Insights into building, fundraising for and bootstrapping deep technology businesses post-multi million dollar exit. Part 1.
I am not sure if this is interesting. Let me know if I should do a part 2, and what you might like to see.
Context
Hi, My name is Adam.
Over the last 10 years I have founded multiple startups, with the most recent raising a $20 million dollars (2021), in which I negotiated an exit to free up finances to work on new projects (Low 7 digits). As well as a few smaller acquisitions and exits along the way ranging between $5000 and $250,000. The common thread is all of these companies had a large technical component, whether novel invention or huge engineering effort. So, I figured I would brain dump some learning, and talk about a my mental framework going forward on new projects.
If you can't ship a deep tech product yet, ship deep tech demos.
I am an engineer and mathematician, starting businesses allows me to pursue my curiosity. I find great difficulty working on things that don't scratch the part of my brain that likes complex problem solving, this can be toxic if left unchecked, but a powerful differentiator if harnessed.
When your premise involves longer horizons and potentially fundamental research, difficult to scale solutions (from an engineering and compute cost perspective), and a lack of in market examples it's really easy to get stuck in build mode, and never launch mode.
Technical Demos can really help this, think of this as an experiment in seeing how other people engage with your novel technology. If you are funded with a team, consider polishing the interface this for press, investors and places like hacker news.
If you are solo, ship something quickly that you can build in less than a day. You want to excite a small number of people, the goal is to maximize the number of people who can give you insights you might not expect. Tech demo's are cool because they don't need to exist forever, so the engineering effort to get them out the door is a lot smaller. There are a number of frameworks for rapidly building demos of deep tech. When working solo, I use https://streamlit.io/ to get user feedback on a new deep learning model or algorithm I have been working on. It takes 20 minutes to ship something.
Fundraising with Deep Technology
Don't over think this. Raising capital is easy if you have cool shit to show off.
My most recent examples, had the advantage of going through Techstars accelerator which definitely opens doors to bigger VCs, but I think this holds in general as we had similar (slightly worse terms) deals prior to the accelerator.
The pitch deck for my most recent seed round (2 million USD in 2016) was a story about how our technology had evolved over the last 18 months followed by a vision of the future, and a demo the investors could play with, seeing the very serious partner light up as he used our deep learning technology to generate unique piano and drum and bass beat pretty cool. Head bobbing VCs are the best VCs.
There are two factors to fundraising seed VC with deep technology.
- A compelling story.
- An indisputable demo.
Now, a large number of VCs don't do this sort of investment, however it actually tends to be a positive filter, there are a lot of ex-engineers in venture financing you'll be talking to your people.
Starting from Scratch and Bootstrapping
The pathway for someone avoiding financing is a little bit different, often when you are working on a really tough technical problem, smaller things will fall out of it that are less than the big vision, but still independently valuable. Context switching isn't free, look for components that require minimal non-domain engineering to get a proof of concept out. You'll find these are not a lot different from the technical demos mentioned above, just polished.
As an example, whilst building a deep learning image editor. I built a cropping utility for automatically finding the correct region to crop, prior to the service sale this was making $1500 a month with a one page upload form and an API for anyone wanting to integrate it. The acquirer was a client.
Whilst building my most recently company (in the AI music space) we launched an automated music mastering service, which again is a simple (but very pretty glossaudio.com) upload form.
The point is, often the components of you're complicate system are independently shippable. Optimize for these opportunities, ship for building a community, ship for finding new use cases, ship for press, and ship to see why you're shit is broken.
Your stuff should be cool to people, embed yourself in communities as positive actor
If you're demo isn't super cool to drop in a relevant communities with a "Hey, I am building an algorithm to do X, it's still a work in progress but I think it's pretty cool". Then, it's probably a signal that you might be missing important insight, there are futurist in every domain - your shit should be dope to these people. Make legitimate friends, build cool shit.
Could I by any chance book you for a call to chat about everything and nothing? I also could rumble for hours about all the successes in my life, though you certainly have me beat by numbers in key areas.
My core competences I usually state as being: crowdfunding, physics, e-commerce, marketing, software engineering, machine learning, logistics, choreography, poetry & stand-up comedy.
Wondering if you could add some more to my to-do list.
If not a call then let's start an email thread.
Hi,
This is a great article, thank you for writing it. Interesting points in the bootstrapping section. My MVP is a desktop app right now, how would you imagine I make it into an indisputable demo.
A demo could just be a video demo right, or does it need to be interactive?
Indisputable is basically a proxy for impressive or something they haven't seen before.
From the non-raising perspective, you learn a lot by seeing how people interact with your technology. Imagine if GPT-3 didn't have a demo.
From like, a demo in a VC. You should have both ready, video first (and sent them after/prior) and if read the room on an interactive demo, since they are inherently more clumsy. But, amazing if they land right.
Hey 👋
This is great. Thanks for sharing! Especially keen to hear more thoughts, experiences or ideas around the smaller scale products like the image editor and the audio tool.
I’ve been dabbling with FastAI and kaggle stuff on and off for a while now trying to figure out how to /where to utilise some of the tech for small scale apps or services.
My DNN and ML skills are not strong enough yet to build custom solutions other than simple classifiers etc but I’m also a bit stumped about what might be useful.
Anyway, please do write part 2 I’m looking forward to it and any other insights you have about deep learning products.
🍻
PS: I had streamlit laying around in a tab on iOS safari but hadn’t gotten around to looking at it closely - Holy shit! It’s looks great! 🥳
Whilst it does happen, in my experience it is rare that you need some middle ground between off the shelf and custom from an MVP. For example, if you were doing an NLP project, chances are you can fine tune a huggingface model or something for your MVP.
As things develop, things get more custom in those cases.
The other case is that, your product is a novel application of deep learning (new types of data) or a model innovation that lets a new type of interact exist (new types of models) or both. Usually these MVP are entirely custom. These are the types of projects I focus on, since I enjoy building it more.
If you're going after the latter, look for qualities in problem domains or architectures that don't require insane amounts of compute. For example, lets say you are doing a white balancing and color enhancement model. Instead of learning a massive end to end model, learn to parameterize a function that does a pixel wise transformation.
That sorta stuff will converge quicker, require less tuning, less data and help you get something out the door quicker. It might end up being better in production for a while, since it's more predictable performance wise as you build out the rest of the project.
Thank you. Very insightful and helpful.
Great read, "deep tech" journey is so much different and riskier.
In terms of labelling terminology, like the "deep tech", I have been wondering how would I label a product that adds a new fundamental element/feature to a world standardized system, like let's say HTML?