Performance Review Cycle

You run the process that tells each person, fairly and with evidence, how they performed against what was expected and what to work on next. A good cycle has a clear timeline everyone can plan around, forms that ask for examples not adjectives, a rating scale with written definitions, calibration that makes ratings consistent across managers, and conversations that contain no surprises because the topics were already raised in one-to-ones. You organise, prepare and check; managers judge; a qualified HR professional handles anything that could lead to a formal process.

When to use

• "Set up our first performance review cycle."
• "Run the mid-year reviews."
• "Write the review forms and the manager guide."
• "Prepare the calibration session for the engineering managers."
• "Our reviews feel random, make them fairer."

What you need

• What the reviews are for: development only, or also pay and promotion. This changes the design (see step 1).
• The expectations people are reviewed against: goals set at the start of the period, role or level descriptions, company values.
• The list of people, managers and reporting lines; who joined during the period.
• The owner's choices: cadence, whether to use ratings, whether peer and upward reviews are included.
• Access: an HR or review tool through a connected app, or forms and documents through the agent's own browser signed in by the owner; otherwise files. Review content is confidential personal data: restrict access to the person, their manager, HR and the owner.

Method

1. Decide the design with the owner. Settle: cadence (annual, or twice a year, with quarterly check-ins); components (self, manager, 2 to 3 peers chosen by the person and approved by the manager, optional upward review of the manager); rating scale (recommended: 4 or 5 points with written definitions, see references/cycle-timeline.md); and whether pay decisions use the outcome. If they do, hold the development conversation first and the pay conversation separately, ideally a few weeks later, so people hear the feedback.
2. Build the timeline. Use the template in references/cycle-timeline.md, usually 6 to 8 weeks: announce and brief managers, peer selection, self and peer reviews (with protected writing time), manager reviews, calibration, review delivery, close-out.
   • Put every date in the announcement.
   • Set cronjob reminders for each deadline.
3. Write the forms. Ask for evidence: "What were your most important results this period? Give examples and their impact." "Where did you fall short of what you planned, and why?" "What will you focus on next period?" Peers answer two questions: what this person should keep doing, and what one thing would make them more effective, each with an example. Keep each form to what can be completed in 30 to 45 minutes.
4. Brief managers. Give them the manager guide: rate against the expectations for the person's role and level, not against teammates; use the whole review period, not the last month; write feedback with Situation, Behaviour, Impact (SBI, from the Center for Creative Leadership); separate the behaviour from the person; nothing in the review should be news to the employee. Include the bias list in references/calibration-guide.md.
5. Track completion. Daily during writing windows: who has submitted what.
   • Nudge with drafts the owner has approved.
   • Report blockers to HR.
6. Prepare calibration. For each group of managers, compile a sheet with each person's proposed rating, a two-line evidence summary and the manager.
   • Show the distribution of ratings per manager.
   • Flag: managers whose ratings are all at one level, top and bottom ratings without specific evidence, and people who joined or were on leave during the period (rate only the time worked and do not penalise protected leave such as maternity, parental or disability-related absence).
7. Run the bias checks. After calibration and only with the owner's permission, HR compares rating distributions across groups (for example by gender, part-time versus full-time, remote versus office) using aggregate monitoring data.
   • Large differences are a prompt to re-examine the evidence, not to adjust ratings by quota.
   • Do not rate or rank anyone using demographic data.
8. Support the delivery. Help managers prepare each conversation: the key message in one sentence, two or three strengths with examples, one or two growth areas with examples, next period's goals, and a question to open discussion. The written review is shared with the person; they can add comments.
9. Route performance concerns correctly. If a manager wants a performance improvement plan, a warning, a demotion or a dismissal, that is a formal process with legal consequences that differ by country.
   • Collect the facts the manager gives, point to the policy, and hand it to the owner and a qualified HR professional.
   • Do not draft warnings or plans for sending without their direction.
10. Close out. Archive the final reviews with access controls, record goals for next period, run a short survey on how the process felt, and write a one-page retrospective for the owner.

Judgement calls

• A very small company (under about 10 people): skip calibration sessions and formal ratings if the owner prefers. Keep the self-review, the manager review and the conversation.
• A change of manager mid-period: the previous manager contributes written input; the current manager writes the review.
• A manager rates everyone at the top: ask for evidence in calibration. You never change a rating yourself.
• An employee disagrees with their review: they can add written comments. A formal appeal follows the policy and goes to HR.
• Ratings linked to pay: set the pay budget and guidelines before calibration, so ratings are not worked backwards from money.
• People with agreed adjustments or adjusted duties: assess them against the adjusted expectations.

Output

• review-cycle-<period>/plan.md: design decisions, timeline, roles.
• Forms (self, peer, upward, manager) and the manager guide.
• A completion tracker and calibration sheets.
• Aggregate results card for the owner: completion rates, rating distribution, calibration changes, process feedback.

Cycle status card layout:

Cycle: <period> · People: <n> · Managers: <m>
Design: self + manager + <2> peers + upward <yes / no> · Ratings <yes / no, scale>
Pay linked: <yes / no; pay conversations on <date>>
Timeline: announce <date> · writing <dates> · calibration <dates> · delivery <dates> · close <date>
Completion today: self <x>% · peer <y>% · manager <z>%
Risks: <late managers, new joiners, people on leave>

Checks before you finish

• Managers received the manager guide and bias list before the writing window opened.
• Calibration changes are recorded with the reason for each.
• Every rating level has a written definition that managers received before writing.
• Every top and bottom rating has specific evidence.
• People on protected leave or new joiners were rated only on time worked, with no penalty.
• Pay conversations are separate from development conversations if pay is linked.
• Access to review content is limited to the right people.
• Any formal performance matter has gone to a qualified human, not been handled by you.

Pitfalls

• Recency. The last month dominates. Ask managers to review notes from the whole period.
• Forced distribution by default. Stack ranking can damage collaboration and invite legal risk. Use it only if the owner chooses it knowingly.
• Adjective feedback. "Needs to be more strategic" helps nobody. Require an example and an expected behaviour.
• Surprises. If the review is the first time someone hears about a problem, the manager failed earlier. Coach managers to raise issues in one-to-ones.
• Gendered language. Analyses of written reviews (for example Kieran Snyder's 2014 study of tech-industry reviews) found women received far more personality criticism ("abrasive") than men. Check feedback for traits versus behaviours.

See also: one-to-one-meetings for the ongoing conversations that make reviews surprise-free.