leading to a successful exit). It’s kind of sad when put that way, but that’s the most capitalistic view on what we do as seed investors (along with partner with great founders, help people achieve dreams, whatever else we want to pander about on twitter to make ourselves feel/look better). For some businesses, related to the first bullet, it’s actually unclear *what* the right inflection point will be. My job as a lead seed investor is to help our founders know these metrics at a market level, and at an individual investor level. Unlike in more well known areas like enterprise SaaS where people generally have an idea of growth rate and ARR required to raise a strong A (until a hot company blows those expectations out of the water), the areas I invest in are less clear. For robotics, some Series A investors will want to see lots of pilots and customer data paired with LOIs, for some they just want to see a scalable MVP and an elite team, and others think they want to do series A robotics investing but really don’t and won’t get comfortable unless another firm outbids them.

• Market Size — I admittedly don’t think too deeply about this nor (despite my hedge fund background) do I care to do a finance-style analysis on TAM, especially not in the businesses that I often am investing in. I have a general view of “can you build a $100M+/year business over the next 4–7 years, or not” and that’s about it. Projecting TAMs for companies that are solving previously unsolved problems is guesswork at best.

• First-time founders — Again, I worry about this from specific skillset angles like ability to manage capital or inspire people to take less salary to join their company, but there isn’t much mitigation. For a few companies a mitigation has been on the acqui-hire value of the people based on their unique skillsets, but this often decays within industries over time.

More case-specific examples of mitigations include:

• Simulation is built in-house — In AV-first simulation this was a key risk where I was concerned that behaviors from players like Waymo and Zoox. which had built fully-featured simulation software stacks, could continue to proliferate to any of the long-lasting AV companies over the next few years, as these businesses built up the early revenue before expanding into other verticals. 
  Mitigation: “Tech-heavy OEMs like Tesla, GM, Ford may be able to attract software talent but with projected rise of auto OEMs + Tier 1s, many won’t be able to hire software expertise to build. The key factor here is the delta in time between proliferation of AV teams (and thus the business that can be built) vs. when someone solves AV and we see industry contraction.”

• Slowed market expansion means fast traction will come from large customers early on which may be hesitant to implement at scale. — This was for a company working on a massive, but highly technical industry and selling infrastructure to those building companies within that industry. My concern was that in this industry I had repeatedly seen companies be able to go from proof-of-concept to pilot but very few had expanded into scaled recurring revenue, leading to a bunch of bridge rounds to nowhere as everyone waits for the market to mature.
  Mitigation (anonymized): (Strategic Investor) is investing $xM into this round and plans to pilot and then scale new tech product across multiple business lines. The company has shown an ability already on old product to break through procurement at large Fortune 1000 customers.

Deal Structure/Offer Context

The structure is pretty simple: How much are we investing, what is the valuation, what is the total price, and if there are any additional terms (is another VC fund or studio getting warrants for some reason? Is the liquidation preference non-market? Will the founders hold legal responsibility post an acquisition?)

The context is what matters far more: Who else is competing for this investment? What is it that we feel we must write check size-wise to win? Where in the process are the founders and what do they value most? Will we need to potentially bridge this company so should we hold reserves between seed and A?

There are many variables here that lead to where in our investment range we invest (both on the founder side and our side).

If you have any thoughts, questions, or feedback feel free to DM me on twitter. All of my other writing can be found here.

On My Mind - by Michael Dempsey

Michael Dempsey@mhdempsey

I just cleared out my inbox and so figured I'd start the end of the "work week" with 5 things that are on my mind, largely inspired by my now-archived inbox.

September 28th 2018

7 Likes

This email spawns from this thread. The process for this will evolve but as you'll see, some thoughts are random, and most are unfiltered or barely edited. Either way, let me know what you like, don't like, or want to talk about more.

Two self promotion thoughts this time!

I wrote about animation (#1).

I was on Erik Torenberg’s podcast (#5).

Main Thoughts

1) Animation is Eating The World

I wrote something on the history of animation, the future of animation, and how different technological breakthroughs have had profound effects at various points in time. This newsletter is focused on animation because I wrote something really long about it. My main view is that animation is vastly undervalued, under appreciated, and on the brink of a new explosion of content that is incredibly valuable to many stakeholders. This piece is a result of lots of research and conversations and copious notes. It’s a long read, but one that I think has lessons that apply to multiple different industries within both tech broadly, consumer specifically, and media of course.

The website I built for the piece is slow to load, but I think visually important to read on, so please wait the full ~10-20 seconds for it to load. Again, it’s a long read so feel free to use the table of contents to skip around and read what interests you.

Also please share it!

2) We have both heavily overestimated and underestimated the commoditization curve in machine learning.

We underestimate ML in certain areas and drastically overestimate it in others, leading to deaths of companies innovating due to the commoditization curve (object recognition as a service companies), and others dying betting on the commoditization curve that never came (Jibo). This has made it incredibly difficult for founders in the ML space to understand what investors want, and for investors to understand what is truly defensible and won’t be pushed down to a near free, general-purpose but horizontal model.

We as investors say hand wavy things like “ability to acquire proprietary datasets” for some vertical ML applications but increasingly areas have shown that this isn’t as advantageous or defensible as some believe. I’m incredibly interested in reading an updated take on Google’s One Model to Learn Them All paper.

Increasingly I’ve distilled my view on some defensibility as an understanding/elite ability to do research adaptation (and then expansion) into a commercializable product. While this sounds simple, it requires a pretty complex blend of research brain and production brain founders that are rare.

Related - It's been fascinating seeing how the "minimum viable implementation" of ML can stay strong for a really long period of time where 10x implementations don't cause any excitement. This specifically I'm thinking of neural style transfer and how applications like Prisma took early steps at this and wow'd consumers (for a brief time). Now we're seeing increasingly scalable or transferrable and bleeding edge research related to this that few people would care about or notice.

