Andy Weissman's posts and talks


Part of the genius of the Internet is its ability to coordinate the actions of many disparate and geographically diverse people by eliminating the marginal cost of sharing information. In the last few decades alone, we've seen this happen again and again in areas such as education, science and commerce with remarkable results.

One question we have asked ourselves over the years is how best to apply the network model to the business of allocating capital. What other components or technologies would be necessary for a global scale community of people making investment decisions?

Numerai is a hedge fund managed by an anonymous community of data scientists. It encrypts its data and allows anyone in the world to continuously apply machine intelligence to the set and anonymously submit price predictions back. Numerai turns these predictions into trades and compensates the best performing models with bitcoin.

Today, we're excited to announce that USV is leading the Series A round of financing in Numerai’s management company.

For an activity so heavily dependent on the efficient transmission and interpretation of information, the business of allocating capital has been slow to adopt the network model. Meanwhile, the pace at which the field of machine learning is advancing is rapidly accelerating. Between breakthroughs in our understanding of the science, platforms such as Kaggle, the Netflix Prize and a wealth of free online learning tools, there is an increasing supply of talent tackling all aspects of computing and data analysis. But these thousands of data scientists around the world with expert knowledge in machine learning are unable to apply that expertise to finance for lack of high quality data and trading capital.

Numerai attempts to fill this gap by acting as an interface between the machine intelligence community and global capital markets with an open-access, open-participation model. Anyone with an email and a bitcoin address can download the company’s data for free and train machine learning algorithms on it.

Every participant approaches the data set in their own unique way, producing many different solutions to the same problem. Numerai then combines each of these approaches into a single meta model, which dictates how to allocate the assets in its investment fund. In return, users are compensated in bitcoin in proportion to how much they help improve the meta model.

By encrypting its data set before releasing it to the public, Numerai turns the challenge of price prediction into a purely mathematical problem by removing the influence of human bias upon the results. Participants don’t know which securities they’re modeling nor what their predictions mean; only whether their model is performant or not. At the same time, Numerai itself does not know what algorithms the data scientists are using; their code and intellectual property remains theirs. Richard Craib, founder and CEO of Numerai, calls this a trustless relationship between Numerai and the data scientists, facilitated by encryption and anonymity.

Numerai is thus not a search for the best model; it is a platform to synthesize many different models, an invisible collaboration to build the meta model. At scale, Numerai’s fund is exposed to every model and a diversified portfolio without the risk of relying on a single and imperfect model.

This is a new kind type of capital allocation business that also has network effects - from the community, from their individual models and from the collective meta model. These network effects result in an open access fund that will generate more intelligence than a closed system built on a pre-internet organizational design.

In the year since its launch, this appears to be working: 7,500 data scientists have created over 500,000 models representing 28 billion predictions. This may be the largest ensemble of stock market machine learning models in the world.

Numerai describes this directly and succinctly: “The world doesn't need another hedge fund, it needs exactly one hedge fund that's powered by every artificial intelligence.”

More about Numerai can be found at


For all the promise of digital technologies, medical care today is still largely provided in the same unstructured manner as it always has been. When you need to see a doctor, there are typically 5 or 6 steps you need to take before a potential outcome: finding the doctor; finding time to schedule the appointment; visiting the doctor; getting a diagnosis and prescription; visiting a pharmacy and paying for your medication. Each one of these steps, while necessary, has to be done with live communication and often in-person visits that requires the parties involved be present at the same time. This can be not only inconvenient, but also an inefficient and unnecessary waste of time for both the patient and the provider. These processes account for a material percentage of the costs of delivering medical care.

We believe digital processes, accessible via a mobile phone, and in an asynchronous manner, will over time transform how medical care is delivered, and how much it costs.

Nurx is a service that today prescribes and delivers medication from a mobile app and in doing so is redefining the doctor-patient relationship and the practice of primary care. We are announcing today that USV had led the company's latest round of financing.

Today, Nurx delivers birth control and Truvada for PrEP, conveniently and quickly, with the user in charge at every step, done via a simple text based mobile interface. We believe this is a radical new way of providing care -  by  changing unstructured interactions into structured care, by shifting work from MDs to algorithms where possible, by automatically and on the fly creating a portable, digital medical record. As a result, this dramatically lowers the cost to the user (to $15 or less) as well as potentially making the lives of medical professionals better by giving them more time for creative work with more patients, resulting in better outcomes.

Since launching earlier this year, Nurx has served thousands of users across 3 states - California, Washington and New York. It will be available in 3 more states this year and nationwide in 2017. Users love it. To supplement this growth, Nurx will be expanding their team to build new applications for messaging, apps, and other integrations that will come to define these new healthcare accessibility systems.

Beyond that, we believe this focused entry point can evolve into being the first point of contact for anything healthcare, giving the user as much control over their care as possible for as low a cost of possible. We are excited to be participating in one of the mobile medical networks for the 21st century.


