Performance Reviews

Reviews Shouldn’t Reward the Last Six Weeks

Recency bias makes a review rate the last sprint instead of the whole year. The cure every expert prescribes is dated, per-person notes you actually read before you rate. Here is the template, the checklist, and the habit that makes it stick.

Recency bias in performance reviews is the tendency to rate someone on what they did in the last few weeks instead of the whole review period. It is why a strong year gets dragged down by one rough sprint in November, and why a quiet year gets a glowing review because the person shipped something big the week before ratings were due. The fix every expert prescribes is the same: keep dated notes on each person all year, then read them before you rate. The catch is that "keep notes all year" only works if capturing a note is effortless. Otherwise you do not do it, and you are back to recalling six weeks at review time.

What recency bias actually is

It is not laziness and it is not favoritism. It is how memory works. A year of someone’s work compresses into a handful of vivid recent scenes, and those scenes feel like the truth because they are the easiest to picture. By review season you are not recalling the year. You are recalling the last month with footnotes, no matter how fair-minded you intend to be. That is the trap, and concentrating harder does not fix it. You cannot strain your way back to information that was never durably written down.

Recency rarely travels alone. Reviews tend to drift through a small family of predictable errors, and they all share one root cause: rating from an impression instead of a record. Knowing the tell for each one makes it easier to catch yourself in the act.

Bias What it does The tell
Recency Overweights the last few weeks against the whole period. Your examples all cluster in the month before ratings were due.
Halo / horns Lets one strong (or weak) trait color every other rating. Every competency lands on the same number.
Central tendency Rates everyone in the safe middle to avoid hard conversations. The whole team scores a 3, regardless of what they actually did.
Leniency / strictness Shifts the whole scale up or down by personal default. You consistently rate higher or lower than peer managers in calibration.

Recency bias is a capture problem disguised as a memory problem

Here is the reframe that changes everything. You do not have a memory problem at review time. You have a capture problem from nine months earlier. The reason the spring is a blur is not that you are bad at remembering. It is that nobody wrote anything down while it was still true. Every piece of advice about "reading the whole year before you rate" quietly assumes the whole year exists somewhere to read. For most managers it does not, so the advice fails not because it is wrong but because the raw material was never kept.

That is good news, because a capture problem has a fix and a memory problem does not. You cannot promise to remember better. You can promise to write one line when something happens. The fair review is decided long before review season, by whether anything got recorded while it was still accurate. Fix the input, and the output takes care of itself.

The per-person review notes template

This is the asset. Keep one of these per report, add a single dated line whenever something worth noting happens, and tag it so the themes assemble themselves at review time. The running log is what you build all year. The theme roll-up and rating draft are what you fill in the week the review is due, by pulling from the lines you already wrote.

Copy the template
PERSON: [Name]  ·  ROLE: [Title]  ·  CYCLE: [e.g., Jan-Dec 2026]

- RUNNING LOG (newest at top; one line each) -
2026-11-08  Shipped billing migration two days early; zero rollbacks.  #delivery
2026-10-22  Onboarded the new hire start to finish; their first PR merged in week 1.  #mentoring
2026-09-30  Missed the QBR deck deadline; deck was thin. Talked it through in 1:1.  #ownership
2026-07-15  Caught the data-leak bug in staging before release. Saved an escalation.  #delivery #judgment
2026-04-03  Pushed back (well) on scope creep in planning; kept the sprint realistic.  #communication

- THEME ROLL-UP (fill at review time, pulled from tags) -
Delivery:        [paste the #delivery lines]
Communication:   [paste the #communication lines]
Mentoring/team:  [paste the #mentoring lines]
Growth areas:    [the honest, dated misses - with what changed after]

- RATING DRAFT (each rating cites >=1 dated line above) -
[Category]: [rating] - evidence: [date + specific moment]

Notice what the running log does. It is dated, so the calendar is visible. It is one line each, so it is cheap to keep. It is tagged, so a whole year of #delivery moments collapses into one theme with a search. And it keeps the honest misses in plain sight (the September slip) right next to the wins, which is exactly what a fair review needs.

The fair-review checklist

Run this before you finalize each report’s ratings. It is the discipline that turns a pile of notes into a defensible review.

