Discover more from Tango.vc
Raising a VC Fund in Public
Tango.vc is now a 506c fund. Read more about what this means and how we're backing seed stage machine learning and automation startups.
TLDR: Tango.vc is raising a seed fund to invest in machine learning and robotics. Back the fund: https://angel.co/i/aPEzB
This post covers:
Why I can now do this in public
Overview of the fund, including my experience, fund structure, what founders say
How my 1st angel investment turned $25K into $1.5M
What to expect in future posts
Future posts going into detail about software, services, incubation, SPVs, partnerships, and more
This kind of announcement is rare because of the rules around fundraising. In this post, I want to explain what changed. I also want to outline our plans to help folks understand what we’re doing differently at Tango.vc. Our ideas don’t fit in a single blog post, so you can expect many more posts on this over the next few weeks.
First, the rules: the SEC wants to prevent naive investors from losing money. They require firms to register (like the overhead of going public), but have exemptions. 506b mandates that you don’t advertise and all your investors are accredited. 506c allows for advertising if you verify that accreditation. Tango.vc just switched from 506b to 506c. This means that unlike most funds, we can write about the process more openly.
You might have heard of AngelList’s rolling funds. Those are also 506c, but coupled with a “subscription” instead of a discrete vintage. Tango.vc Fund I is not a rolling fund, in part because I believe the fund mechanics make more sense in the traditional way. LPs that commit will just have their total commit amount to consider.
I’m so excited to share more about how VC works. Fundraising in public means much more content will be open on this blog.
To start, let me share what has been stuck in private fundraising conversations: my experience, track record, the fund structure, and more.
I’ve been obsessed with AI my whole life, starting with loving SciFi, from Data & Deckard to Hal & Mr. Smith. It’s why I studied computer science. I quickly learned that SciFi lies and nothing works yet, so I wanted to learn more and fix that.
At Carnegie Mellon University I got a Master in Robotics, focused on computer vision for field robotics. I worked at iRobot on the research team, which included helping the DARPA Urban Grand Challenge for autonomous vehicles.
In 2007, I decided there wouldn’t be new robots for a while (which proved correct), and founded Tipjoy, a social micropayments startup. We were backed by Y Combinator in Winter 2008. We ultimately shut down, and while at Facebook in 2009 & 10, I learned what real growth looks like.
I joined Dropbox to help them drive growth. I was #22 as one of their first product hires. Dropbox grew 12X in 2 years. I helped build data systems and run dozens of experiments around sharing, referrals, and onboarding.
In 2012, my 2nd startup YesGraph focused on machine learning over social graphs. We were backed by A16Z, Accel, YC, NextView, Bloomberg Beta. Lyft acquired YesGraph in 2017. At Lyft, I focused on user acquisition, building machine learning systems to manage $250M+ in marketing spend. I decided to go back into robotics while earning out the acquisition and worked on scenarios and simulation at Lyft Level 5, their autonomous driving division.
In late 2019, I started angel investing full time. Those early investments are rolled into Tango.vc Fund I, started in the middle of 2020.
Track Record: Cruise Automation & Powerlaws
When talking about track record, it’s hard to overstate the degree to which returns look like a powerlaw. This means a few top investments determine the whole fund outcome.
My very first angel investment was in Cruise Automation. It returned 62X, turning $25K into $1.5M. This was in just a few years, representing 294% IRR. If I had 19 other failed angel investments that literally netted $0, the overall return would still outperform most funds.1
The fund has made around 20 investments, which you can see below. At the seed stage, it is very hard to predict what might happen long term. That said, it’s worth exploring this first investment to understand how I think and work.
At iRobot in 2007, I worked on the DARPA Urban Grand Challenge in 2007 and on adding better DSPs for video processing to bomb disposal robots. When Justin.tv was announced, I got in touch with the team. This was one of my first connections to Justin Kan, Kyle Vogt, and the larger YC community. It helped me push to leave iRobot, found my first startup, and go through YC in the next batch, W08.
When you’re an expert in a field but don’t work on it for a while, ideas accumulate. At Facebook and Dropbox, I learned what high scale web systems look like. This combined into a novel strategy of scaling training data for autonomous driving. I was telling my friend Aydin Senkut from Felicis Ventures my ideas in 2013, and he told me Kyle was working on a similar approach at Cruise. (Thanks Aydin!)
