Lucas Siow, Co-Founder @ ProteinQure,
Aug 13, 2018
On June 7th, ProteinQure graduated from the quantum computing stream of the Creative Destruction Lab. While this marks a significant accomplishment, it is only the beginning. We’ve raised our pre-seed, begun hiring and are moving quickly to build the next generation of computational drug design tools.
The Creative Destruction Lab (CDL) is the premiere incubator in Canada and one of the best in the world. With mentors and participants coming from across the globe (including many who fly from NYC and Silicon Valley). This was the first year they had a quantum computing stream. The quantum stream had the highest attrition rate of any stream; not everyone admitted in to the CDL makes it through all 10 months. We are proud to say we made it!
The potential returns (both extrinsic and intrinsic) from working on a deep tech startup are great, but founding a startup is hard. And doing it with nascent technology comes with additional challenges. So far I have learned three personal takeaways for creating a successful quantum computing (software) startup.
- If your problem is worth solving (hard and valuable) it is very unlikely quantum computers are going to help solve it.
- You need to talk to customers.
- Your job is to find a lot of good advice and filter 50% of it.
The unfortunate thing about hard problems (Rule #1)
If your problem is worth solving its very unlikely that quantum computers are going to help you solve it.
The first issue with any startup is finding a problem that is actually worth solving. This is especially difficult for a quantum computing startup. You need to be working on one of those ‘can’t live without the solution’ problems. If your customer would only ‘like’ to be able to solve the problem you are working on, then it doesn’t justify making a quantum computing startup to solve it. The return won’t justify the risk of using these emerging technologies. If you are only going to make something 50% faster or cheaper, try a different approach. To warrant a quantum approach (or hybrid approach) you should be looking to enable something novel or for at least 10x improvements. Don’t attempt a moonshot to just build a better mousetrap.
The more common problem that we’ve seen is that people pick problems where quantum computers are not going to help in the near term. If its going to take 10+ years of hardware improvements to have viable product, you aren’t in the sweet spot for good venture backed projects.
Some examples of things that are NOT good reasons to try and use quantum computers for your particular problem:
- There is a lot of data and finding useful relationships is hard.
- The data is sparse/inaccurate. Often because it is hard to collect.
- Standard ML, regression, statistical techniques etc… haven’t worked so far.
- The problem happens to be computationally intensive.
The issue with any of these is that near term quantum devices are only good for very specific problems. Some day they might help with a much more general class of problems. But, just because you cannot make headway with classical computers today; doesn’t mean you can with quantum computers. In fact, classical algorithms are very good for most things these days. So if they aren’t working very well the likelihood is that the challenges are much deeper than you suspect.
Quantum computers also don’t help you acquire better data sets or teach you what causal relationships in the data exist. Domain expertise and industry relationships are often a much easier (if less cool) way to make headway on your problem.
If your problem is worth solving, make sure you’ve fully explored the non-quantum ways of solving it.
That being said, there is no reason to be pessimistic. Many high value problems are likely to be amenable to quantum computers and with enough ingenuity you can find ways to make them relevant commercially in the short term.
Know your customer (Rule #2)
You still need to talk to customers.
This is a problem ProteinQure faced early on in our existence. We knew that protein folding on computers was a very hard problem. So we thought that if we were just better at it we could worry about monetizing or creating a product later.
It turns out, that because the problem hasn’t been tractable historically; Pharma has done a good job of building billion dollar drug discovery processes that don’t actually require being able to fold proteins on a computer.
ProteinQure needed to figure out how we could take our technology and try to fit into the industry’s established workflows. Eventually computers will be surely integrated into every step of the drug discovery process. But early on you have to find ways to fit into the existing paradigm. As the tech becomes more established you can completely reimagine new ways to do things, but trying to do that today is an unlikely road for success.
Don’t just build algorithms for the sake of it. Understand what are the pain points for your customer that they would love to solve. Then find the intersection with where your tech might be useful. In our case it only took about ~10 conversations with drug discovery experts to better understand where we might fit into the landscape. Here are just a subset of problems we heard about:
- novel peptide drug discovery
- target and binding site identification
- lead optimization (e.g. stability, toxicity etc…)
- biosimilar drug development
- assay development
Four of these weren’t even on our radar before we began our customer conversations. And it wouldn’t be feasible to chase down every possible use case. So as a company founder its your job to prioritize and make sure you don’t just fall in love with your first idea. The two key criterion are feasibility (is this a problem we can solve in the near term) and value creation. Keep in mind that solving some of the easier problems will allow you to build knowledge and unlock the hard ones.
Once you start really building your tech it will require singular focus. So do this step as early as possible. Also make sure you onboard domain expertise to help execute. These can be in the form of employees or advisors, but there will always be a lot you need to know about your industry when building software. Pretending that best technical solution always wins is an easy way for great scientists to turn into failed founders.
No one has done this before (Rule #3)
Your job is to find good advice and ignore the other 50%
Quantum computing startups are a pretty new thing. So no one yet has really had the experience of creating/growing a commercially successful quantum computing startup (though there are plenty of people on the cusp). While there is never a straightforward path for any startup founder its doubly hard when there aren’t case studies for you to follow.
That means you need to get advice and be prepared for two unfortunate facts:
1) most of it will contradict some other piece of advice you’ve received
2) operator advice will be context specific, but people won’t always say that
That doesn’t make getting advice a waste of time. Your job is to try and understand where people are coming from and whether you fit the same context. Trying to follow everyone’s advice (often in the form of pivoting to the latest piece you got) is tempting for many first time founders. After all, you may be getting advice from CEOs, successful venture capitalists, world class academics or the occasional astronaut. Those people are going to give you help based on their experiences and extrapolating experiences to a field that no one has ever worked in.
Avoid the temptation to just listen to what you want to hear or to simply follow what you heard last (recency bias). Focus on synthesizing the information you receive. I have three tips that reflect what I found useful when trying to find advice.
Look for people who’s advice is different than exactly what they did. Look for people who can imagine how your particular circumstances would change how they think about the problem (even if they faced a similar problem).
Find people with domain expertise who are passionate about tech. It will be easy to find great entrepreneurs who want to give advice without knowing your industry at all. And people from the establishment will often privilege the status quo. Find the sweet spot of those who know about where your playing and are imaginative about the future. Though basic skepticism is a great way to pressure test your assumptions.
Finally, ask for another connection. If you ever find a person who gives great advice, ask them if there is someone else you should meet. Networks are a huge part of any startup ecosystem and smart people tend to hangout with each other.
ProteinQure is proud to be part of Toronto’s (& Waterloo’s) successful quantum computing ecosystem. There is no better place in the world to be working on these technologies and trying to advance the frontier of what is possible.
We hope to see many other startups come here and have success as well. This post is intended to help the next group avoid some of the early challenges that we all have to overcome.