How to run a performance review

You have four or five reviews to write, a form somebody else designed, and at least one rating you're not sure you agree with. Most managers open that form the week it's due and try to reconstruct nine months from memory, which is how a report who carried the team through a bad quarter in March ends up with a review that's mostly about April.

This article covers what the review is actually being used for, how to write one from evidence, and how to deliver it. It also covers what happens in calibration that your report never sees, and what to do when you have to give a rating you disagree with. Read the section on the argument against reviews before you write your next one, because the CTO of Netflix will tell you her company doesn't run them at all, and her reasoning changes how you should think about everything after it.

The sources are Elizabeth Stone, CTO of Netflix, a company that has no performance reviews at all; Kim Scott, who managed teams at Google and Apple and wrote Radical Candor; Deb Liu, CEO of Ancestry and previously a VP at Facebook, PayPal, and eBay; Shishir Mehrotra, CEO of Coda and formerly VP of Product at YouTube; Jules Walter, a former PM at Slack and YouTube; and the Mochary Method curriculum, the operating manual executive coach Matt Mochary uses with startup founders.

What the review is actually being used for

Companies pile several unrelated jobs onto the performance review, and most of the frustration you feel writing one comes from trying to do all of them in the same paragraph:

You can't remove those jobs, but you can stop letting them collide. If your company lets you, deliver the rating and the pay decision in one conversation and the development conversation in another, a week apart. People can't absorb coaching while they're doing math about their mortgage.

The argument that reviews shouldn't exist

This isn't a fringe position. Netflix, a company with thousands of employees, doesn't run performance reviews, and Elizabeth Stone said so on the record while she was CTO.

Notice what's still there. Netflix keeps an annual 360 feedback cycle and an annual compensation cycle. What it removed is the rating itself, and any link between the feedback you give and a score. Stone's point is that once you delete the process, performance has to live in the daily operating rhythm, because nothing else will bring it to the surface.

Before you take this to your leadership team, hear the caveat Stone gives later in the same conversation. Netflix ran this way partly because it had no leveling system for individual contributors at all until about 2022, so promotions barely existed as a thing to argue about.

So the honest version of the anti-review argument is narrower than the slogan. Reviews are load-bearing when people compete for a limited number of promotions and a fixed pay budget, and when the manager making the case wasn't present for most of the work being judged. If your company has no levels and no promotion queue, you can drop the review too. Most of us don't work somewhere like that.

Kim Scott makes the softer version of the same point, and it applies whether or not your company has reviews. Feedback you saved for a scheduled meeting is already months old by the time you give it.

That gives you the standard to hold yourself to for the rest of this article. If the review is the first time your report hears something, that's a mistake you made months earlier, and writing the review is how you notice.

Nothing in the review should be new information

The no-surprises rule gets repeated so often that it sounds like a platitude, so it's worth hearing why it exists from someone who applies it to the hardest conversation there is. At Netflix, the toughest version is telling someone they're not the right fit, and Stone's argument is that you only earn the right to have that conversation by having smaller ones for months beforehand.

The practical test: before you submit anything, read your draft and mark every criticism your report has never heard from you. If there are any, you have two obligations. Write them anyway, because hiding a real problem to avoid an awkward hour is worse. And say out loud in the meeting that you should have raised it earlier, name roughly when you first noticed, and don't let the rating carry weight it hasn't earned.

The way to stop the problem recurring is a monthly rating in your one-on-ones. The Mochary Method calls this absolute feedback: tell each report where they stand on a 1 to 5 scale, what you liked, and what would move them up a level. Mochary's line is that bad news is less anxiety-inducing than no news, and that people carry constant low-grade worry when they don't know where they stand.

Do that ten times a year and the annual review turns into a summary of conversations you've already had. If your reports want the other side of this, send them how to run a one-on-one with your manager.

How to write one your report can act on

The Mochary Method teaches a review structure modeled on how Bill Campbell, the coach who worked with Steve Jobs and the founders of Google, ran them. It's worth copying because it spends most of its questions on results and forward motion rather than personality.

Mochary adds a step most managers skip: ask the same management questions of that person's reports. If a manager says they hold weekly one-on-ones and their reports describe something different, you've found the gap before it becomes an attrition problem.

Put evidence under every claim

Every claim in the review needs something underneath it. The pattern that works is: what happened, when, what effect it had, and what it tells you about their level.

Three or four examples like that are worth more than a page of adjectives. If you can't produce a single specific example for a rating you're about to give, that rating is an impression, and your report will be able to tell.

Make the goals section usable

The last section of the review is the only part your report will reread in August. Give them two or three goals, each one written so both of you could tell in six months whether it happened. "Improve stakeholder communication" fails that test. "Run the monthly business review yourself for the next two quarters, with me in the audience for the first one" passes it.

How to deliver it

Most delivery problems come from the same source: the manager is trying to soften a number, and the report is trying to figure out what the number is. Get the number out of the way first and you can actually talk.

