Work Rules!: Insights from Inside Google That Will Transform How You Live and Lead

by Laszlo Bock

Adobe, Expedia, Juniper Networks (a computer hardware manufacturer), Kelly Services (a temporary worker agency), and Microsoft have all eliminated performance ratings.

The academic research suffers from inconsistent measurement, where “real time” can mean anything from “immediately” to “days later.” Most real-time feedback systems quickly turn into “attaboy” systems, as people only like telling each other nice things. And how often are your comments structured in a way that actually causes behavior to change? Saying “Great job in that meeting” is far more common than “I saw that you noticed when the customer pushed back from the table and seemed to lose interest, and you responded by asking if they had any concerns. You did a great job of re-engaging them. You should continue to pay close attention to body language in meetings.”

Google board member John Doerr introduced us to a practice he had seen Intel use with much success: OKRs, or Objectives and Key Results. The results must be specific, measurable, and verifiable; if you achieve all your results, you’ve attained your objective. For example, if the objective is to improve search quality by x percent, key results that contribute to that would be better search relevance (how useful the results are to the user), and latency (how quickly the results show up). It’s important to have both a quality and an efficiency measure, because otherwise engineers could just solve for one at the expense of the other. It’s not enough to give you a perfect result if it takes three minutes. We have to be both relevant and fast.

location. It’s important that there’s a way to find out what other people and teams are doing, and motivating to see how you fit into the broader picture of what Google is trying to achieve.

On the topic of goals, the academic research agrees with your intuition: Having goals improves performance.113 Spending hours cascading goals up and down the company, however, does not.

The science on rating systems is inconclusive.

said, how many rating categories you have is the least important issue here, even though it was one that Googlers were stunningly passionate about.

the soul of performance assessment is calibration.

So what is it? Google’s rating system was (and is) distinctive in that it isn’t just the direct manager making the decision. A manager assigns a draft rating to an employee—say, “exceeds expectations”—based on nailing OKRs but tempered by other activities, like the volume of interviews completed, or extenuating circumstances such as a shift in the economy that might have affected ad revenues.xliv Before this draft rating becomes final, groups of managers sit down together and review all of their employees’ draft ratings together in a process we call calibration. Calibration adds a step. But it is critical to ensure fairness. A manager’s assessments are compared to those of managers leading similar teams, and they review their employees collectively:

Calibration diminishes bias by forcing managers to justify their decisions to one another. It also increases perceptions of fairness among employees.

recency bias is when you overweight a recent experience because it’s fresh in your memory.

Ratings are tools, simplifying devices to help managers make decisions about pay and promotion.

Intrinsic motivation is the key to growth, but conventional performance management systems destroy that motivation. Almost everyone wants to improve. Traditional apprenticeship models are based on this notion. An inexperienced worker wants to learn, and will learn best when paired with a more expert partner who teaches them. Remember the first time you rode a bike, or learned to swim, or drove a car? The thrill of mastery, of accomplishment, is a powerful motivator.

But introduce extrinsic motivations, such as the promise of promotion or a raise, and the willingness and ability of the apprentice to learn starts to shut down.

Workplaces that permit employees more freedom tap into that natural intrinsic motivation, which in turn helps employees feel even more autonomous and capable.

Evaluation is necessary to distribute finite resources, like salary increases or bonus dollars. Development is just as necessary so people grow and improve.”121 If you want people to grow, don’t have those two conversations at the same time. Make development a constant back-and-forth between you and your team members, rather than a year-end surprise.

future. The standard deviation describes how likely a certain amount of variation (or deviation) is to happen. For example, the average woman in the United States is 5 feet 4 inches tall,128 with a standard deviation of just under three inches. This means that 68 percent of women are between 5 feet 1 inch and 5 feet 7 inches. That’s one standard deviation. Ninety-five percent are within two standard deviations of the average, between 4 feet 10 inches and 5 feet 10 inches. And 99.7 percent are no more than three standard deviations away from the average, between 4 feet 6 inches and 6 feet 2 inches. If you look around your office or neighborhood, you’ll see this feels about right. (For men, the average height is 5 feet 10 inches tall, with the same range of about plus or minus 3 inches at every standard deviation. You might notice that in Blakeslee’s photo the average man is about 5 feet 7 inches. Thanks to better nutrition, Americans have grown taller over the last hundred years.)