3) Is specializing in Consumer early in your VC career a bad idea?

I originally thought the best advice for new VCs is to not do consumer, but now I'm not 100% sure. It may be smart to do consumer because you'll get a super fast feedback loop (though ability to sift signal vs. noise as to why something worked in short-term but didn't generate a long-term valuable business is tough). My early thought as to why you don't do consumer (specifically consumer social or digital consumer businesses) is because the failure rate is just significantly higher/faster, with little process or understanding of failures vs. wins, and often a difficult to support investment thesis if the company fails.

It could be dominant to do consumer though as a junior VC if you can capitalize on a few hot, early deals and then leverage that into a more stable, longer-term role, before your consumer company burns out or sells in a premium acqui-hire.

Within non-digital consumer (think consumer brands/CPG and even consumer healthcare products), we’ve seen an ability to build single-digit $M/year businesses occur at a faster rate, but it’s still incredibly difficult to break past that $10-$20M/year ceiling. I haven’t heard an incredibly compelling vision as to how to understand those types of companies vs. capped-upside companies from some of the elite consumer investors. There’s an argument to be made that for D2C businesses, you’ll have a lower failure rate so you won’t have to burn as much capital making mistakes early on (though in reality, a 1x isn’t too different from a 0x when it comes to burning capital). The issue is, you will also probably have lower upside, so if you’re doing these investments in the scale of a traditional technology VC firm, your results may not be as valued or matter vs. a firm built around the dynamics of D2C businesses. And the lower failure rate today (paired with larger funds that need to pour $$ into companies) can lead to overpriced early rounds with small signs of traction, as full-stack VCs (more on them later) can see an avenue to quickly putting tens of millions of dollars to work on CAC).

On the consumer digital side, many firms seem to be sticking to the playbook of wanting to play the call-option game (write tiny seed checks into companies out of a $250M+ fund with the hopes of leading their A), but I don't know if that really works in a highly competitive series A+ environment.

This may be all a moot discussion though as I’m not sure being a horizontal “consumer investor” means the same thing anymore/is possible at an individual level. What I mean by this is that post-facebook/twitter/zynga/snap we had a rush of people wanting to find the next mobile consumer social win, which led to bets on companies like Houseparty/Meerkat, Secret, Whisper, Peach, and more (I don’t think any of these companies returned a fund, except maybe a Discord, which could but is a fundamentally different product). Now however the explosive consumer companies have looked more like Uber/lyft, bird/lime, hims/Ro, Allbirds, and some others i'm forgetting. I’m not sure it’s the same profile of investor (or even common thread between companies) that will see and get excited by all of those at seed enough to lead. Bird/Lime will feel complex and capital intensive to an investor that loves Hims/Ro’s ability to instead spend their VC money on digital acquisition. Allbirds’ upside will feel capped or moat will feel weak compared to Hims/Ros recurring nature and ability to go incredibly horizontal. Hims/Ro will feel like a regulatory risk down the road to Allbirds’ responsible brand and clear ability to become the next big shoe company, etc.

Back to the point about call-options, I will say that I'm confident that having a mandate to spray and pray at consumer seed is probably the worst of both worlds. Many full-stack firms have started to do this and while you may get some early founder face time with a now-hot series A consumer deal, I'd be surprised if it leads to a meaningful win-rate at A rounds (does anyone want to share data?).

There are a lot of dynamics at play here that I’ve written horribly about above, so let’s get back to the core question of is doing consumer dominant early on for your career in VC? The answer is, it depends. A less cop out answer could be: If you’re a new GP who is now going to be measured on the economics/returns you bring into the fund and you have 2 funds to prove it, maybe not. But if you’re a junior person looking to parlay a brand elevation into a better role (or learn quickly and get out of venture), maybe so.

4) Applying deep learning to identify patterns in animal behavior could lead to new understandings of how they work.

This paper digs into detecting pain of horses via DL. You can start to see a slope where we end up being able to better understand animal health and preference through patterns, similar to how we can identify these things visually in humans with enough pattern recognition (although thus far computer vision and deep learning have proven quite poor at truly understanding emotion from visual cues alone). Nothing concrete here but an interesting rabbithole to go down in academia and an even more interesting future to imagine where we both communicate with animals or have a much deeper understanding of them.

5) I did my first podcast since 2015. I need to be better at this.

I went on Erik’s podcast (listen here) and discussed a bunch of things that I care about including machine learning, robotics, family planning/women’s health, animation, gender fluidity, and more. I definitely spoke too fast and used the word “like” too much but it was a lot of fun. We didn’t go aggressively deep into each of the topics but if you have any thoughts or want to further discuss, feel free to email me or tweet at me and happy to!

Other papers/things I’ve read

• Generating 3D models of clothing from 2D Images - There are multiple obvious use-cases for this research, but the maybe under-the-radar one is for use on avatars and digital celebrities. One of the bigger issues people likes CJW (creator of Shudu) has talked about is creating digital clothing. Miquela has historically had similar problems and thus their approach has been often not digitally recreating the clothing at all.

• A dataset for facial recognition in cartoons - Here are some thoughts I tweeted related to this. TLDR is that while these results actually weren’t great, it’ll be interesting to see more datasets emerge surrounding cartoons, as I wrote about 4 issues ago.

• FaceSpoof Buster - This paper is another in a string that I have continued to read and catalogue related to either tricking facial recognition systems, or identifying the various types of spoofs within them. If you’re interested in this topic I’d recommend reading through some of the related works within this paper.

If you have any thoughts or feedback feel free to DM me on twitter. All of my other writing can be found here.