One thing I have seen frequently with many of the biotechs I advise, as well as at Recursion where I worked for close to 8 years, is an undervaluing of the importance of investment in sound software engineering. Even companies that are founded upon a core belief that technology and science together will unlock the next wave of discoveries find themselves believing that the engineering systems are not foundational to their business. They often remind themselves and others that they are not software companies - they are drug discovery companies, CROs, service providers - and thus the software is just a means to a greater end. They are quick to explain that their central thesis, their primary hypothesis, is not centered on the software but on the science or the machine learning. They will tell you how the engineering components are all clearly solvable problems, and that the real questions - the real uncertainty - hinge on the biology, the chemistry, the machine learning. And the thing is, these statements aren’t wrong; they simply miss the point.
Just because something has low uncertainty does not mean it is easy or trivial. Just like how something having high uncertainty doesn’t mean it’s inherently difficult or complex. They are two largely independent axes that we often superimpose upon each other, conflating them as one and the same. To provide clarity for the rest of this post, let me provide some pithy, personal definitions of uncertainty and difficulty:
Uncertainty: a state of limited knowledge where one cannot determine if a solution to some problem exists or whether or not you will be able to find a solution if it does.
Difficulty: a measure of the amount of time, money and effort required to find and develop a solution to some problem.
Among modern biotechs it is thought (often correctly) that the software challenges faced have low uncertainty, because they have been solved before by other big tech corporations. The science or ML proposition at the core of the company’s central thesis has likely not been solved before, and thus has very high uncertainty. So they focus primarily on reducing the uncertainty of the science or ML, in many cases trivializing the difficulty of the other necessary components.
This isn’t wholly wrong as long as you are consciously aware of the separation between uncertainty and difficulty and are making a clear (and hopefully) temporary decision. There are many reasons why you’d want to focus first on hammering down uncertainty (read this great post for a very brief explanation). But if you persistently believe that uncertainty is all that matters, you will eventually find yourself in a very challenging position where those difficult-but-low-uncertainty problems have piled up to the point that you cannot adequately further reduce your uncertainty in your central thesis. And because those problems are difficult, it will take time, money and effort to address them, further delaying your ability to hammer down the uncertainty at your core.
So remember: uncertainty and difficulty are not synonymous. They are two orthogonal axes that you need to keep your eye on as you build your business. Because if you consistently downplay or ignore the difficult-but-low-uncertainty problems, they’ll eventually sink your ship.