I have worked full-time at three tech companies in the US: a startup with fewer than 20 people, a solar company with about 15,000 employees, and a FAANG-level company with roughly 50,000. I went through at least two review cycles at each of them. Drawing on my own experience and what I have seen happen to friends, here is how performance reviews generally work at US tech companies.

How often reviews happen

Review cycles run anywhere from quarterly to once a year. My first two jobs used an annual review at the end of the year; my current company reviews every six months. Quarterly reviews have one advantage: the period is short, so the evaluation stays focused. But honestly, that is too often. Some projects simply need more time before they show results, and quarterly cycles mean managers spend a large share of their time writing and calibrating reviews instead of doing their day jobs. Annual reviews give longer projects a fair window to be judged, but the low frequency puts more pressure on employees, because a weak cycle or a missed promotion means waiting another full year. Annual reviews are also unfair to people who leave in the second half of the year, since they walk away without the rating and bonus their work earned. A cycle every six months is the compromise, and it is the one most companies seem to land on.

The three parts: self, peers, manager

A review usually has three parts: a self-review, peer reviews, and your manager’s review. The self-review is your own write-up of what you delivered during the cycle. Because you are hardly an objective judge of yourself, coworkers you have worked with are invited to write about your contributions too. A peer can choose not to share their review with you (only your manager sees it), so you may never know what a coworker wrote. And even if you do not invite someone, they can submit feedback on their own. In my experience, peer reviews tend to be reasonably objective and fair. Your manager then reads your self-review and the peer feedback and writes the final assessment.

The self-review is an essay contest

Because the outcome directly affects your bonus and your chances of promotion, everyone writes their self-review very seriously. There is a running joke that every review season turns into an essay contest: doing great work matters, but at review time what matters is how well you present it. Self-reviews tend to be long, because nobody knows your contributions better than you. Some people apparently write close to 10,000 words, which no manager has time to read. If a manager has 6 to 10 reports, and each report comes with a self-review plus 4 to 6 peer reviews, the reading load is enormous. Companies now generally advise keeping your self-review under 1,500 to 2,000 words, though I still go slightly over every time.

The four dimensions

At the companies I have worked at, engineers are evaluated on four dimensions: Impact, Engineering, Direction, and People. Other roles swap the Engineering dimension for their own craft metrics.

Impact is measured against whatever your team produces. On a product team, it might be shipping a product, growing the user base, or building features that make people use the product more or save them time. On an infrastructure team, it might be how much efficiency your work created internally or how many resources it saved. Every team has different metrics, so Impact looks different everywhere.

Engineering is about the quality of your technical work: the quality of your code, whether your designs are sound and scalable, and whether you raise the engineering bar for the team as a whole.

Direction is about leadership within your projects and team: setting and driving a project’s direction, spotting pain points in the product, and coming up with new ideas worth trying. People covers things like interviewing and hiring, mentoring less experienced engineers, coordinating across teams, and sharing knowledge internally or externally.

Every company has its own variations, but the broad shape is similar. Much of Silicon Valley’s approach seems to trace back to Microsoft’s review system, adjusted by each company to its own needs, at least according to an article I read years ago and can no longer find. Corrections welcome.

Peer reviews follow the same dimensions, but they are usually much shorter, since a peer can only write about the parts of your work they actually collaborated on. Your manager then pulls the self-review and peer reviews together into the final version.

How companies try to keep it fair

The whole point of the process is to give everyone a rating, and that rating drives bonuses and promotions. Once a company reaches a certain size, rating people fairly across different departments becomes a genuinely hard problem. The usual solutions: (1) managers from different departments calibrate together, and a rating only stands when the managers in the room reach consensus; (2) an independent review committee of people with no direct stake in the outcome makes the call; (3) ratings are compressed into a narrow range, so most people land at “meets expectations” and only a small fraction get the very top or very bottom ratings.

Promotion means performing at the next level

Each level carries different expectations on these same dimensions. The common practice in Silicon Valley is that you have to consistently perform at the next level before you get promoted into it. That protects people from being promoted too early and then struggling to meet a bar they were not ready for.

No surprises

No system is perfect, and every cycle length has its tradeoffs. The way to make sure the result never blindsides you is to use your regular 1:1s with your manager to check where you stand, and to raise problems and ask for help early, while you can still address them.

That is how performance reviews work, based on what I have seen. How does your company do it? If you are not an engineer, what dimensions are you reviewed on? I would love to hear in the comments.

Originally published on my former blog on June 12, 2020. That blog is now offline; the text is preserved via the Wayback Machine. Lightly edited for this site.