My failed startup: Lessons I learned by not becoming a millionaire
How do you find a co-founder?
Back in early 2019, I left my job at IQVIA to join the London cohort ofEntrepreneur First. EF is an amazing incubator / accelerator program that has given the world the likes ofMagic Ponyandhundreds of other exciting companies.
‘Entrepreneur’ what?
The main thing that sets EF apart from the dozens of similar programs is they focus on the earliest possible stage of companies: sole founders. They take in a hundred talented and ambitious individuals in each cohort (actually almost 200 when you count the other European EF sites: Berlin and Paris).
Then the good people of EF will make everyone go through a weird, speed-dating meetsLove Islandtype of co-founder finding process.
In this phase, for weeks on end, you are simultaneously reading the CVs of super interesting cohort members and having ludicrously blue-sky chats with them, about potentially world changing companies the two of you could build. By the end of this process, you should end up with a co-founder.
EF also gives you a stipend, a place to work and a ton of high quality material on the basics of entrepreneurship and startup building. But I think the real USP of EF, is the access to this pool of pre-screened, highly talented and motivated people, who all left their (sometimes lucrative) day jobs to become members of the cohort and build companies.
From beers to business
As a naturally skeptical person, I wasn’t sure what to think about this process. I did what was expected of me: read CVs, talked to people, came up with outlandish ideas. But I didn’t really click with anyone.
Finally however, after many beers and chats, I found my co-founder, with whom we just had enough in common (interest in healthcare and machine learning) to make our discussions fruitful, but with whom our skills were complimentary enough that we could take advantage of having two people on the team.
This latter point is in fact quite crucial: too often, we’re naturally attracted to people with similar interests, skills and thinking to our own. This is great for friendships, but can be devastating in a startup where you want the co-founders to cover each other’s blind spots.
Both my co-founder and I are “techies” in broad terms. But it was quite clear from the beginning that I am more of a CTO type who loves to build, whereas he liked the process of selling and pitching, which is quite important because that’s pretty much all you do as an early stage startup CEO.
Fortunately (as this is quite rare AFAIK), we both had a good share of the skills of the other’s role too. This really helped with both the technical and business discussions between the two of us.
Do you have a $1 billion idea?
Something we all learned fairly quickly in EF was the economics of venture capital. It’s really quite simple and goes something like this:
Venture capitalism 101:
Companies from the first group are called successful medium sized business. The latter ones are called unicorns: startups in hyper-growth mode, overtaking and disrupting a whole industry like AirBnB or Uber did for example.
Not all ideas are created equally
Even though probably both investment models could be equally successful, VCs want to invest in high impact firms that will change the world. Think Google, Facebook, and Twitter.
I’ll leave it to your judgement whether this is a healthy and sustainable business model. I will certainly write more about this in the future, but I will say this: VCs serve a super important function in the global innovation market and without them, the world would be different in countless ways.
The corollary of all the above is that if you go down the VC backed startup building path, you need an idea that can become a 1 billion-dollar company. Yes, that’s a billion with a capital ‘B.’
Start me up!
We spent weeks ideating with my co-founder, but after dozens of discussions we came back to the idea I arrived to EF with. It was the only one that seemed simultaneously ambitious and high impact enough that a VC might take a liking to it, while also having the potential to grow into a multi billion-dollar company. Also, neither of us hated it, which is quite important, I heard.
The idea
I had this idea for a company for months before applying to EF:
I couldn’t wait to tell all the people at the cohort about it, so I did. And they all made the exact same face you just did when you read the sentence above.
The problem with deep tech ideas (besides not fitting on a napkin) is that for the most of us not well-versed in the particular field, they are often indistinguishable from vaporware and snake-oil.
You really have to know a lot about healthcare informatics, medical machine learning, federated deep learning and also the financial incentive structure of the US healthcare industry to properly evaluate that one sentence above.
If you do that, it will not only make a lot of sense but you’ll find that it’s offering one of the very few viable long-term solutions to the impossible trade-off healthcare faces in the 21st century:
This is bad news, because without the deployment and proliferation of AI,healthcare is doomed. Another indication that the above idea might not be the worst ever is that Google Ventures already backed anamazing teamto do exactly this. Yeah, finding that out wasn’t fun…
The idea explained
Probably it’s easiest to explain the idea for our startup by enumerating some of the relevant problems that plague healthcare currently:
Once you write down these six simple facts about today’s healthcare systems, it really doesn’t take long for the light bulb to start flashing.
