Innovation Runs on Failing Well

Picture a pharmacy director who pilots a new discharge workflow on one unit. Two weeks in, it is clearly not working. Technicians are skipping a step because the timing collides with their delivery runs, and the handoff to nursing is slower than before. She pulls the plug, writes up what happened, and sends it to her leadership team. What happens next tells you almost everything about whether that organization will ever innovate. In one version, someone asks, "What did we learn, and what is the next smallest test we can try?" In the other version, someone asks, "Who on earth approved this?"

Everyone on that unit hears the answer. They hear it more clearly than anything written in the organization's strategic plan or listed on its values posters. After hearing the second response, people will bring proposals that are safer, slower, more heavily studied, and a good deal less innovative. When this happens, an organization begins to stagnate, not because anyone decided to stop innovating, but because the organization's reactions simply taught its people that the smart move is to avoid being associated with anything that might not work.

The Arithmetic of Innovation

There is simple math underneath this. Anything genuinely new involves uncertainty, and uncertainty means some of your attempts will not work. You cannot collect the successes without the failures, because you cannot know in advance which attempts belong in which pile. If failure is punished, you get fewer attempts. Fewer attempts means fewer discoveries. An organization's capacity to innovate is therefore proportional to its people's willingness to be wrong.

That can sound like permission to be reckless, and this objection to "embracing failure" deserves a real answer. In healthcare especially, some failures hurt people. A wrong dose is not a learning opportunity. Amy Edmondson, the Harvard researcher who has spent her career on this question, draws exactly that line in her book Right Kind of Wrong. She sorts failures into three kinds. Basic failures occur in known territory, where a proven method exists but was not followed due to simple human error. Complex failures are multi-cause breakdowns in which several small breakdowns line up badly. Both of those call for vigilance, checklists, and human-centered redesign. The third kind of failure is the intelligent failure: an undesired result in new territory, in pursuit of a meaningful goal, from a test that is as small as it can be while still teaching you something, without spending more time, money, or risk than the knowledge is worth.

That last clause is the whole idea. Failing well means failing fast and cheap. The goal is not to simply fail more. The goal is to get the answer to "will this work?" at the lowest price available, before the price goes up. An organization that does this well does not have fewer failures than one that does it badly. It has smaller failures that serve as the basis of rapid learning and innovation.

Confidence in the Process, Not the Decision

Here is the uncomfortable part. You can never be completely confident in a decision made under uncertainty. If you could, it would not be uncertain. Good decisions sometimes produce bad results, and bad decisions sometimes get lucky, because outcomes depend on things no one controlled or could see in advance. That leaves a leader with a choice about what to judge: the result, or the way the decision was made. One of those choices builds an innovative organization. The other quietly destroys it.

The approach that destroys innovation is what Annie Duke, the former professional poker player, calls "resulting": judging a decision by how it turned out. Under a resulting mindset, if the pilot project worked, the decision was good. If it flopped, the decision was bad, and someone should answer for it. This feels like accountability, and it is easy, because the outcome is the only thing anyone can see. But unfortunately, the resulting approach fails for two reasons. First, it gets the facts wrong. It labels a sound decision that hit bad luck as a mistake, and it rewards a careless decision that happened to land well. Consequently, people learn nothing about what made a decision good or bad. Second, it pushes behavior in the wrong direction. If you are judged on outcomes you cannot fully control, the rational response is to stop making decisions where the outcome is uncertain. People defer, escalate, form committees, and wait for data that will never settle the question. Resulting does not produce caution about bad bets. It produces avoidance of all bets, including the good ones.

The approach that builds an innovative organization is to judge the process. Before anyone looks at the result, look at how the decision was made. What did we know at the time? What did we assume, and did we write those assumptions down? How big was the bet relative to what we could afford to lose? Did we find out early or late? This works for two reasons that mirror the first two. First, it measures something real. The quality of the inputs, the size of the risk, and the speed of the feedback are things a person actually controls, so judging them is fair and instructive. Second, it points behavior in the right direction. When people know they will be evaluated on how they decided, they have a reason to make their thinking visible, state their assumptions, and keep the bet small, which is exactly what failing well requires. That is what it means to be confident in the process. You cannot guarantee the result, but you can know the decision was made well.

Go back to the pharmacy director's flopped pilot. Under resulting, the review asks who approved it and treats the two wasted weeks as a mistake. Under process review, the questions find that she named her assumptions, tested on one unit, and caught the problem in two weeks instead of six months. The pilot failed, but the decision was sound, and she has earned more trust for it. If the questions find instead that she skipped the assumptions and rolled out to three units at once, you have found something fixable. It is the process you repair, not the person you replace.

What Makes a Good Process

Two ideas together give you most of what a good process needs. The first is discovery-driven planning. The second is matching your level of preparation to your level of risk.

