This article originally appeared on HBR.org.

Autonomy may be the single most important element for creating engagement in a company. How can anyone feel engaged, let alone inspired, if she feels that some supervisor is always looking over her shoulder? But autonomy is a double-edged sword. On the one hand, it spurs creativity and involvement. On the other, unchecked autonomy can lead to ambiguity and inefficiencies, even organizational chaos. To find the right balance, you have to wrestle with three challenges:

Balancing autonomy and accountability. An essential counterweight to autonomy is strict accountability for results, and for the actions and behaviors that deliver those results. A company has to establish a strategy and a purpose that provide context for employees’ actions. It has to put the strategy into practice with measurable objectives, consistent measurement of progress toward those goals, feedback systems to monitor activities along the way, and appropriate consequences for reaching or failing to reach the goals. At their best, companies realize that not everything is easily measurable, or should be measured, and that constant temperature taking and micromanagement are inefficient and demoralizing. They establish transparent boundary conditions and clear expectations. Employees and teams know they will be held accountable, and they know where the guardrails are. They understand the objectives, and they have a great deal of freedom in determining how to reach them within those guardrails. Clarity of purpose and what we call high-resolution strategies, which give people a clear view of where they’re headed, provide the compass that can guide the choices that teams and individuals make when working autonomously.

Balancing freedom to innovate versus following proven routines. The art and science here is determining how to get both outcomes—consistency and innovation—in the right proportion and in the appropriate parts of your organization. In many areas, freedom to innovate is the critical need. Think of new product development, or the parts of the company’s value chain and business model that are undergoing significant reinvention because of digital transformations. In these activities, speed of innovation is critical, and the rallying cry should be autonomy, small teams, and organizational agility. Other areas, however, may benefit from standardized approaches. These are areas where consistent outcomes are essential and where speed of execution comes from deploying common methods, best practices, and enforced routines. The focus here should be on repeatability and efficiency. Each requires speed in different areas, innovation versus execution, and achieves these results in different ways. The challenge in striking the right balance is to know which method should predominate and how to design appropriate ways of working for each area. The wrong approach leads to confusion over goals and to ineffectiveness.

Balancing alignment with control. This task is closely related to the other two. In traditional hierarchical organizations, managers direct the work of subordinates and thereby ensure alignment with broader organizational goals. Spans of control are limited to a reasonable number—typically eight people or fewer—so that managers can effectively oversee their subordinates’ efforts. This organizational model can work well in relatively stable business environments, where the pace of change is modest and where annual planning cycles suffice for managing strategic changes and course corrections. In dynamic business environments, where innovation cycles happen in days or weeks rather than months and years, and where much of the work is cross-functional in nature and is undertaken by small, agile teams, this type of organizational model can be slow to respond and innovate. Companies that take the approach of empowering autonomous teams must find ways to ensure that coordination and connectivity happen among those teams without relying on controlling managers. Again, it’s a matter of managerial art as well as science to achieve alignment without excessive control.

Our favorite example illustrating how to approach these three challenges is the Swedish company Spotify. Spotify is a 10-year-old music, video, and podcast streaming company with 30 million paying subscribers and about $3 billion in revenue. Its more than 2,000 employees are organized into agile teams, called squads, which are self-organizing, cross-functional, and colocated. Spotify has largely succeeded in maintaining an agile mindset and principles without sacrificing accountability. It enables innovation while keeping the benefits of repeatability, and it creates alignment without excessive control. Its lessons apply to many companies, not just digitally enabled service providers. Here’s how.

Spotify’s core organizational unit is an autonomous squad of no more than eight people. Each squad is accountable for a discrete aspect of the product, which it owns cradle to grave. Squads have the authority to decide what to build, how to build it, and with whom to work to make the product interoperable. They are organized into a light matrix called a tribe. Tribes comprise several squads linked together through a chapter, which is a horizontal grouping that helps to support specific competencies such as quality assistance, agile coaching, and web development. The chapter’s primary role is to facilitate learning and competency development throughout the squads.

Leadership within the squad is self-determined, while the chapter leader is a formal manager who focuses on coaching and mentoring. Spotify believes in the player-coach model: Chapter leaders are also squad members. Squad members can switch squads and retain the same formal leader within their chapter. Spotify introduced a third organizational element, known as a guild. Guilds are lightweight communities of interest whose primary purpose is to share knowledge in areas that cut across chapters and squads, such as leadership, continuous delivery, and web delivery.