I met up with Kyle at The Creamery and asked to invest the next day. Keep in mind this is before YC, when he had a few interns and some 3D printers. This access matters: the last investor joining at Demo Day would have 80% lower returns, despite also investing in the seed stage. Around Demo Day W14, I wrote publicly about my ideas, keeping Cruise’s name out of it. Here is that contemporary blog post, published mainly to help with recruiting (which worked).
The rest is history, with a strong Series A and an early, enormous exit to GM.
What are the lessons here?
Understand the tech to see the connection between video streaming on the web and high scale ML training
Use your network to get access to these great opportunities
Pounce on great opportunities when they present themselves
This is how I’m running Tango.vc.
What Founders Say
"Beyond just being a very clear and independent thinker, Ivan was constantly sending me highly qualified candidates during the early days of Cruise - exactly when I needed them most." Kyle Vogt - Cruise cofounder
“Ivan’s experience shows when you sit down with him and talk about your business. He gets to an understanding of the business quickly and then provides multiple contextually relevant ways he can help you succeed. He is the type of investor you want on your side.” Hiten Shah - FYI cofounder
"Ivan has been a dream investor for Zitara. He's generous and proactive with useful connections to clients and industry players, and has quickly earned our trust as a thoughtful and experienced business and product thinker with a uniquely clear understanding of ML technology development. Ivan is one of our go-to investors as founders when we're looking for honest feedback on real problems." Shyam Srinivasan - Zitara cofounder
"Apart from the investment, Ivan takes time to deeply understand the product and brings pragmatic go-to-market ideas to the table. His network has helped us a lot in business intros, backchannels with investors and angel connects. If you're building developer tools, highly recommend getting the initial check from him." Jai Pradeesh - DeepSource cofounder
Our Fund I target is just under $10M with 20% carry, and 2% fee average, front-loaded. Currently, our check size to startups is around $100K. This smaller check size is very agile in co-investing with other leads, compared to writing $1M checks and competing with other leads.
Because we’re targeting under $10M, there is a 250 investor limit. Previously, we were bumping the minimums as we got more commitments to manage this limit. Now the minimum will stay $25K for individuals and $250K for institutions and family offices. See more in this updated post about the change.
We have $3.8M committed with 62 LPs.
For funds & family offices: $250K minimum https://angel.co/i/aPEzB
For individuals: $25K minimum https://angel.co/i/xXmYi
We held an open LP Q&A, and here are the notes.
My anchor LPs helped me raise enough to start deploying quickly. I think it’s an incredible signal of support when my past investors in my startup back this new fund. This includes Roy Bahat from Bloomberg Beta, Trevor Blackwell from Y Combinator, and Lee Hower from NextView ventures.
Because we had LP commitments from my past investors right at the start, we’ve been raising and deploying in parallel. I set 3 capital calls to make it even easier for friends to access. We just had our 2nd, so when an LP commits, it means wiring ⅔. The last ⅓ will be called in H2 2021.
Companies & Performance
(Stealth) - Resilient hosting
Albedo - high res satellite imagery
Almanac - Docs & workflows for async collaboration.
Climate Robotics - Robots to fight climate change
DeepSource - Automated code reviews
Depict.ai - Automatic product recommendations
Edia - AI for math education
Faction - driverless delivery
Fathom Computing - 10,000X performance AI chips (exit)
FYI - Search docs and manage access
Freshpaint - Heap autotrack with Segment syndication
Hypotenuse.ai - Automatic product descriptions
Locomation - Autonomous trucking
Quartz - Computer Vision for construction sites
Red Leader Tech - HD Lidar
Rhetoric - Spellcheck for speaking (incubation)
Stoke Space - Sprinter vans for Space
Teleo - Heavy construction vehicle teleoperation
Zitara - Enterprise battery analytics
Because this is a public post, I can’t release as many details about individual companies and their recent performance. I invested at the seed stage for each of them within the last 18 months. 3 have raised on SAFEs at higher caps representing 42%, 52%, 205% increase in enterprise value. Fathom was acquired. The companies want the details here outside a public blog post, but I can share more in a conference call. 👉 https://airtable.com/shrQnGGXbSbJ0zDOH
Tango.vc incubated Rhetoric and has common stock, making it special. The enterprise value created is already more than 3X the $1.1M we’ve deployed, which highlights how exceptional incubation economics can be. A future blog post will go into more detail about this math.