Statistically, these phenomena are better described by a “power

The name “power law” is used because if you wrote an equation describing the shape of the curve, you’d need to use an exponent to describe it, where one number is raised to the power of another number (e.g., in y = x-½, the exponent is −½ and x is “raised to the power of −½.” This equation gives you a graph roughly like what’s shown on the top right).

In fact, human performance in organizations follows a power law distribution for most jobs. Herman Aguinis and Ernest O’Boyle of Indiana University and the University of Iowa explain that “instead of a massive group of average performers dominating… through sheer numbers, a small group of elite performers [dominate] through massive performance.”130 Most organizations undervalue and underreward their best people, without even knowing they are doing it. In chapter 10 I’ll explain why and suggest a better way to manage and pay people.

More important is to learn from your best performers.xlvii Every company has the seeds of its future success in its best people, yet most fail to study them closely. This is a missed opportunity, because as Groysberg demonstrates, high performance is highly context dependent. Benchmarking and best practices tell you what worked elsewhere, but not what will work for you. In contrast, understanding precisely what makes your best people succeed in your unique environment is the natural extension of Groysberg’s findings. If success depends on specific, local conditions, then you are best served by studying the interplay of high performance and those local conditions.

Project Oxygen initially set out to prove that managers don’t matter and ended up demonstrating that good managers were crucial. Project Gifted Youngsters was targeted at explaining what people who sustain the highest performance for long periods of time do differently from everyone else. They explored the top 4 percent compared to the other 96 percent, then dug into the top 0.5 percent versus the other 99.5 percent. The Honeydew Enterprise (named for Bunsen Honeydew, the intrepid innovator from the Muppets) strove to understand the behaviors and practices that most foster and inhibit innovation among software engineers. Project Milgram explored the most effective ways to mine social networks for knowledge within Google. (This was named for the same Stanley Milgram who studied obedience.

Engineers generally think managers are at best a necessary evil, but mainly they get in the way, create bureaucracy, and screw things up. It was such a deeply held belief that in 2002 Larry and Sergey eliminated all manager roles in the company. We had over three hundred engineers at the time, and anyone who was a manager was relieved of management responsibilities. Instead, every engineer in the company reported to Wayne Rosing. It was a short-lived experiment. Wayne was besieged with requests for expense report approvals and for help in resolving interpersonal conflicts, and within six weeks the managers were reinstated.138 Managers clearly served some purpose, but by 2009 engineers’ innate leeriness about management had reasserted itself.

mediocre. This is called a double-blind interview methodology, because it prevents the interviewer from biasing the interviewee, and the interviewee doesn’t know which category they are in either. In other words, both the interviewer and interviewee are “blind” to the experimental condition.

The 8 Project Oxygen Attributes Be a good coach. Empower the team and do not micromanage. Express interest/concern for team members’ success and personal well-being. Be very productive/results-oriented. Be a good communicator—listen and share information. Help the team with career development. Have a clear vision/strategy for the team. Have important technical skills that help advise the team.

And unexpectedly, we found that technical expertise was actually the least important of the eight behaviors across great managers. Make no mistake, it is essential. An engineering manager who can’t code is not going to be able to lead a team at Google. But of the behaviors that differentiated the very best, technical input made the smallest difference to teams.

Reading this, I realized that management too is phenomenally complex. It’s a lot to ask of any leader to be a product visionary or a financial genius or a marketing wizard as well as an inspiring manager. But if we could reduce good management to a checklist, we wouldn’t need to invest millions of dollars in training, or try to convince people why one style of leadership is better than another. We wouldn’t have to change who they were. We could just change how they behave.

