Memory is unreliable. Recency bias means the last six weeks dominate the year. The halo effect means a single strong project shapes how you see everything else. Affinity bias means the people you most enjoy working with quietly get more credit. None of this is malice - it is just how human assessment works when there is no record to check it against. Data does not solve every problem in performance reviews, but it does the one thing that matters most: it gives you something to compare your impressions against, and forces you to notice when those impressions have drifted from what actually happened.
Fair reviews do not happen by intent. They happen by structure. The structure is what stops you praising the person who happened to do something good last week while overlooking the person who quietly delivered all year.
Where bias creeps in
Bias in reviews is rarely the obvious kind. It is the small, structural kind - the parts of the year you can no longer remember, the people whose work happens out of your sight, the moments that stuck for reasons unrelated to performance.
- Recency biasThe last six weeks of the year carry more weight in your memory than the previous nine months combined. Without notes, every review starts skewed.
- Visibility biasLoud work gets noticed; quiet work does not. People who do the unglamorous, behind-the-scenes work that holds the team together are the most often under-rated.
- Halo and hornsA single strong project lifts your view of everything else they did. A single bad moment can drag everything else down. Both warp the picture.
- Affinity biasYou collaborate more naturally with the people you click with. They get more of your time, more feedback, more chances to shine. Without checking, that becomes "they are higher performing."
- Comparison biasReviewing someone right after you reviewed someone else changes how the second one lands. Order effects matter; tooling and structure absorb them.
The data that helps
Not all data is useful for reviews. Vanity metrics make people defensive; the wrong measures encourage gaming. The right data is rich enough to tell a story and modest enough to leave the judgment with the manager.
- Catchup notesNotes from a year of one-to-ones are the single most useful evidence base for a fair review. They surface patterns, capture small wins, and remind you of conversations you have long since forgotten.
- Target progressTargets set at the start of the period and tracked throughout give an honest picture of what was agreed versus what got done. They also surface where priorities shifted, which is itself a fair conversation.
- Action follow-throughHow many of the actions agreed during the period actually got delivered? This is one of the most honest behavioural signals you can capture, and it cuts across both performers and quiet contributors.
- Sentiment trendsA year of catchup sentiment shows the shape of the period - when things were hard, when they felt good, when something changed. It contextualises the rest of the data.
- Peer inputStructured peer feedback fills in the angles you cannot see. Specific questions to specific colleagues beat a broad survey to everyone.
The data that does not help
Not every measure earns its place in a review. Some create the appearance of fairness while quietly making things worse.
- Activity countersNumber of meetings attended, hours logged, lines of code written. Counters reward people who optimise for the counter and penalise people who do focused, lower-volume work.
- Single-source surveysA pulse survey result interpreted as a performance signal is a misuse of the tool. Surveys are great for sentiment and patterns; they are bad as individual evaluations.
- Cross-team comparisonsDifferent teams produce different shapes of work. Cross-team rankings tell you who plays the calibration game well, not who is performing well.
- Anything cherry-pickedA single chart that supports a conclusion you already had is rhetoric, not data. If the data only gets quoted when it agrees with you, it is not making the review fairer.
How to use the data well
Data informs a review; it does not write one. The manager still has to do the judgment work. The trick is to use the data to challenge your impressions, not just to support them.
- Form the view first, then checkWrite your draft assessment from memory and judgment. Then walk through the data and look for places it does not match. The mismatches are where bias lives, and where the data is most useful.
- Compare across the teamLook at the same data points across all your direct reports. Patterns you would have missed for one person become visible side by side, especially follow-through and cadence of contribution.
- Show the data in the reviewWhen discussing follow-through, show the number. When discussing focus areas, show the catchup notes that surfaced them. People accept hard feedback better when they can see how you got there.
- Ask the data what is missingLook for absences as well as presences. A direct report whose name does not appear in any peer feedback all year is not invisible by accident.
Frequently asked questions
Less recency bias, more evidence
Manager Toolkit pulls a year of catchup notes, targets, sentiment, and follow-through into one review so the data is there when you need it.