Across all our seed investments, here are some of our coinvestor angels, funds, and individuals.
Automation accelerates progress, which helps build enormous defensible businesses. We believe investors need to deeply understand technology in order to predict which will power those companies. This technical credibility helps win deals, with founders seeking investors who understand their products to help inform strategy.
Here are some of the patterns that get us excited.
Computers Can See
I first studied Computer Vision as a specialization in Robotics in 2003. Notable about that time: nothing worked! Since then, there has been a revolution with Deep Learning where tracking, classification, and segmentation are all very effective. There are very many verticals where using computer vision will enable new capabilities. Consider robots that don’t interact with an environment, but need to localize and move from A to B without hitting obstacles. This includes all AVs, delivery robots, and most drones. Solving computer vision means all these applications are now possible.
Computers Can’t Read… Yet
Can a computer analyze this paragraph and answer this question? Probably. Can it understand the context and infer related meaning? Probably not. There is currently a renaissance in natural language processing which enables whole new applications, stemming from the same progress in deep learning that helped computer vision. But text and conversation still require a great deal of common sense reasoning and knowledge. This means some new applications will be possible, while others will only become available years from now. A good example is marketing, where so much of the work is writing: ad copy, blogging, landing page copy, support email & chat, a knowledge base, and more. I soon expect to see teams of humans and machines, also known as “centaurs”.
Computers Can’t Touch… Yet
Manipulation is the part of robotics where the robotic systems interact with the world. Think about your vacuum cleaner bristles that get stuck with hair, the clogged drain in your dishwasher, or the paperjam in your printer. These systems are expensive to develop and prone to errors. It’s no coincidence that the robots we most commonly see don’t actually touch their environment. An autonomous vehicle’s interaction with the world is just the tires. A drone has its rotors. In both of these cases, the robot needs to move from A to B safely and efficiently (a perception problem), but can’t do much when it gets there. This will change, and when it does, the application space for robotics will explode.
Machine Learning Dev is Still Hard
There is an enormous difference between research risk and execution risk. It used to be the question of machine learning systems was whether something might work or not at all. Now enough research and system development is done so that execution risk dominates. This means that the tools we use to build machine learning systems matter more than ever. But it’s still too hard. Analogously, databases went from rolling your own with specialized staff to standard open source solutions with a cornucopia of tools to help. We expect ML tools to follow this pattern: prototyping, collaboration, deployment, and monitoring. Some tools will be new, like managing training data or finding subtle errors in live models.
Vertical vs Horizontal
New technology can unlock multiple markets. Startups could take on a broad area directly, or sell to developers at other companies. It could also be that a new technology needs to be paired with customer and market insight, where a vertical deployment will win the space. There isn’t a general answer but it puts the other technologies in context. Maybe it isn’t a new robotic hand, but a custom robot to process recycling. Maybe it isn’t a developer-facing API, but an end application service enabled by the same technology.
There is so much more to say. Here are working titles for future posts, and I’ll link from here when ready.
Fundamental math behind VC funds
Why funds never hire, and how that affects diversity in VC
How Tango.vc Fund II will be different
Why management fees should be spent on software and services.
How Tango.vc will fill the gap in seed stage recruiting
The math behind the incubation model for VCs
Back the fund: https://angel.co/i/aPEzB
Want to learn more? https://airtable.com/shrQnGGXbSbJ0zDOH
Subscribe to get future updates.
2013 Vintage top quartile VCs are above ~30% net IRR. To highlight the incredible cruise return, consider this in a full fund context with 19 other failed $25K investments. That would mean $500K invested, instead of just $25K. If the return profile were the same, the IRR would be 39%, beating 80% of 2013 vintage funds. See more here.
Past performance is not indicative of future returns. There is no guarantee that any current or future fund will achieve the same exposure to, or quality of, startups held by any existing fund. This post and the information contained herein is provided for informational and discussion purposes only and is not intended to be a recommendation for any investment or other advice of any kind, and shall not constitute or imply any offer to purchase, sell or hold any security or to enter into or engage in any type of transaction. Any such offers will only be made pursuant to formal offering materials containing full details regarding risks, minimum investment, fees, and expenses of such transaction. The terms of any particular fund, including size, costs and other characteristics, are set forth in the applicable constituent documents for such fund and may differ materially from those presented in this presentation.