Successful Failure: Experiment, Invent, Innovate
Run the sequence everyone skips: you cannot innovate without first failing at experiments
- Difficulty
- Advanced
- Time to result
- ~months to results
- Steps
- 6
- Confidence
- 88%
Anderson takes the phrase from Apollo 13, where Jim Lovell called the aborted mission NASA's most successful failure, and maps it onto a theme running through every Bezos letter: failure is necessary for success. The mechanism is a sequence most companies invert. You must experiment in order to invent, and only then can you innovate — but everyone talks about innovating while skipping the work underneath it. Since an experiment whose outcome you already know is not an experiment, experimentation must produce failure by definition. The blocker is not psychological but structural: Anderson argues employees understand failure is part of learning and are afraid of its consequences, not of failure itself. So the fix is removing the consequence. And the bets must scale — Bezos argued that if the size of your risks does not grow with the size of your organization, billion-dollar failures included, you will not move the needle.
Origin
Extracted from Young and Profiting
Core principles
- 01If you know the experiment will work, it is not an experiment
- 02Experimentation must produce failure or it is not experimentation
- 03The order is experiment, then invent, then innovate — never the reverse
- 04Employees are not afraid of failure; they are afraid of its consequences
- 05Risk size must scale with organization size to move the needle
How to run it
- 1
Test whether it is really an experiment
If you already know it will work, you are executing, not experimenting. Only bets with a real possibility of failure qualify.
Pro tip Ask the team to state the odds out loud before starting; certainty is a red flag.
- 2
Run the sequence in order
Experiment first, which produces the raw learning. Invention comes out of that learning. Innovation is what you do with the invention. Reversing the order produces slogans, not products.
Pro tip When someone demands innovation, ask what experiments are funded to produce it.
Watch out Most companies start at step three and wonder why nothing arrives.
- 3
Remove the consequence of failure
Change what actually happens to people whose experiments fail — assignments, promotions, standing. The fear is of the consequence, not the outcome.
Pro tip Bezos claimed Amazon is the best place in the world for an employee to fail, because failure is understood as part of the process.
Watch out Declaring failure safe while quietly sidelining the people who failed teaches the real lesson.
- 4
Take the write-off in the open
When a bet fails, absorb the loss visibly rather than burying it. Amazon wrote off $178 million in inventory and development loss on the Fire Phone at the end of 2014.
Pro tip Publishing the number makes the next big bet easier to authorize, not harder.
- 5
Redeploy the team, not just the learning
Keep the people who ran the failed experiment together and pointed at the adjacent problem. The knowledge lives in them.
Pro tip The Fire Phone hardware team from Lab126 demoed the Echo to Bezos four months after the phone's announcement.
Watch out Dissolving a failed team destroys the asset the failure just bought you.
- 6
Scale the size of the bets
Match risk size to organization size. A big company placing small bets is running experiments that cannot change its trajectory even when they win.
Pro tip Bezos: if we're not taking billion-dollar risks and making billion-dollar failures we're not going to move the needle enough.
Watch out Bet sizing must respect what the balance sheet can absorb — Amazon's scale is not a template for a small firm's.
In the wild
The Fire Phone was Bezos's pet project, announced in 2014 as a phone designed for shopping on Amazon — into a market that already had Android and iOS and needed no more phones. Amazon tried selling it for 99 cents and could not give it away, writing off $178 million at the end of 2014. But four months after the phone's announcement, the same Lab126 hardware team gave Bezos his first demo of what became the Echo: hardware that could recognize a voice from across the room, married to the Alexa machine learning platform to understand a question, look up the answer, and speak it back.
→ Amazon's biggest hardware failure directly produced the team and technology behind a product now in nearly every household.
Amazon went through two dead ends before landing on Marketplace. First it opened an auction site to compete with eBay, but people did not come to Amazon for auctions — eBay simply had it. Then it tried zShops, a separate website with a separate login where third-party sellers could reach Amazon's customers; nobody wanted to log in somewhere else. The third iteration put third-party sellers on Amazon's own product pages, which looked like a crazy idea internally since those sellers competed directly with Amazon's inventory.
→ Marketplace became one of Bezos's named big bets, with third-party sellers paying fees for access to the site and the fulfillment network Amazon spent billions building.
Common mistakes
Demanding innovation, punishing experiments
Most companies punish employees for failure while requiring them to innovate. The result is that people only propose bets they already know will work, which by definition are not experiments.
Fixing the language instead of the consequence
Calling failures 'learnings' changes nothing if the person who failed loses standing. Employees read the consequences, not the vocabulary.
Betting small at large scale
A big organization running only small experiments cannot move its own needle even when they succeed, so the innovation program stays permanently theatrical.
Is it for you?
Best for
Leaders who want genuine invention and are willing to change what happens to people whose bets do not work.
Not ideal for
Regulated or safety-critical contexts where the cost of a failed live experiment is measured in harm rather than money.
From the transcript
“experimentation absolutely has to lead to failure because if you know the experiment's going to work it's not an experiment”
“you have to start with an experiment in order to invent and only then can you innovate but everybody's talking about needing to innovate but…”
“i'm convinced employees aren't actually afraid of failure we all understand that that's part of learning but they're afraid of the consequences of that failure”
From the episode
Steven Anderson: Grow Your Business Like Amazon
Steven Anderson