Why annual reviews under-deliver
An annual review compresses twelve months of judgement into one conversation. Recency bias does the rest. The employee remembers the last six weeks; the manager remembers the last project. Anything from quarter one might as well not have happened.
Mid-year reviews force a second data point. Two checkpoints across the year produce four to five times the usable signal for compensation and promotion decisions, in our internal data.
The sixteen anti-patterns
Common anti-patterns include: ratings inflated to avoid hard conversations, identical wording across reports (copy-paste from a template), no examples cited, no forward-looking development goals, calibration sessions that turn into manager turf wars, and feedback that is conflated with rating.
The single most damaging one is using performance reviews to deliver feedback that should have been given six months ago. The review is for synthesis, not surprise. Calibration sessions are where the rating gets normalised.
The lightweight cycle we use
Two checkpoints a year. Self-evaluation, manager evaluation, one calibration session per skip-level, and a one-page output covering rating, rationale, and three forward-looking goals. The whole cycle takes about ninety minutes per employee end to end.
The supporting tooling lives in our Performance module; the cadence is what matters. Tooling without cadence ships rating fatigue.
What we measure to improve
Rating distribution by manager (looking for inflation outliers), rating-to-promotion correlation at twelve months (validating that ratings predict trajectory), and self-vs-manager delta (where the conversation actually needs to happen). All three live in the manager dashboard.
If the dashboard shows a manager whose ratings consistently outperform their team's actual outcomes, the issue is in the review process. The fix is calibration, not retraining individual managers.