We need to build a marketplace where providers can sell access to their data without compromising patient privacy and where vendors can get access to this data to learn from it and train models on it. Furthermore, once vendors have a model trained, they need to be able to easily deploy it to network of hospitals, where those models could start to generate useful (and potentially life-saving) predictions.
The answer is quite remarkably simple, and it’s been powering all sorts of technologies (like predictive texting on your phone) for years. It’s called federated learning and it’s first real-world, large scale application came from Google (as far as I know) followingthis paper. Then,many–more–followedand amazing open source libraries were released likePySyftbyOpenMined, orFATEbyWeBank.
Federated learning’s simplest explanation can be boiled down to a single sentence:
Suffice to say that this idea has found applications in several security and privacy conscious industries, but its impact on healthcare will truly be transformational. Have a look at these two recently published Nature articles to see what federated learning holds for the future ofmulti-institutional medical studiesanddigital health applications.
If you’re curious about the technical details and implementation of our idea, make sure to check outthis pageabout the MVP I built, or watch my demo of it below.
To recap:
Here’s a flowchart of the smallest possible incarnation of our platform with just two hospitals connected to the federated data network.
The rollercoaster life of a founder
The EF program is structured according to a fairly well thought out schedule that gives a nice framework for working on your idea.
Hm… That was depressing to learn, but don’t worry, it’ll get worse.
The up
As anyone who’s ever tried will tell you, building a company is like being on a rollercoaster ride. Sometimes you have days that feel like none of the days in your previous office jobs. You feel amazing, fulfilled and working on something truly meaningful (at least to you).
Even better, sometimes things just work out. Like when we started to ask huge and well-known US research hospitals if they would partner with us (two random unfunded guys from London) to build the world’s first federated machine learning platform for healthcare and they just said “yeah sure”.
Or when the second and third meetings started to occur with these teams, and the term “letter of intent” started to fly around a lot. We felt invincible. Even if it took more than a hundred Zoom calls and thousands of emails sent out, we thought we are clearly making progress with business development here.
The top
Then the rollercoaster continued and took us to this amazing top. Armed with our letters of intent and signed project proposals with these hospitals, EF’s IC said they’d invest in us and just like that, we got the pre-seed funding.
Roughly 30% of the initial EF cohort make it this far so we felt great. Even though we knew the work is just about to get a whole lot more challenging and crazy, there’s a moment when you have your first little success and you start quietly wondering to yourself:
I was also super excited that I get to finally build again after months of not coding. Don’t get me wrong, conducting BD calls, writing marketing materials, technical documents, grants, drafting SOWs and LOIs was extremely educational too and they are necessary and unavoidable parts of every company formation. Nonetheless, I was itching to actually get back to the thing I love doing and I’m reasonably good at.
The down
After the pre-seed investment, I was 100% on building and my co-founder was 100% on business development. We only had 2.5 months till demo day to:
Our plan seemed straightforward. I am going to take care of the first one and my co-founder is doing the second. Even though we knew the second is probably dependent on the first being ready, we agreed to work as hard as we can and try to have our first customers by demo day in late September 2019.
However, the cracks started to show quite early on… The business development process went from painfully hard (pre-IC) to near impossibly hard (post-IC). Many people told us that letters of intent are worth the same as toilet paper, and we should try to convert those into signed contracts for paid or unpaid trials ASAP, so that’s what we tried to do.
Despite my co-founder’s heroic efforts however, after months of negotiations, meetings, project proposals and more meetings with US hospitals, every single lead of ours simply dried up. Emails went unanswered. Crucial final meetings with the key stakeholders did not happen.
When the discussions turned real, and when hospitals realized that for this fancy-sounding research project they’d need to actually provision resources for us so we can integrate our software into their IT systems, they went cold turkey on us.
