Speed of Iteration
Collison: The main thing companies screw up at the pre-product-market-fit stage is speed of iteration. If you have some meaningful, albeit perhaps small, initial set of users — and you're rapidly iterating in response to their feedback and observed behavior — that's a really good spot to be in.
At that juncture of pre-product-market fit, you should be doing everything you can to tighten that feedback cycle.
The main pre-PMF mistake: not tightening the feedback cycle fast enough.
The Fighter Pilot Analogy
Collison: There's a fighter pilot named Boyd who revolutionized airborne combat in the US from the Korean War onwards. He had this concept of the OODA loop. Previously people thought you wanted the fastest aircraft or the most sophisticated weaponry.
Boyd was all about — no, you actually want aircraft, pilots, and training that are really oriented around the fastest responsiveness and iteration to the particulars of the situation. That went on to really inform modern aircraft design. You want to be like one of these modern fighters — optimized to respond as quickly as possible to rapidly evolving situations.
Boyd said you don't want the fastest plane. You want the most responsive. That changed everything.
The Optimal Team Size Is 2 to 10
Collison: From a hiring standpoint pre-PMF, it should be about — what's going to get you there faster? At an early stage, it's most likely just people who help you build the product. But not too many — at some point you might be able to do more, but you're actually less responsive because you have a bigger team to manage.
As an empirical matter, somewhere between two and ten people — depending on what you're building — is the optimally responsive size.
Somewhere between 2 and 10 people is the optimally responsive size.
The Calculation
Collison: Each successive person takes time to hire, slows you down. Takes time to onboard, slows you down. Takes time to meld with the culture and learn the stack. Then there's the persistent ongoing cost of coordination and alignment — and that's not necessarily linear, it can be quadratic given the multi-way communication problem.
The question is — is the benefit of this additional person worth all these costs? The ultimate arbiter: will we be more responsive to what we're learning about our users with this additional person?
Will we be more responsive to what we're learning about our users? That's the ultimate arbiter.