It’s easier to link to this text than it is to repeat the intuition every time.

Those who learn fastest win

One of the core reasons why, as I write this in mid-2018, Silicon Civilization has the world in their teeth is because they figured out that it wasn’t just about learning. It was about learning quickly.

Look at it from their perspective. A startup is a hypothesis looking for validation. Those startups that are able to learn fastest have a greatest chance of pulling up before the runway runs out. Those that learned survived takeoff. Those that really thrived never stopped learning.

They win because they got really good at learning.

It isn’t purely about data, it’s about how it’s used to learn

If it was entirely about access to data, and only about data, then the world’s middle management data hoarders would be masters of the universe. They aren’t. There’s way too much data chasing way too few brains. There’s very little opportunity for many of our duopoly minders.

Data, for many, is simply there for the veneer of credibility it gives to gut decision making. Somebody somewhere has a genuinely stupid idea and they need the data to prove to decision makers that it is a good one. Or, worse, somebody somewhere has a genuinely fantastic idea and they can’t rally enough data to symbolically counter somebody who has rallied contrary data. As one middle manager at a large telco once said: “Oh, I see you brought a binder of data, next time I’ll bring three binders of data and we can play that game.”

When data about the competitive or internal environment is rallied to make decisions about things other than optimization tweaks, real second order learning can happen. This is really important because one can outlearn a strong competitor, but one can’t out-optimize their way out of a truly terrible situation.

It’s about learning. Does information from nature seep into the mental models of how managers make decisions? Do models update? If not, why not? If so, how often?

Rate of learning

Most valley startups are optimized for speed. This approach has been taken right into stack. A lot of the impulses around modern CI/CD are about deploying stable code really quickly. Why? Because the code, the very source of enablement, was often getting in the way of learning. Make it easy to ship and one makes it easy to learn. Friction is reduced and the rate of learning is increased.

They have cultures that are optimized for speed. They can take some pretty bold risks because they only expose the craziest things to a small portion of their user base. They release slightly embarrassing things all the time. They’re fail tolerant. Managers that continuously improve and who exude an indefatigable (and saccharine) sense of optimism are selected and rewarded.

Vector of learning

Their learning is often directed and aligned against at least one thesis. Google didn’t start with testing 41 shades of blue. It learned all the big things early and got really big into optimization once it had traction. You only get to 41 shades of blue once the big battles are fought. I think it’s great that they’ve earned the priveledge of optimizing something that specific. And that tale of woe dates to 2009.

The learning isn’t entirely undirected. (To retain a strong sense of empowerment, it’s important to be open to tons of tests.) Typically the learning is directed at testing a direction. It might have the outward appearance of scattershot. And some degree of randomness is important for discovery. But there’s a vector.

Who decides the vector, and if that direction of learning is producing dividends, is the stuff of management. Is management able to accept that a given vector may not have been a good one, and can they modify the angle of it over time? If so, they earn another life. If not, they die. I won’t list the number of companies who have just failed to adjust their vector.

A vector-line that doesn’t have an arrow is a dot. You can’t ever validate or reject a thesis if there’s no learning the direction of the arrow.

Those who learn fastest, win

In general, we should wherever possible, enable learning to happen faster.

That means making it easier to ship. Because learning.

That means democratic access to even slightly dirty data. Because learning.

That means empowering action. Because learning.

That means recognizing and rewarding learning. Because learning.

Because those that learn fastest, win.