First time reading your work, was linked her from The Diff, by Bryne Hobert.
As I read this I thought of the old sport axiom of basketball... You never want to be mediocre. You want to bottom out for draft picks, or go for the title. Going for the middle is death.
Teams in the middle, are expensive, and won't win. You never learn anything from them.
To expand the analogy, better to lose OFTEN with young, high upside players (learn, iterate)
than middle along with known players with no room to improve - (mediocre success)
I love this analogy. Bottom out, get lottery picks, wipe the slate clean. Or go all-in for a title. But getting stuck with 40 wins and a first-round exit year after year is purgatory -- what do you do next? You have no good options. Mediocre success at its best (worst)!
Yep, exactly right: most of my time is spent on investing and advice/mentorship (and writing, which is a big component of both of those). But I'm also carving out some bandwidth to explore LLMs and all things data+AI. Not enough hours in the day!
I'm struck by how much this sounds like test driven development. The idea that you say what you want to happen; tests fail, cos it ain't happenin': write code to make it happen: tests pass.
The fact that you you are articulating what success looks like before start.
Very interesting article. Definitely counterintuitive. But playing devils advocate, at least in consumer space it seems like repeatable iterative growth is the name of the game. The proverbial 1.01^182 = 6xx. In this sense isn’t a series an incremental wins (mediocre success) a good thing as long as you have the ability to repeat and rinse and even hit a 50% success rate?
Great article. However I have an alternative view.
Startups are like control systems with feedback impacted by the environment its operating in.
Rather than treating the outcome as a qualitative metric (good|bad, yes|no) , it needs to be transduce'd and fed-back to the input. At best it can turn out to be a virtuous loop (snowball) or a vicious loop (diminish) with baseline thresholds to break the circuit.
The word agile leads to "hurried" and "many more" experiments which in itself can be expensive in terms of cost and time.
First time reading your work, was linked her from The Diff, by Bryne Hobert.
As I read this I thought of the old sport axiom of basketball... You never want to be mediocre. You want to bottom out for draft picks, or go for the title. Going for the middle is death.
Teams in the middle, are expensive, and won't win. You never learn anything from them.
To expand the analogy, better to lose OFTEN with young, high upside players (learn, iterate)
than middle along with known players with no room to improve - (mediocre success)
I love this analogy. Bottom out, get lottery picks, wipe the slate clean. Or go all-in for a title. But getting stuck with 40 wins and a first-round exit year after year is purgatory -- what do you do next? You have no good options. Mediocre success at its best (worst)!
Thanks for the response. I knew watching sports my whole life was worthy of something!
A lot of what you say applies to science too, nice stuff Thomas :-)
Couldn't agree more! That is an excellent way of explaining it. It is way more painful to be in the mediocre realm than failing.
I enjoyed this post as well as your deeper dives on data topics. Thanks for sharing Abraham!
You might have mentioned elsewhere — are you spending most of your time investing and advising these days?
Yep, exactly right: most of my time is spent on investing and advice/mentorship (and writing, which is a big component of both of those). But I'm also carving out some bandwidth to explore LLMs and all things data+AI. Not enough hours in the day!
I agree that the worst outcome is a mediocre success, but concretely what are some ways to make the outcome of experiments more extreme?
Good question. Here are some tactics that I've used, and found effective:
- write down your hypothesis clearly and explicitly beforehand; if you can't articulate it, how can you test it?
- define your success/failure criteria before you take any actions (write a single line of code, talk to a single customer)
- share both of these written documents with multiple (internal) people, to create accountability
- for products, launch earlier than you think you should; avoid polishing too much
- set hard deadlines for your experimental window, and stick to them
- don't accommodate customer requests; listen to feedback, but don't change anything during the course of the experiment
- have a designated skeptic / devil's advocate whose job it is to argue against your positive signal
- think of all your excuses in advance ("we didn't get clear results because of X") -- and then find ways to mitigate them
I'm struck by how much this sounds like test driven development. The idea that you say what you want to happen; tests fail, cos it ain't happenin': write code to make it happen: tests pass.
The fact that you you are articulating what success looks like before start.
Yep, great comparison!
Same for project stati. Red and green work, you know what to do. But with yellow and orange? What does this mean?
Very interesting article. Definitely counterintuitive. But playing devils advocate, at least in consumer space it seems like repeatable iterative growth is the name of the game. The proverbial 1.01^182 = 6xx. In this sense isn’t a series an incremental wins (mediocre success) a good thing as long as you have the ability to repeat and rinse and even hit a 50% success rate?
Abraham
Great article. However I have an alternative view.
Startups are like control systems with feedback impacted by the environment its operating in.
Rather than treating the outcome as a qualitative metric (good|bad, yes|no) , it needs to be transduce'd and fed-back to the input. At best it can turn out to be a virtuous loop (snowball) or a vicious loop (diminish) with baseline thresholds to break the circuit.
The word agile leads to "hurried" and "many more" experiments which in itself can be expensive in terms of cost and time.
Agreed, startups are all about feedback loops!