4 Aug 2016

Summary of Astro Teller's Talk on Failure

Astro Teller, the head of Google X, gave a good talk on failure:


Here's a summary:

- I'm a cultural engineer.
- OKRs don't work. They are a stick for your boss to hit you with. He wants them as high as possible, and you want them as low as possible. You end up with a compromise neither side is satisfied with. You don't own the resulting OKRs. They're not yours.
- Set audacious goals, which you'll achieve around a tenth of the time. Lower, and people won't even try. Higher, and it's not audacious, by definition.

- There's a big gap between what companies say (lean, fail fast, pretotype) and what they do.
- If the boss says "fail fast", but it might get you fired, lose your reportees, be demoted, lose your bonus or not get promoted, would you still fail fast?
- People will follow the path of least resistance.
- People will recognise failure and end their projects only if you make it the path of least resistance to do so: encouragement, applause and respect from the boss and from colleagues, and a bonus and vacation.
- Failing should get you what you want.

- It's bad to fail for no reason and it's meaningless. I'm not pro-failure. I'm pro-learning.
- X wants projects to fail quicker and with less cost.
- Innovation means reducing the time or cost to fail.
- We don't want projects that take years or tens of millions of dollars before we learn we're wasting time. We're not a "take big risks blindly" organisation.

- X does premortems. Learn from failure before it happens. If we fail three years from now, what will the reason be?
- X does rapid eval followed by the foundry.
- Rapid eval is: How can we figure out as soon as possible why something is a bad idea?
- You need to create a safe space for people to express ideas. The only acceptable response to an out of the box idea is "Awesome". And that's for the out-of-the-box thinking. Immediately followed by, "How are we going to figure out that that's not going to work?"
- We were all creative at 6. It gets beaten out by society.
- Rapid eval evaluates the technical merits of an idea, while the foundry evaluates the market size, regulatory environment, and how much good would it good for the world.
- The goal is for foundry to kill more than half the projects, and these are the survivors from rapid eval.
- Since it's common for projects to die, people don't feel bad and are willing to kill their projects.
- The pressure is not: How can we get this to work. The pressure is: how can we figure out as soon as possible why this won't work so that we can move on to something better?

- Astro often asks teams to make a list of ten things the team needs to do. This invariably ends up being ordered by how important each item is. He then asks the team to re-order it by learning. The one at the top should cause us to learn the most. Just do the first two things on the new list. Then make another list of 10 things by importance. This list is completely different from the first list. The learning caused the change. You could skip doing the other 8 that were made irrelevant by the learning. Always do the learning items first.
- All this doesn't work if your time horizon is 3 months — if you have to show results in 3 months. In that case, it works best to find great people and use fear and greed on them. Conversely, if you have decades, you can't over-trust and over-empower people. Let them spend 6 months investigating something. It will eventually pay off.
- X's time horizon is: We're looking for things that that can have a huge meaningful impact on the world in 5-10 years.
- I love questions like, "Can we put a copper ring around the north pole to generate electricity?" because it makes me believe we haven't even scratched the surface of our collective creativity as a species.
- Better to reject a great idea than spend years and a lot of money and management attention on a bad idea. There's no shortage of problems, or great solutions to those problems.
- Ideas should make sense both on the technology and business fronts.
- X gave out a "get weirder" award to people who proposed great experiments. It's not about outcomes, which you can't control. In fact, the award is given out before the experiment is done. As a scientist, you care only about the quality of the experiment: how thoughtfully and cheaply did you create the experiment?
- X is the worst moonshot factory in the whole world, except for all the other ones.

No comments:

Post a Comment