Managing your two tails I’ve gone into detail on Project Oxygen and the bottom 5 percent for three reasons. First, it’s an exceptional illustration of what can be learned and accomplished by focusing on the two tails of performance. Looking at average managers didn’t help, nor did benchmarking. Comparing the extremes allowed us to see meaningful differences in behavior and outcomes, which then formed a basis for unceasing improvements in how people experience Google. Second, it illustrates the notion of compassionate pragmatism. Letting those who are at the bottom of the performance distribution know it, without tying that directly to pay or career outcomes, alerted and motivated them in as positive a way as possible.

American companies spent $156,200,000,000 on learning programs in 2011,141 a staggering sum. A hundred and thirty-five countries have GDPs below that amount. Roughly half the money went to programs put on by the companies themselves, and the other half was paid to outside vendors. The average employee received thirty-one hours of training over the year, which works out to more than thirty minutes each week. Most of that money and time is wasted. Not because the training is necessarily bad, but because there’s no measure of what is actually learned and what behaviors change as a result. Think

Why then is so much invested in corporate learning, with so little return? Because most corporate learning is insufficiently targeted, delivered by the wrong people, and measured incorrectly.

I can, however, tell you exactly where to find the best teachers. They are sitting right next to you.

minimum. In theory, you want the best person, the one with the maximum expertise, to be delivering training. But in mathematics there’s a more refined concept: the local maximum. The local maximum is the highest value within a more constrained range of values. The largest number is infinity, but the largest number between one and ten is ten.

because individual performance scales linearly, while teaching scales geometrically.

You don’t even need to pull your best salesperson out of the field to make this work. If you break selling down into discrete skills, there may be different people who are best at cold-calling, negotiation, closing deals, or maintaining relationships. The best at each skill should be teaching it.

For the learner, having actual practitioners teaching is far more effective than listening to academics, professional trainers, or consultants.

It is generally far better to learn from people who are doing the work today, who can answer deeper questions and draw on current, real-life examples.

conditions. Organizations always seem to have more demand for people development than they can satisfy, and Google is no different.

For the past forty-plus years, HR professionals have measured how time is spent, asserting that 70 percent of learning should happen through on-the-job experiences, 20 percent through coaching and mentoring, and 10 percent through classroom instruction.152

Scott DeRue and Christopher Myers of the University of Michigan did a thorough review of the literature: “First and foremost, there is actually no empirical evidence supporting this assumption, yet scholars and practitioners frequently quote it as if it is fact.”156

In 1959, Donald Kirkpatrick, who was a professor at the University of Wisconsin and past president of the American Society for Training and Development, came up with a model that prescribed four levels of measurement in learning programs: reaction, learning, behavior, and results.

Frank Flynn, a professor at the Stanford Graduate School of Business, once told me the secret to high student evaluations: “Tell lots of jokes and lots of stories. Grad students love stories.”

Stories key into a human hunger for narrative, rooted in wisdom that’s passed from generation to generation through myths and folklore. They are an essential part of effective teaching. But how students feel about your class tells you nothing about whether they have learned anything.

Moreover, the students themselves are often unqualified to provide feedback on the quality of the course. During the class, they should be focused on learning, not on assessing whether the balance of presentation to team exercise to individual exercise is correct.

A thoughtfully designed experiment, and the patience to wait for and measure the results, will reveal reality to you.

as a management team we have probably spent more time thinking through compensation issues than any other people issue, save recruiting. Recruiting, you’ll recall, always comes first, because if you’re hiring people who are better than yourself, most other people issues tend to sort themselves out.

we fine-tuned the extrinsic rewards as well. It came down to four principles: Pay unfairly. Celebrate accomplishment, not compensation. Make it easy to spread the love. Reward thoughtful failure.

At the same time, companies in lower-margin industries have found that paying people well—even when they don’t need to—can be smart business. Costco and Wal-Mart’s Sam’s Club are both warehouse retailers. Wayne Cascio, of the University of Colorado Denver, compared the two in 2006.

Most companies manage pay like this to control costs and because they think the range of performance in a single job is somewhat narrow.

Fairness is when pay is commensurate with contribution.

O’Boyle and Aguinis did five studies encompassing a population of 633,263 researchers, entertainers, politicians, and athletes. The table below compares how many people in each group you would expect to be at the 99.7th percentile of performance using a normal statistical distribution, and how many there are in reality.