Rita McGrath and Ian MacMillan introduced discovery-driven planning in a 1995 Harvard Business Review article, and its central insight is that conventional planning quietly assumes you already know things you do not. A conventional plan starts with a forecast and then measures you against it. Under real uncertainty that is backwards, because the forecast is mostly assumption dressed up as a spreadsheet. Discovery-driven planning starts from the other end. You define what success has to look like, numbers included, and work backward to ask what would have to be true for that result to happen. Then you write those conditions down as assumptions, separate from what you actually know, and rank them by how much damage it would do if they were wrong. Finally, you fund the effort in stages, with checkpoints designed to test the riskiest assumptions first. Each checkpoint is a real decision: continue, change course, or stop. Stopping at a checkpoint is not the plan failing. It is the plan working, because it bought you the answer for a fraction of what the full commitment would have cost.

Notice what this does for failure. It builds failing well into the structure. The cheap experiments come first, the expensive commitments come last, and nobody is surprised when an assumption turns out to be wrong, because the plan was written expecting that some would be.

The second idea is about proportion. James Clear has written about a simple way to sort decisions by how reversible they are: hats, haircuts, and tattoos. Hats are cheap and easy to undo. If you pick the wrong one, you take it off, so try a bunch of them and move quickly. Haircuts can be fixed, but not fast, and you may feel foolish for a while. They are worth the risk, with some thought. Tattoos are close to permanent, so slow down and think carefully. Jeff Bezos made a similar point in his 2015 shareholder letter with the language of one-way and two-way doors. The logic is the same. The cost of reversing a decision should determine how much deliberation it gets.

Most organizations get this wrong in both directions at once. They treat hats like tattoos, running a simple change to a meeting format, or a pilot on a single unit, through three layers of review. That approach kills any chance for rapid, low-cost learning. Meanwhile they treat real tattoos like hats, signing a multi-year automation contract after a persuasive demo because everyone is tired of deliberating. The fix is a habit of asking one question early: what would it cost, and how long would it take, to undo this? A good leader asks it out loud so everyone learns to ask it. The follow-up is often the more valuable question: can we turn this haircut into a series of hats? A pilot on one unit, a thirty-day trial, a limited rollout with a defined exit. Most apparent tattoos can be broken into smaller, reversible steps, which is exactly what discovery-driven checkpoints do.

Psychological Safety Is What Makes It Work

All of this can be designed perfectly and still fail, for a reason that has nothing to do with the planning. A good process only works if people tell you the truth about what is happening. That requires psychological safety.

Edmondson defines it as a climate in which people feel safe to take interpersonal risks: to voice an uncertain idea, disagree, admit a mistake, or say that something failed, without fear of being rejected or punished. It helps to be precise about what it is not. It is not niceness, and it is not low standards. A team can be pleasant and quiet and have very little of it. Her work points to high safety paired with high standards as the combination where learning actually happens. Her early research on hospital nursing units produced the reverse of what she expected: the better-led units reported more medication errors, not fewer. The likeliest explanation was that people on those units were willing to say so. The units with low error counts were probably not safer, only quieter. Google's Project Aristotle later landed in similar territory, finding psychological safety the most important of the five dynamics that set its most effective teams apart.

For a leader, three things matter most. The first is how you frame the work. If you present an initiative as an execution problem, with a known answer and flawless delivery expected, every deviation looks like a failure. If you present it as a learning problem, with real uncertainty and a request for help finding out what is true, candor becomes part of the job. The second is modeling fallibility. People watch what the most senior person does with their own mistakes. A leader who says "I got that wrong, and here is what I am changing" hands everyone else a template. A leader who never does hands them silence. The third thing that matters most is what the leader does during the first thirty seconds after bad news. Your reaction to the first person who tells you something is going wrong decides whether you ever hear from the second. Thank them first. Analyze later.

There is a legitimate worry underneath all of this. Doesn't a culture without blame become a culture without accountability? It would, if psychological safety meant never holding anyone accountable for anything. Just culture, the framework developed by David Marx and Sidney Dekker out of aviation and healthcare safety work, is the answer to that worry. It separates behavior into three categories and responds to each differently. Human error, an honest slip, is consoled while the system is examined. At-risk behavior, where someone has drifted into a shortcut and stopped seeing the danger, is coached. Reckless behavior, a conscious disregard of a known and substantial risk, is met with formal discipline. The distinction matters because it tells people exactly where the line is. Most failures in innovation work land in the first two categories, and treating them like the third is how organizations teach people to hide things and never take calculated risks. With a just culture, accountability survives intact. It simply moves to where it belongs: the quality of the process, the speed of speaking up, and what you did with what you learned, instead of outcomes no one could have guaranteed.

Making It Real

None of this lives on a values poster. It lives in a handful of repeatable practices. Tag decisions as hats, haircuts, or tattoos so the level of scrutiny is visible and agreed on in advance. Write down assumptions and kill criteria before launching anything, so that stopping is a pre-agreed decision instead of a political one. Review failures by asking what was learned and what the next smallest test is, before anyone asks anything else. And as a leader, go first. Share one of your own failed bets, with the reasoning behind it, at your next team meeting. It will do more than any memo.

Organizations rarely announce that they have stopped innovating. They just get a little better each quarter at avoiding being wrong, one cautious decision at a time, until that habit becomes the culture. The organizations that keep building something new are not braver, and they are not luckier. They've figured out how to make failure small enough to survive, visible enough to learn from, and safe enough to report. That is a design choice, and it belongs to the people in leadership.