A culture conducive to innovation is not only good for a company’s bottom line. It also is something that both leaders and employees value in their organizations. In seminars at companies across the globe, I have informally surveyed hundreds of managers about whether they want to work in an organization where innovative behaviors are the norm. I cannot think of single instance when someone has said “No, I don’t.” Who can blame them: Innovative cultures are generally depicted as pretty fun. When I asked the same managers to describe such cultures, they readily provided a list of characteristics identical to those extolled by management books: tolerance for failure, willingness to experiment, psychological safety, highly collaborative, and nonhierarchical. And research supports the idea that these behaviors translate into better innovative performance.

But despite the fact that innovative cultures are desirable and that most leaders claim to understand what they entail, they are hard to create and sustain. This is puzzling. How can practices apparently so universally loved—even fun—be so tricky to implement?

The reason, I believe, is that innovative cultures are misunderstood. The easy-to-like behaviors that get so much attention are only one side of the coin. They must be counterbalanced by some tougher and frankly less fun behaviors. A tolerance for failure requires an intolerance for incompetence. A willingness to experiment requires rigorous discipline. Psychological safety requires comfort with brutal candor. Collaboration must be balanced with a individual accountability. And flatness requires strong leadership. Innovative cultures are paradoxical. Unless the tensions created by this paradox are carefully managed, attempts to create an innovative culture will fail.

1. Tolerance for Failure but No Tolerance for Incompetence

Given that innovation involves the exploration of uncertain and unknown terrain, it is not surprising that a tolerance for failure is an important characteristic of innovative cultures. Some of the most highly touted innovators have had their share of failures. Remember Apple’s MobileMe, Google Glass, and the Amazon Fire Phone?

And yet for all their focus on tolerance for failure, innovative organizations are intolerant of incompetence. They set exceptionally high performance standards for their people. They recruit the best talent they can. Exploring risky ideas that ultimately fail is fine, but mediocre technical skills, sloppy thinking, bad work habits, and poor management are not. People who don’t meet expectations are either let go or moved into roles that better fit their abilities. Steve Jobs was notorious for firing anyone he deemed not up to the task. At Amazon, employees are ranked on a forced curve, and the bottom part of the distribution is culled. Google is known to have a very employee-friendly culture, but it’s also one of the hardest places on earth to get a job (each year the company gets more than 2 million applications for about 5,000 positions). It, too, has a rigorous performance management system that moves people into new roles if they are not excelling in their existing ones. At Pixar, movie directors who cannot get projects on track are replaced.

It sounds obvious that companies should set high quality standards for their employees, but unfortunately all too many organizations fall short in this regard. Consider a pharmaceutical company I recently worked with. I learned that one of its R&D groups had not discovered a new drug candidate in more than a decade. Despite the poor performance, senior leaders had made no real changes in the group’s management or personnel. In fact, under the company’s egalitarian compensation system, the scientists in the group had been receiving approximately the same salaries and bonuses as scientists in much more productive R&D units. One senior leader confided to me that short of ethics violations, the company rarely terminated anyone in R&D for subpar performance. When I asked why, he said, “Our culture is like a family. Firing people is not something we’re comfortable with.”

The truth is that a tolerance for failure requires having extremely competent people. Attempts to create novel technological or business models are fraught with uncertainty. You often don’t know what you don’t know, and you have to learn as you go. “Failures” under these circumstances provide valuable lessons about paths forward. But failure can also result from poorly thought-out designs, flawed analyses, lack of transparency, and bad management. Google can encourage risk taking and failure because it can be confident that most Google employees are very competent.

Creating a culture that simultaneously values learning through failure and outstanding performance is difficult in organizations with a history of neither. A good start is for senior leadership to articulate clearly the difference between productive and unproductive failures: Productive failures yield valuable information relative to their cost. A failure should be celebrated only if it results in learning. (The cliché “celebrating failure” misses the point—we should be celebrating learning, not failure.) A simple prototype that fails to perform as expected because of a previously unknown technical issue is a failure worth celebrating if that new knowledge can be applied to future designs. Launching a badly engineered product after spending $500 million developing it is just an expensive flop.

Building a culture of competence requires clearly articulating expected standards of performance. If such standards are not well understood, difficult personnel decisions can seem capricious or, worse, be misconstrued as punishment for a failure. Senior leaders and managers throughout the organization should communicate expectations clearly and regularly. Hiring standards may need to be raised, even if that temporarily slows the growth of the company.

Managers are especially uncomfortable about firing or moving people when their “incompetence” is no fault of their own. Shifting technologies or business models can render a person who’s very competent in one context incompetent in another. Consider how digitization has impacted the value of different skills in many industries. That sales representative whose deft interpersonal skills made him a superstar may no longer be as valuable to the organization as the introverted software engineer who develops the algorithms used to predict which customers are most likely to buy the company’s products. In some cases, people can be retrained to develop new competences. But that’s not always possible when really specialized skills (say, a PhD in applied math) are needed to do a job. Keeping people who have been rendered obsolete may be compassionate, but it’s dangerous for the organization.

Maintaining a healthy balance between tolerating productive failures and rooting out incompetence is not easy. A 2015 New York Times article about Amazon illustrates the difficulty. The piece, which was based on interviews with more than 100 current and former employees, labeled Amazon’s culture as “bruising” and recounted stories of employees crying at their desks amid enormous performance pressures. One reason striking a balance is so hard is that the causes of failure are not always clear. Did a product design turn out to be flawed because of an engineer’s bad judgment or because it encountered a problem that even the most talented engineer would have missed? And in the event of bad technical or business judgments, what are the appropriate consequences? Everyone makes mistakes, but at what point does forgiveness slide into permissiveness? And at what point does setting high performance standards devolve into being cruel or failing to treat employees—regardless of their performance—with respect and dignity?