USV Opportunity Fund, circa 2016

Over five years ago USV launched its first “Opportunity Fund” - a pool of capital meant as a complement to the core activities of our early stage funds. A few years thereafter, in 2014, we raised a second Opportunity Fund. Both of those funds have been continuously active. In light of making our first investment in our 2016 early stage fund, we thought this would be a good time to take a look back (and forward) at this investment strategy.

To date, the Opportunity Funds have invested in 15 companies in sectors such as financial technology, infrastructure, marketplaces and consumer services, located in both North America and Europe. The initial investments have ranged from $7.5 to $15 million, and we have made follow-on investments in many cases. About two-thirds of these investments were in companies in which we had previously invested out of our early stage funds.

The Opportunity Funds give us the ability to invest in more developed companies where we think our perspective can add value to the business. It also gives us flexibility to invest in companies that we missed in the early stage. As we wrote in 2011: “We hope you'll think of USV as stage-agnostic, highly-focused investors who can add value to your company.”

Importantly, these funds are designed to operate in a way that complements our core activities.

The first way they do this is by letting us support portfolio companies across a wider range of stages and at higher valuations than we can with our core funds. An example of this is the round we recently led for Foursquare.

The second way is by developing relationships with entrepreneurs over longer periods of time, which in ultimately results in an investment opportunity outside of a bidded investment round. A good example of this is Cloudflare, which was led by Brad.

The third way gives us a very real opportunity to invest in companies that squarely fit within our thesis but we missed during their earlier stages. Lending Club and Realty Shares fit into this bucket.

In many of these examples, founders felt that USV's approach and experience was relevant and we structured a deal outside of the usual fundraising process that worked for both sides.

We have plenty of capacity in our existing Opportunity Fund and are actively looking for new investments. If you think this focus is relevant to your business, please reach out to us.

The Reordering of Medicine

"This is less about the unbundling of medicine (which may also be happening) but instead the reordering of it."


"This picture of Justin Bieber getting choked at a nightclub looks like a Renaissance painting"

This photograph, and its caption, were posted on Imgur on April 17th. It quickly made it to the front page of Reddit, ultimately garnering over 5,000 comments.

As the picture made its way through my streams, I was struck by the many details: the different people, some interested, some disinterested; the seemingly different vertical levels of the bystanders; the colors. But I also found myself interested in something more: Who was the photographer? What was she doing to get that photo? Did she take others around that time? Was this taken, as it seems, from a phone?  While we're great at sharing pictures on the web, we're also notoriously bad at maintaining their context, their stories, as they spread across different networks and platforms. And not only photos, but maybe more generally, any media.

Looking to learn more, I turned to Google Image search for answers, but instead received a cacophony of results sorted by Google's ranking algorithms, linking out to celebrity gossip aggregators. Attribution for the creator was nowhere to be found.

Mediachain approaches the problem millions of creators and users encounter daily in a novel way. This team asked, “What if information about media were open, and developers could utilize technology similar to that which powers Shazam or Google Image search to easily retrieve it?”

This is precisely what Mediachain is: a universal media library.

Had the above image been registered in the network, just by seeing it in my feed, I could look up its Mediachain and find its history. Why would that matter? I might be able to find out identity of the photographer, send her a micropayment, see what social network it originated on, see the press the image received, find more images by her and any number of other things. All of this would be possible via any number of applications that a developer might build that implements the Mediachain protocol.  

How does this work? The protocol allows anyone to attach information to creative works, make it persistent and discoverable in a blockchain-based database. And because it is completely decentralized, there is no central point of control or failure. The data is maintained by participants of the network and no permission is required to contribute or access it, making it an ideal place for collaboration between creators, developers, platforms, and media organizations. It is applicable to any form of media - images, gifs, videos, written works, and also music.

Imagine if the monetization of media followed the content, instead of flowing from a platform. For example, being able to pay a musician directly by virtue of hearing their song play, no matter where you got it: from an mp3 in an email, via a link on Soundcloud, or even by hearing it on the radio.

Mediachain represents the type of permissionless innovation that we continue to support at USV.  We are excited to be investing in Mediachain, along with a16z, RRE, and other investors to advance the sharing and programming of media everywhere.

Developers who are interested in the project can find out more and get involved by joining the community on GitHub or through their public Slack.

USV Thesis 2.0

Union Square Ventures has always been a “thesis” driven firm. We maintain specific principles about the internet that guide our investment decisions. While other things like stage and to a lesser extent geography, also matter, our thesis or point of view is the primary thing that guides our decision making.

We last wrote about this a few years ago, in Investment [email protected], where we attempted to describe this view of the world. There, we also tried to describe how dynamic the thesis is, or can be. A few years later, we have a better idea of how our thesis has evolved and now presents, circa 2015.

Since USV was founded, we have focused on the applications layer of the internet. The layer that sits on top of the relatively open and robust infrastructure of the internet, the infrastructure that allows for permissionless connectivity.