If the hard part is the part you keep softening, how to give feedback that lands goes deeper on the wording. Two rules from the Mochary Method's feedback curriculum apply here. Deliver it in person or on video, never by written message alone, so you can see when someone gets defensive and address it instead of letting it turn into resentment. And say why you value the person before you say the hard part, every time, even when it feels formulaic. It feels formulaic to you because you already know that you value them, and your report doesn't.

Calibration, and what your report can't see

At most companies your rating isn't final when you write it. Other managers compare their people against yours in a calibration meeting, and most of them have never watched your report work, so all they have is what you wrote.

This is where the case for formal reviews is strongest, and Shishir Mehrotra makes it from experience. He ran product at YouTube and now runs Coda, and at Google he was elected by the other product leaders to lead the effort to keep the product management job description consistent across the company, because otherwise every team's definition of a good PM would drift apart until the title meant nothing.

That's the honest defense of the whole apparatus. One manager rating one person is a rating of that manager's standards as much as the person's, and a committee of people with no stake in the outcome is the cheapest correction anyone has found. Mehrotra is not sentimental about how well it works, though. He printed Google's PM level guide, cut it into slices with the level numbers removed, handed the slices to eight product leaders and asked them to say which level each one described, and nobody could do it. So the committee is a real check, and the rubric it applies is often much vaguer than the people being rated imagine.

Deb Liu ran this exact process as a VP at Facebook, and she tells a story about a product manager on her team whose work she could see every day and nobody else could.

Her story is the part of the argument that matters most to you as the person writing the document, and it cuts against the Netflix model. When a group of peers decides ratings and promotions, quiet people get penalized for being quiet unless something written speaks for them. The document is how you argue for someone who won't argue for themselves. So write the review knowing that a manager three teams away will read it cold, and put the scope, the numbers, and the hardest problem your report handled near the top.

Liu also has the reframe worth passing on to reports who won't write a real self-review because it feels like bragging.

Calibration is where you do that on your report's behalf, in a room of managers who have never seen their work. Have this written down before you walk in:

There's one thing you owe your report that this process makes easy to skip. Jules Walter, who was a PM at Slack and YouTube, points out that the feedback most likely to cap someone's career is also the feedback managers are least willing to say to their face.

So after calibration, write down every objection raised about your report that you haven't already told them, and then go tell them. If someone said your report is hard to work with and you deliver a review about missed deadlines, you've protected yourself and left them stuck at their current level with no idea why.

Tell your report the calibration exists too. Most people find out that other managers shaped their rating only after a decision goes against them, and by then it looks like something was hidden from them. If they've just been turned down for a promotion, the questions to ask after a promotion denial is a useful thing to send.

When you disagree with the rating you have to give

It happens to every manager eventually. You wrote a 4, calibration made it a 3, and now you have to deliver a number you argued against.

Fight it while the number can still change. Bring the specific evidence to calibration and ask what would have to be true for the higher rating, then write that answer down, because it's the roadmap for next cycle. Once the number is final, there are two temptations worth avoiding.

If the rating carries a pay or promotion consequence your report thinks is unfair, they may come back asking about compensation. That's a separate conversation and it deserves a real answer rather than a deflection; asking for a raise after taking on more responsibility covers the version they'll be reading.

Running the continuous version instead

You probably can't delete your company's review process. You can make it almost irrelevant by the time it arrives, which is what the Netflix model actually amounts to in practice.

The failure mode of continuous feedback is that it quietly becomes no feedback. Scheduled reviews are bad at a lot of things, and they're excellent at forcing an uncomfortable conversation onto a calendar. If you drop the ceremony, you have to replace it with a discipline you'll actually keep.

Frequently asked questions

Do performance reviews actually work?

They work for the jobs that need a written record: calibration across managers who don't know each other's people, promotion cases, and pay decisions that have to survive scrutiny. They're poor at development, because feedback arrives months late attached to money. Netflix removed them entirely and replaced them with continuous candor, which worked partly because it also had no levels and no promotion queue. Most companies have both, so the review is doing real work even when it feels like theater.

How long should a performance review be?

Long enough to hold three or four specific examples and two or three goals. That's usually one page. Length is not the signal people think it is; a two-page review of adjectives helps nobody, and a report can tell when you padded it.

What if I inherited the person halfway through the cycle?

Say so in the first line of the review, then write only what you observed and clearly label what came from their previous manager or their peers. Don't pass secondhand impressions off as your own observations. Ask the previous manager for specifics rather than a rating.

Should the self-review change my rating?

It should change your rating when it contains work you didn't know about, which happens more often than managers expect. Confident writing on its own shouldn't move it. Read every self-review looking for facts you're missing, then set the tone aside.

What do I do about someone who's performing well but in a role that's shrinking?

Rate the performance honestly and have the scope conversation separately and early. Merge them and you hand someone a review that feels like a punishment for something they didn't do, and they'll spend the next quarter fixing the wrong problem.

The performance review on one page