Performance reviews are one of the highest-stakes conversations a manager has with their team — and most of the tools we build for them treat the process as a forms problem. Type into a box. Pick a rating. Submit.
But the work happens somewhere else. It happens in pull requests, tickets, channel threads, customer calls, end-of-week emails. By the time review season arrives, that work is scattered across a dozen systems and a year of memory. So managers reach for what they can remember — which is usually the last month, or the loudest voice, or the person who spoke up in the standup that morning. The work deserves better than that.
Owned the Q1 onboarding revamp end to end. Delivered it two weeks early, helped close the Acme renewal, and kept follow-through steady across teams.
The status quo.
Recency bias creeps in because you can only weigh what you can recall. Reviews drift toward the visible work — the demo, the launch, the escalation — and away from the steady, quiet contributions that actually keep teams running. And the writing itself is hard: even a fair manager spends hours per cycle staring at an empty box, trying to find the right phrase. Most teams either accept this or skip reviews altogether.
Our take.
Reviews should be grounded in what someone actually did — and the evidence already exists in the systems your team uses every day. PerfCopilot reads the metadata of that work (PR titles, ticket transitions, channel activity, email headers) and uses it to draft a review with citations: every claim in the draft links back to a specific signal you can audit. The manager edits, refines, decides on the rating, ships.
AI helps with the synthesis. It doesn't replace the judgment. The manager is still the author; the system is the research assistant.
How we think about data.
We read metadata, not contents. PR titles and ticket statuses, yes; message bodies and email contents, no. OAuth tokens are encrypted at rest. We don't train AI models on your data — generation runs through Anthropic's API under terms that preclude training. You own your data, including when you cancel. The full legal version lives on the privacy page; controls, sub-processors, and the DPA live on security.
— Nick
Founder, PerfCopilot · get in touch