Copy the checklist
FAIR-REVIEW CHECKLIST (per report)
- Read the WHOLE period before writing a single rating
- Pull 2-3 dated examples per competency (not 1, not 10)
- Check the calendar: am I over-weighting the last 6 weeks?
- Balance: one strong recent moment AND one strong early one
- Separate the person from one bad sprint (or one lucky launch)
- Tie every rating to a specific dated entry, not an impression
- Note what changed over time (trajectory, not snapshot)
- Write developmental feedback as behavior + date, not character

Why the usual way fails

The standard advice is "just keep notes" and "read the whole year." Both are correct and both quietly fail, because they skip the step where the notes get created. The win happens on a Tuesday afternoon mid-sprint. You think you will remember it. You do not open a document, because opening a document, finding the right person’s file, and typing a paragraph is friction you will not pay in the middle of real work. By Friday the moment is gone, and by November the whole spring is a blur.

So when the review comes due, the blank form helpfully offers you the last three weeks: the deck someone pulled together, the deadline another nearly missed, the thing said in standup on Monday. This feels like remembering. It is the most recent sediment, and you are about to mistake it for the riverbed. The fix is not heroic recall at review time. It is removing the friction from capture so the year writes itself down as it happens.

What changes when the year is written down

The difference is not subtle. Below is the same employee, the same year, reviewed two ways: once from memory in November, once from a year of dated notes. Nothing about the person changed. Only the record did.

From memory (November) From a year of dated notes
"Had a rough end to the year." "Shipped 6 of 7 committed projects across the period; one slip on Sep 30 (QBR deck), addressed in 1:1 and corrected by the next cycle. Caught a data-leak bug in staging (Jul 15) that prevented an escalation."
Overall rating: Meets. Overall rating: Exceeds, with one dated growth area.

The November version is not unfair on purpose. It is unfair because the rough patch was recent and loud, and the six shipped projects were spread across months that memory had already sanded smooth. The dated version is not kinder. It is just more accurate, and accuracy is what a review owes the person being rated.

How a per-person memory makes this effortless

The cure everyone prescribes (continuous, dated, per-person notes) is exactly what Notivo is built to make effortless. When something happens, you type one quick note and tag the person with @Sarah. It files itself into Sarah’s timeline, dated automatically. Add a topic tag like #delivery and the same line also joins a topic timeline you can read across the whole team. At review time you open the name, the year is already there in order, and prep becomes reading instead of remembering.

That is the whole loop: capture in seconds, organized by person and topic, surfaced when you need it. Today that means web capture, @people and #tags, per-person and per-topic timelines, and search and export. Your notes are private by default, account-scoped, and stored on secured infrastructure. To be clear about what that means: it is platform-level security, not end-to-end or zero-knowledge encryption, and we will not pretend otherwise.

(An AI assistant that works from your own notes is available on Plus, with a 14-day free trial. This checklist itself needs neither.)

Frequently asked questions

What is recency bias in performance reviews?

Recency bias is the tendency to weight recent work more heavily than the full review period. It is normal memory at work, not favoritism: the last few weeks are vivid and easy to recall, so they crowd out months of earlier work. The result is a rating that reflects a season instead of a year.

How do you avoid recency bias in reviews?

Keep dated, per-person notes throughout the year, then read the whole timeline before you write a single rating. Pull two or three dated examples for each competency, balance an early moment against a recent one, and tie every rating to a specific entry rather than an impression.

How far back should performance review notes go?

Cover the full review period. Month one matters as much as last week. If your cycle is annual, your notes should reach back a full year so a strong but quiet spring is not erased by a loud autumn.

Is recency bias the same as halo bias?

No. Recency bias is about time: recent work overweights the rest of the period. Halo (or horns) bias is about one trait coloring everything else. They are different errors, but the same fix reduces both, because dated, specific evidence forces you to rate each competency on its own record.

What are good review examples that avoid recency bias?

Evidence-based, dated specifics rather than impressions. For example: "During the March outage, posted a status update every 30 minutes and kept stakeholders calm (Mar 12)." A dated, behavior-based line proves the work happened across the period, not just before ratings were due.

How do I prepare for a performance review as a manager?

If you kept notes all year, prep is reading: open the person’s timeline, group by theme, and draft each rating from a dated entry. If you did not, gather what you can from your own records, be transparent about gaps rather than inventing a verdict, and start a running log now so next cycle is reading instead of remembering.

Keep reading

Prep from the whole year

Your notes don’t have to be tilted toward what’s recent.

Type @Sarah, write one line, and the year files itself by person. Walk into the review reading, not remembering. Private by default. Free to start.