The small up before…
We realized that we made our own lives impossible by trying to aggressively innovate on two fronts at once:
Even just one of these goals would have been insanely hard for a two-person team to pull off in 6 months, but doing them together was clearly way too much. So we pivoted and said:
This wasn’t the worst idea to be honest. Pharma companies already have longstanding relationships with hospitals and outpatient clinics, as these connections are essential for recruiting patients for new clinical trials.
Healthcare – as we found out to our own detriment – is built on trust and slowly evolving long-term partnerships. There’s very little place for disruptive technologies, especially pushed by small companies. Due to the heavily regulated and risk averse nature of healthcare, you can have the best tech product and it’ll still be extremely hard to sell (this was confirmed to us by numerous companies on the market).
So we thought, let’s sell the technology to the innovation leads and heads of data science of large pharma companies. Then they will take care of getting the tech to the data, i.e. inside the hospitals.
The crash
This hasn’t really changed the MVP I had to build, only the way we’d demo it to users potentially, so I proceeded further with full steam.
My co-founder in the meanwhile was trying to line up as many demo opportunities as he could, so we would have a good chance of converting at least one or two of those into an unpaid trial by demo day. Then we’d have our hard-won“traction”in our hands for our discussions with VCs.
I finishedthe MVPand we started to demo it to large pharma companies. Quite miraculously, none of them thought it was terrible. In fact, some of them even murmured something positive about it during the call.
But crucially, none of them jumped out of their chair singing Hail Mary, thanking us for finally solving their long standing hair-on-fire problem. Instead, they all said, they’ll get back to us after they had some internal discussions.
Although sales cycles in healthcare are notoriously long, and selling to pharma as a tiny startup is near impossible, we knew we missed the mark. We could simply feel after these calls that we aren’t solving any of these teams’ top 3 problems. What we offered was definitely in the top 10 or 15, but not even close to top 3.
And then, to make things worse (or better?), I learned through a series of honest chats with my co-founder that his heart wasn’t really in federated learning anymore.
Turns out, it’s a pretty big problem when the CEO of a company is not emotionally invested in and hyped up about the very thing the company is selling. It’s especially bad if this happens literally weeks before he’s supposed to stand in front of a hundred VCs and convince them that our business is the one that’s going to change the world and which is worth their £1M investment.
Operation salvage
I totally empathized with the frustration of my co-founder after he struggled for months to sell without much to show for it… However it felt rather bad to give up on the idea at the very last minute.
Granted, our traction was practically non-existent at that point, but I suspected from the beginning that selling this product will be incredible hard and might potentially take years. We also knew for a fact that selling our product isn’t impossible, as we have foundcompanies that managed to get into US hospitalsand others thatworked successfully on something similar to our idea(although both with incredible funding behind their backs). So personally I wasn’t disappointed, and I wasn’t that surprised with our lack of progress. But clearly my co-founder was and it was him who had to muster the energy day after day to try and sell this thing.
It was pretty clear to me that just like you cannot make someone love another human being, you cannot make someone feel passion for something for which they simply don’t. So reluctantly, I agreed to move on and use the rest of our pre-seed funding to build something else for the next demo day (which was scheduled for 6 months later).
We told our plans to EF (who were really nice and supportive about it), then proceeded to conduct one of the most painful 4 weeks of my life. We brainstormed in the Google Startup Campus, the British Library and countless pubs and cafes of London for days on end. We were desperately trying to come up with an idea which we both felt good about, meaning we would be willing to dedicate a couple years of our lives to it.
Whiteboard sessions came and went, beers and many more coffees were had. We tried for hours each day, but one of the two following scenarios kept happening:
The end
After weeks of this, I proposed to leave the company and sell my shares to my co-founder. Again, EF was incredibly nice about this and signed all the necessary contract modifications that allowed me a clean exit.Then, on a rainy October day, six months after joining EF, my co-founder and I said goodbye in a typical London pub and that was the end of my first startup.
As I walked towards the tube, lots of thoughts and feelings were rushing through my head… None of them particularly positive or pleasant, but I knew I made the right decision.
Takeaways
I probably learned more about myself and the world in those six months than in the preceding 2.5 years in my typical 9 to 5 corporate data-science job. I couldn’t possibly summarize all of it here and some (or most) of it might be completely trivial to some readers. So here’s a list of my top learnings.