This unusual combination of squads, tribes, chapters, and guilds is the organizational infrastructure that underlies Spotify’s operating model. At first reading, it might sound like just another way to define a conventional organizational matrix in Millennial- and digital-friendly terms. But a closer examination reveals just how different the model really is and why it seems to work so well.

The squad structure achieves autonomy without sacrificing accountability. Every squad owns its features throughout the product’s life cycle, and the squads have full visibility into their features’ successes and failures. There is no single appointed leader of a squad; any such leadership role is emergent and informal. Results are visible both through internal reviews and through customer feedback, and squads are expected to fully understand successes and failures. Squads go through postmortem analyses of failures to ensure learning, and some squad rooms have “fail walls.” Every few weeks, squads conduct retrospectives to evaluate what is going well and what needs to improve.

To ensure that the feedback process is effective for individuals as well as for the squads, Spotify redesigned its performance management system to separate salary discussion and performance evaluations from coaching and feedback. Before, peer feedback was incorporated into salary reviews; in Spotify’s words, that “incentivized people to gather as many favorable reviews as possible rather than getting feedback around their biggest areas of potential improvement.” Now, colleagues use an internal tool to invite anyone—including managers, peers, and direct reports—to provide feedback on results and on what an individual can do to improve. Employees may solicit feedback as often as they choose. Spotify employee Jonas Aman told us, “The result is a process that everyone needs to own and drive themselves—it is about development and personal growth.”

Spotify encourages innovation without losing the benefits of repeatability. Since squads are the primary centers of innovation, Spotify introduced its chapters as the matrix to connect competencies across squads. Chapters in some ways are like a function-led center of expertise in a traditional model, which links center-led functions with business units. In Spotify’s case, chapters have less formal authority, and they are organized around discrete competencies as opposed to broad functions. Guilds were added to facilitate experience sharing for horizontal topics of interest that are at a higher level than a specific competency. In the traditional model, central functions define and enforce standards and routinized processes from the top down. At Spotify, best-practice methods are discovered over time and determined by popular adoption from the bottom up. A practice or tool becomes a standard only when enough squads have adopted it to make it a de facto standard.

Culture plays a big role in keeping the innovation engine firing on all cylinders. Spotify has an experiment-friendly culture with an emphasis on test-and-learn approaches and contained experiments. If people don’t know the best way to do something, they are likely to try alternative approaches and run several A/B tests to determine which is preferable. In place of opinion, ego, and authority, Spotify works hard to substitute data, experimentation, and open dialogue about root causes. It lowers the cost of failure through a decoupled architecture, so that a failure has a “limited blast radius” and affects only part of the user experience.

Spotify fosters alignment without excessive control. The central organizational feature that shapes Spotify’s model is the concept of “loosely coupled, tightly aligned squads.” The key belief here is that “alignment enables autonomy—the greater the alignment, the more autonomy you can grant.” That’s why the company spends so much time aligning on objectives and goals before launching into work. The leadership model at Spotify reinforces this alignment. A leader’s job is to figure out the right problem and communicate it, so that squads can collaborate to find the best solution. Coordination comes through context and through a deep understanding of the company’s priorities, product strategies, and overall mission. The release process decouples each element for feature squads, infrastructure squads, and client application squads. The ability to release features and then toggle them on or off enables full releases even before all features are fully operational. Here, too, the culture acts as a support. The watchword at Spotify is “be autonomous, but don’t suboptimize—be a good citizen in the Spotify ecosystem.” A common analogy at the company is a jazz band: Each squad plays its instrument, but each also listens to the others and focuses on the overall piece to make great music.

Clearly, not all of Spotify’s choices will be appropriate for every company; that’s not the point. Rather, the point is that a company must make explicit choices in its operating model, ways of working, and culture that address the three core tensions between individual autonomy and organizational goals. Systematically aligning all elements of your operating model and working environment to create autonomy without sacrificing accountability, to get innovation where it matters most without sacrificing the benefits of scalability and repeatability, and to get alignment without excessive control are all at the heart of building an engaging and inspiring working environment.

Adapted from the Harvard Business Review Press book Time, Talent, Energy: Overcome Organizational Drag and Release Your Team’s Productive Power.

Michael Mankins is a partner in Bain & Company’s San Francisco office and a leader in the firm’s Organization practice. He is a coauthor of Time, Talent, Energy: Overcome Organizational Drag and Unleash Your Team’s Productive Power (Harvard Business Review Press, 2017). Eric Garton is a partner in Bain & Company’s Chicago office and leader of the firm’s Global Organization practice. He is coauthor of Time, Talent, Energy: Overcome Organizational Drag and Unleash Your Team’s Productive Power (HBR Press, March 2017).