Performance reviews are widely disliked for good reasons: recency bias, inconsistent standards, and managers writing under time pressure about a year they half-remember. AI addresses some of those and can worsen others.
- ›The three real problems with reviews
- ›What AI fixes
- ›What it must not do
- ›Running a cycle with AI support
The three real problems
Recency bias. A year assessed from the last six weeks, because that is what is in memory.
Inconsistent standards. Two managers applying different bars, so a rating means different things across teams.
Effort concentration. Twelve reviews written in the same week produce twelve rushed documents.
What AI fixes
Evidence assembly. Pulling the actual record of the period: goals set and met, projects delivered, feedback received, contributions logged. This directly addresses recency bias, and it is the single most valuable contribution.
Consistency checking. Flagging where a manager's ratings diverge sharply from peers, or where language is unusually vague. Calibration input, not a verdict.
Drafting time. Turning evidence plus a manager's assessment notes into a structured, readable draft. The judgment is the manager's; the prose is assisted.
Language quality. Catching vague or non-specific feedback, which is the most common complaint employees have about reviews.
What it must not do
Form the judgment. If the AI decides the rating and the manager approves, the manager has outsourced the part that requires accountability. Employees detect this quickly, and a review that nobody actually thought about is worse than no review.
Score employees from activity data. Task counts, message volume, and hours are motion metrics. A review built on them measures visibility, not contribution.
Predict future performance or flight risk as an input to the review. This crosses from assessing work into profiling a person.
Important: The test is whether the manager could defend every sentence in a conversation with the employee. If the answer requires reading what the AI wrote to remember what it says, the review is not theirs.
Running a cycle with AI support
What would help more than AI
Honestly: more frequent, lighter feedback throughout the year, so the annual review summarizes a continuous conversation rather than substituting for one. Companies that do this find review quality improves more than any tooling change achieves.
FAQ
Should employees know AI helped draft their review?
Yes. Disclosure costs nothing and discovering it later damages trust disproportionately.
Can AI write self-reviews?
Employees can use it the same way, drawing on their own evidence. Same caveat: the judgment should be theirs.
Does this help with bias?
Evidence assembly reduces recency bias, which is real. It does not address bias in what gets recorded or in how contributions are attributed, and it can encode both.
Brainis assembles review evidence from goals, feedback, and delivered work, and leaves the judgment with the manager. See People OS.
Sharing insights on business operations, AI, and modern team management.
Run your company on Brainis
All 11 Operating Systems on every plan, from $29 a month. No per-seat pricing — you pay for AI capacity, not headcount.
See pricing