Initially, the investments related to that applications layer were what we called “large networks” - that is - broad based, mostly consumer-oriented networks that could, or at least aspired to, touch many many people (hundreds of millions or more).

Brad reduced this to 140 characters a number of years ago:

This thesis brought us to companies like Twitter, Tumblr, Etsy and Soundcloud - large networks that, to this day, have proven defensible through network effects.

Over time, it became harder - and it’s still hard - for newer entrants, newer broad consumer networks - to gain scale because to do so requires them to displace the time users devote and spend on the new incumbent networks, such as Facebook.

As a result we turned our attention and applied the thesis to those services that support the larger networks - so called “enabling technologies” - that were horizontal in nature, yet also broad with respect to the numbers of networks they could potentially support.

These enabling technologies are basically businesses that provide essential services to the new crop of web companies.

These are investments such as Twilio (a communications service), MongoDB (a database service), Cloudflare (a network and security service), SiftScience (fraud protection), and Firebase (a synchronization service). And a more recent investment - Clarifai (image and visual recognition service).

Then, roughly in the 2012 time frame, we also turned our attention to thinking about market-specific networks: networks in high-value niches that are differentiated and defensible, partially because they are domain-specific. These networks generally have more subtle or less obvious network effects, precisely because they involve something more specific and tight.

These often fall into specific categories - like education or learning (Edmodo, Codecademy, Skillshare, Duolingo, Quizlet, Stack Exchange), financial marketplaces (Lending Club, Funding Circle, Circle Up, C2FO) healthcare and medicine (Figure 1, Human DX, Clue), science and engineering (Science Exchange, SimScale), the law (Casetext) and company ownership management (eShares).

These more subtle network effects also include platform shifts, such as mobile (Amino, Figure 1 or Duolingo), venue shifts (enterprise security delivered in the cloud, such as Cloudflare), and data networks like SiftScience, which delivers fraud protection by aggregating data points across thousands of domains.

Finally, and more recently we have been thinking and talking about the blockchain and bitcoin. When we analyze the network effects of the large internet platforms, it appears that part of their defensibility is through the centralization of data - user data, interaction data and transaction data.

We started to see that blockchains - by basically being a decentralized data layer - could over time erode those advantages.

So we turned the thesis to the exploration of services that could undermine larger networks by decentralizing the data asset that the large networks have. While this area is obviously early, we have made a handful investments in this decentralized layer including Coinbase (banking and brokerage), OB1 (buy and sell marketplaces) and Onename (identity).

Finally, as infrastructure providers gravitate towards the applications layer, they are underinvesting in connectivity itself at a time when the demand is growing and new technologies are available. Inasmuch as the internet itself is an enabler of creation and creativity, we believe that businesses like Veniam (the "internet of moving things") and one other unannounced investment we have made will be foundational layers for future generations of technology. So, we have also made a few investments in those telecommunications infrastructure companies with innovative technologies or business models (Access 2.0).

Importantly, the way in which we invest against this thesis is also cumulative - we don’t simply stop investing in any one area as we uncover other ones. It looks something like this, a chart of our active investments over time from 2004-present:

USV Investments by Thesis over time

To capture this image into a current version of our investment thesis, we’ve reduced it again today to 140 characters.

This is USV, 2015: 

No Stack Startups

The unbundling of the full stack startup. Where instead of being good at many things, companies can just focus on the last mile of value they provide, the one thing they can excel at better than anyone else. Maybe this should be called the No Stack Startup - services that can focus on doing only one thing - hopefully well - and utilize other services for everything else.

Eben Moglen - Snowden and... on Livestream

Moglen's talk from a few weeks ago.

He makes a bunch of points, one of the interesting ones is buried at the end, about generational change and this issue

Redefining Mental Illness

Very interesting article suggesting for some mental health problems, medications may not actually held fundamental symptoms. Instead:

"It is “vital” that those who suffer with distressing symptoms be given an opportunity to “talk in detail about their experiences and to make sense of what has happened to them” — and points out that mental health services rarely make such opportunities available."

Being able to "talk in detail" of course can take many different forms, including ways that connected Internet services can play a big role.

Why You Should Live in Ammon, Idaho

Municipal fiber - where a locality builts the infrastructure - then opens if up for service providers to build on top of it, available on a neutral basis to any operator, leading to a vibrant market for connectivity. Cities that do this: "Stockholm, Palo Alto, Rockport, Maine, and the tiny, highly-conservative town of Ammon, Idaho "

"What do Stockholm, Palo Alto, Rockport, Maine, and the tiny, highly-conservative town of Ammon, Idaho have in common? Answer: they all have dark fiber networks — thin, flexible strands of glass that are capable of carrying an unlimited amount of information in the form of pulses of light, but haven’t yet been “lit” by the electronics that trigger lasers to create those pulses and transmit them through the glass. Independent operators with access to these passive, dark networks provide their own electronics, creating a competitive, diverse, and choice-filled marketplace for unlimited-capacity, symmetrical (equal upload to download) high-speed Internet access. "