Performance Reviews

The performance review is the most visible, most dreaded, and most misunderstood part of performance management. Done well, it is a consolidation — a periodic gathering-up of signal that already exists, into a fair written record and a forward-looking development conversation. Done badly, it is a once-a-year ambush where people learn for the first time how they're seen, fight over a number, and leave demoralized.

The single test of a healthy review: it contains no surprises. If your 1:1s and continuous feedback are working, the review summarizes a story the person already knows. If the review is the first time they hear something important, the weekly system failed, not the annual one.

What a Review Cycle Is For

• Fairness across the team. A structured cycle with calibration is the main defense against the bias that 1:1s alone can't catch — one manager grading easy, another hard.
• A durable record. Promotions, compensation, and (if it comes to it) managing someone out all need a written history. Verbal-only management leaves nothing to stand on.
• A forced zoom-out. The day-to-day rewards shipping. The cycle is the calendared excuse to step back and ask where someone is going over a year, not a sprint.
• Linking growth to reward. Reviews are where development connects to compensation and level. Decoupling them entirely makes the cycle feel pointless; coupling them too tightly poisons the development conversation.

The Inputs

A good review triangulates from several sources rather than resting on one manager's memory.

A few practices that make inputs useful:

• Ask peers narrow questions. "Give one example of this person doing something at or above their level, and one thing they could do to have more impact" beats "rate this person 1–5."
• Let the person nominate some reviewers, but not all. Self-selected reviewers skew kind; you add a couple they didn't pick.
• Read the self-review before writing yours, not after. It tells you where your view and theirs diverge — those gaps are the most important part of the conversation.

Calibration

Calibration is the meeting where managers compare their assessments across the org to align on what each level and rating actually means. It is the most important and most-skipped part of a fair cycle.

Why it matters:

• It normalizes managers. Without it, "exceeds expectations" means something different from each manager, and your rating depends on who you report to — the definition of unfair.
• It catches halo and horns. Defending a rating to peers forces specifics. "I gave them top marks" collapses fast under "compared to whom, on what?"
• It surfaces level inflation. Two people both called "senior" get held to the same examples in the same room.

How to run it well:

• Bring evidence, not adjectives. Each rating is defended with specific examples mapped to the level definition.
• Anchor on a shared scale. Walk a few well-known people first so everyone is calibrated on what each band looks like before the contested cases.
• Beware forced distributions. A mandated bell curve ("only 10% can be top-rated") manufactures unfairness on small teams and pits managers against each other. If imposed on you, fight to apply it at the largest possible population, never per-team.
• The loudest advocate shouldn't win. Watch for outcomes driven by which manager argues hardest rather than which case is strongest.

The Ratings Trap

Ratings — the number, the label, the bucket — cause more damage per word than anything else in the review.

• The label overwrites the message. Whatever thoughtful paragraphs you wrote, the person remembers "3 out of 5." The rating is a lossy compression that erases the development conversation it was supposed to support.
• Threat response kills learning. The moment someone feels judged and ranked, they stop hearing development feedback and start defending themselves. Neuroscience aside, you've all seen it.
• Reductive by design. A year of varied, spiky work does not collapse to one number without losing most of what mattered.

You usually can't abolish ratings — comp systems need them. What you can do:

• Separate the conversations in time. Have the growth/development discussion in one meeting; deliver the rating and comp news in another. Mixing them means only the number gets heard.
• Make the rubric explicit and shared. People accept a rating they can predict far more than one that feels arbitrary. If they could have guessed it, the system is working.
• Lead with the narrative, end with the label. The words are the value; the rating is the index.

Writing Useful Feedback

Most written reviews fail the same way: they're vague, generic, and backward-only. Useful written feedback is specific, evidenced, and points forward.

• Specific over general. Not "strong technical skills" — "led the payments migration, including the reversible-cutover design that let us roll back the Tuesday incident in minutes."
• Behavior and impact, not personality. "When you pre-wrote the design doc, the review took 20 minutes instead of an hour" tells someone what to repeat. "You're a clear thinker" tells them nothing actionable.
• Tie to the level. Map examples to the career framework so "growth area" means "here's the specific senior-level behavior to build," not a fuzzy ding.
• Balanced and honest. A review that's all praise helps no one grow; one that's all criticism is unfair and demoralizing. Both strengths and growth areas, both specific.
• Forward-looking close. End with two or three concrete things to focus on next, not a verdict on the past.

A reliable structure per section: Context → specific example → impact → what it means for their level → what to do next.

Handling Underperformance

This is the part managers avoid until it's a crisis. The kind, fair approach is to act early and clearly — letting weak performance drift is unfair to the person (who's blindsided later) and to the team carrying them.

A staged approach:

1. Name it early and specifically, in a 1:1. "Here's the gap I'm seeing between your work and the bar for your level, with examples." Not a formal process yet — a direct, private, honest conversation. Many situations resolve right here because the person genuinely didn't know.
2. Diagnose the cause before the cure. Skill gap, motivation, personal circumstances, wrong-role fit, or unclear expectations all look like "underperformance" and need completely different responses. Ask, don't assume.
3. Set clear, written expectations with a timeline. "Here's what 'meeting the bar' looks like concretely, here's the support I'll give, here's when we'll check in." Document it even before any formal process.
4. If it doesn't improve, a fair PIP. A Performance Improvement Plan should be a genuine, supported attempt to succeed — not a paper trail for a firing you've already decided on.

What makes a PIP fair (vs. a sham)

A sham PIP is both cruel and a legal liability. If you've genuinely decided someone has to go, that's a different (and more honest) conversation than dressing it up as improvement. Use the PIP only when you actually want them to succeed.

A guiding principle throughout: document contemporaneously. Fair management of underperformance is impossible to reconstruct from memory and easy to support with dated notes from your 1:1s.

Anti-Patterns

• The annual ambush. Covered everywhere in this section because it's the cardinal sin. The fix lives in the weeks between reviews, not in the review.
• Recency bias. The cycle reflects the last six weeks. Defended only by year-round notes.
• Sandwiching. Burying real feedback between two compliments so it's missed entirely. Be direct; people can handle it.
• Comp-driven ratings. Reverse-engineering the rating from the raise you can afford, then back-filling justification. People can smell it.
• The copy-paste review. Generic language that could apply to anyone on the team. If you could swap two people's reviews and nobody'd notice, you wrote nothing.

The Bar To Hold

A review the person could have substantially predicted, built from dated evidence and multiple inputs, calibrated against peers, where the development conversation is protected from the rating, and where any underperformance was named months earlier — not revealed on the day.

Reviews are the low-frequency consolidation layer. They only work if the high-frequency layer — 1:1s and feedback — is already running.