The problem with ratings that arrive once a year
A yearly rating asks a manager to compress 200 interactions into a number in one sitting. Recency bias decides it: whatever happened in the last three weeks dominates. Employees learn the game quickly — visibility near review time matters more than consistent delivery — and the rating stops measuring work at all.
Rating at the moment of approval fixes the memory problem, but only if you keep the rubric narrow. Five dimensions invite negotiation; ten invite fiction. Four is enough to separate did it right, did it on time, did it without creating work for others, and did it make a good call.
A four-factor rubric with anchors
| Factor | What it measures | 1 — not yet | 3 — meets | 5 — exceeds |
|---|---|---|---|---|
| Accuracy | Correct on first submission | Material errors; rework needed | Minor fixes only | Nothing to change; anticipates edge cases |
| Timeliness | Delivered against the agreed deadline | Late without warning | On time | Early, with room for review |
| Teamwork | Handover quality, clarity, no new work for others | Handover needs chasing | Clear notes and files | Makes the next person faster |
| Judgement | Escalated the right things, chose sensible defaults | Wrong calls or silent guesses | Sound calls within scope | Flags risk before it becomes one |
Anchors matter more than the scale. "3 = meets" is not an anchor; "minor fixes only, no change to the conclusion" is. Write the anchors once, with the team, and then stop editing them — a rubric that moves is a rubric nobody trusts.
Weighting, and a worked example
A default that behaves sensibly in finance and operations: accuracy 40%, timeliness 30%, teamwork 20%, judgement 10%. Statutory or client-facing work can justify shifting weight to accuracy; delivery-heavy teams often move five points to timeliness. Put the weights in the open, apply them consistently, and never tune them to produce a particular answer.
| Task | Accuracy | Timeliness | Teamwork | Judgement | Weighted |
|---|---|---|---|---|---|
| Q2 sales tax return | 4 | 5 | 4 | 4 | 4.3 |
| Vendor reconciliation | 5 | 3 | 4 | 5 | 4.2 |
| Board pack — monthly | 3 | 3 | 5 | 3 | 3.3 |
Weighted = (accuracy × 0.4) + (timeliness × 0.3) + (teamwork × 0.2) + (judgement × 0.1). Rounded to one decimal. In the product this is computed for you and stored with the task.
Five bias controls that actually work
- Rate only approved work. If it was reverted, the rating is not ready. This alone removes most of the "they're a nice person" effect.
- Hide individual scores from peers. Confidentiality is what lets an approver write a 2 without social cost. Aggregates can be public.
- Publish the distribution. If 90% of scores are 5s, the scale has failed and every number after that is noise.
- Separate the process metric from the person metric. A missed deadline caused by a checker sitting on the file is a latency problem, not a maker problem — the trail tells you which.
- Never rate in bulk on a Friday. Ratings written in batches drift toward the average; a 15-second rating at approval time does not.
What not to do
- Do not average across unequal work. A 5 on a tea run and a 5 on a statutory filing are not the same signal — weight by task importance, or at least report them separately.
- Do not rate the person you are managing out. If the rating is pre-decided, the rubric becomes evidence against you; people notice within a month.
- Do not publish a leaderboard. It converts a developmental tool into a zero-sum game and reliably produces sandbagging.
- Do not score speed without quality. Timeliness is measured against the deadline that was set with the maker — not against "as fast as possible".
Running it without extra meetings
The rating should be a field in the approval step, not a separate ritual. In TeamControlSheet the approver rates while signing off; the score is written to a confidential table, the approver's own latency is recorded too, and the monthly view shows distribution per reviewer and per person. The team leader reviews exceptions — scores of 2 or below, and reviewers whose spread never moves — instead of reading 200 ratings.
If you already run appraisals, this becomes the evidence pack: three metrics (weighted score, on-time %, rework count) for the period, each traceable to a specific approved task. No reconstruction required.
Ratings questions
How many rating factors should I use?
Four. Accuracy, timeliness, teamwork and judgement cover the dimensions that matter for controlled work. More factors reduce consistency because raters start improvising definitions.
Should employees see their own ratings?
Yes, with the task and the note that came with it. Seeing your own evidence is developmental. Seeing each other's is corrosive — keep individual scores confidential to the team leader and approver.
How do I prevent a manager from rating everything a 5?
Publish the distribution per reviewer and review the spread monthly. A reviewer with zero variance is the finding; the fix is calibration, not re-rating the past.
Does a weighted score replace a proper appraisal?
No. It replaces the evidence gathering. The conversation about development, scope and pay still needs a human, but it can start from facts instead of recollection.
What about personal tasks?
Never mix them into a performance score. Personal tasks exist so the maker can track themselves; they are private by policy, and in TeamControlSheet they are excluded from ratings and hidden from the team leader.
Put the control in place this week
Free for five users. No card, no trial countdown. Bring the backlog with a CSV.