What does a Skunk Works do?

When an established firm aspires to experiment in a radical direction, management gurus recommend opening a skunk works—an organizational home for high-priority original thinking and projects. It is housed away from the organization’s main operations, sometimes in secret or with organizational barriers. Typically the projects involve something of value to the future but are not directly connected to the present operational or service missions. Sometimes a skunk works has the approval of senior management, and sometimes it does not.

The phrase “skunk works” originated from the aeronautics industry, and in that context it had a specific meaning (and still does). The meaning of the phrase has evolved, and today it means something broader outside of aeronautics; that causes confusion, which further fosters poor managerial decisions. Today’s column maps the scope of change.

Appreciate how far the phrase has come. The very first skunk works began many decades ago with a project for the air force at a division of Lockheed in Burbank, California. It initially had called itself the “Skonk Works” in a bit of salty humor about its own secrecy. The phrase came from Al Capp’s Lil’ Abner cartoon—the skonk works was a “secret laboratory” that operated a backwoods still. Eventually the label became well known. Too well known, as it turned out.

Even in those days, publishers tried to protect their copyrights, and Lil’ Abner’s publisher eventually asked Lockheed to stop calling itself that. Long story short, smell played a role thereafter, because the Lockheed facility was next to another plant that emitted fumes—like a skunk.

In the modern usage, a skunk works still seeks to depart from existing operational processes, but pursue a variety of goals. Despite that variety, all skunk works face similar issues. Without daily operations to ground routine, there are few measurable benchmarks that refer to existing operational metrics. That makes them challenging to manage and monitor, which makes it difficult to align costs with benefits. That is a fundamental economic challenge.

Science model

One type of skunk works focuses on science, or, more commonly, its application to frontier engineering. Most computer scientists and engineers observe this model in university research or government laboratories. There the skunk works supports the discovery of scientific breakthroughs or facilitates elements of the scientific process, such as refining scientific implications, developing prototypes, and implementing initial frontier proposals at scale.

Measurement of progress takes familiar scientific forms for advanced engineering, such as papers published or patents applied for and granted. In short, there is a way to measure major incremental discoveries, and at intermediate stages of progress. This works well when these standard measures actually correspond with meaningful progress, which sometimes happens.

These types of skunk works tend to become institutionalized inside labs or, from time to time, small satellite organizations housed away from headquarters. For example, many large firms have started a skunk works in Santa Clara, sometimes to connect with researchers in the area. That follows a long history going back to Xerox PARC, which stood for Palo Alto Research Center. PARC hired some of the best computer scientists in the area at the time, and they invented some of the most canonical prototypes of the 1970s, including the mouse, LAN, and GUIs. (And, infamously, Xerox HQ had difficulty commercializing these inventions.)

There are many variants on this model. For example, some skunk works devote themselves to producing a “white paper” with employees who get temporarily relieved of regular duties. The white paper investigates a previously unexamined topic, and sometimes develops a prototype or a strategy for scaling the frontier. Although these projects might sound like “skunk-works-lite,” the organizational form serves its purpose, not letting organizational politics get in the way of the analysis.

The most famous of this “lite” model was the small group inside Microsoft who investigated the Internet in the summer of 1994, before Bill Gates took much interest. The people in the works put their jobs at risk, and, luckily for them, eventually got the boss’s attention.

Research community model

Another model of a skunk works focuses on building research communities. Leaders focus on developing a research community with a broad mission. Individual projects, or their success and failure, matter less than the development of a group with an identity built around shared interest. Leaders develop communication channels for the community and initiate institutional support to make the community self-sustaining.

DARPA has used this model for many decades, and it can also be found at many other government agencies with a mission to develop external research. Although DARPA’s development of the Internet is the best-known example, there are also variants outside of government, especially in engineering fields where evolving standards require frequent updates and revision. That is why, for example, Vic Hayes, who led the first committees to develop what became Wi-Fi, focused on developing his community into something self-sustaining. That helped the standard persist and grow.

Related indicators of progress include size of membership and other milestones related to its sustainability. Indirect benchmarks include the members’ prestige, symptoms of their devotion to their activity, and the spread of mission to groups outside of the core initiators.

Because some community members contribute without expecting direct compensation, assigning responsibility and credit can become an issue. For example, think of the Mosaic browser team at the University of Illinois in the mid-1990s. This small team of programmers built on top of the rapidly diffusing web and many years of accumulated processes for the entire infrastructure, and the team generated visible gains and received lots of publicity. Who should have gotten credit for this advance—the software team who built Mosaic, the research institute that paid their salary, or all the others who made the Internet reliable in the prior decades? Two decades later, many of those involved still argue about these matters.

Lead user model

In a third model of a skunk works, the works acts as a lead user. Governments use this model to meet a unique government mission, such as a military need or NASA’s early efforts. The lead user pays for pushing the scientific and engineering frontier.

Lead users tend to measure progress with new metrics developed during experience. Certainly this was a feature of the situation during the NSF sponsorship of the Internet, when it was acting as a lead user. The management had visible symptoms of growth to show others, but it was challenging for them to explain the value of what they were developing.

More generally, lead user efforts are benchmarked against aspirations, and any benchmark compromises between the practical and the dream. Nothing challenges dreamers more than learning from (bitter) experience that their goals need to pivot. Leaders have to be visionaries and good managers. Those people are rare.

Many startups act as glorified versions of this type of skunk works outside of government, often with the explicit goal of being acquired later. Why did these lead-user projects get housed in a startup and not as a skunk works in an established firm? Usually there are dozens of reasons.

Kickstarter is full of firms like this. A recent known example is Oculus Rift, which pushed the envelope just enough and hit it big with an acquisition. Occasionally, a major venture capitalist will take on one of these projects. An example is SpaceX, which decided to develop frontier engineering and processes for reusable rockets parts.

Moon shot model

There is a fourth model of a skunk works. The organization defines an audacious target—it shoots for the moon. The audacious target acts as a unifying goal for a team and every participant. You will notice that this model contains elements found in all the others. Leaders develop communities, the participants tend to develop frontier engineering using science, and the organization also acts as a lead user.

Private firms occasionally like to shoot for the moon. Google used this type of organization to start its autonomous car project, and IBM used one to beat Jeopardy! with Watson.

The best part of this model arises at the end. It is easy to measure when it is done—a goal is either achieved or not. It can be extremely challenging, however, to indicate progress until that moment. For example, the team that developed Watson started as a scientific skunk works, as a part of IBM Labs.

Because progress had slowed, management decided to convert the skunk works to the moon shot model. Hindsight makes this conversion look smart, but it was not easy to put all of the team’s eggs in one basket. It took the team four years to meet its goals. In the meantime, they published nothing, and they almost failed their trial with the Jeopardy! producers. Kudos to the team leaders for keeping this together.


Even when skunk works successfully meet their targets, they face issues transferring their success into regular operations. Firms and government both tend to put off the hard questions affiliated with such integration. That is a difficult topic, and one best left for another day.

Suffice it to say, there is no magic to integration. Because the projects deliberately varied from existing processes, making it work with everything is usually an effort as large as the skunk works itself.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s