Skip to content
The Growth Track
← All articles
  • Performance reviews
  • Engineering management

How to run fair performance reviews for software engineers

Run engineering performance reviews against shared expectations: preparation, self and peer feedback, evidence, calibration and the conversation itself.

Rachid Baaziz

· 3 min read

Most engineers don't object to being evaluated. They object to being evaluated against criteria they never saw, by a process that seems to depend on which manager they happen to have.

Fair reviews are mostly decided before review week. This guide covers what to put in place, how to run the cycle, and how to keep the conversation about evidence.

Fairness starts before the review

Gallup has found that only 14% of employees strongly agree that their performance reviews inspire them to improve, and only 29% that their reviews are fair. A better form won't fix that. Three things set up a fair review long before it starts:

  1. Published expectations. Every engineer can read what is expected at their level and the next one, area by area.
  2. A known placement. Everyone knows their current track and level.
  3. Goals during the year. Concrete goals with success criteria give the review real material to look at.

If an engineer learns what "senior" means during their review, the review is already unfair, however carefully it is written.

Decide the review structure once

Make the structural choices once, for everyone, and keep them for the whole cycle:

  • Who reviews whom. A self review and a manager review at minimum. Peer feedback adds perspective on collaboration and technical influence that managers often don't see directly.
  • What is assessed. Each area of the engineer's level, using the expectations in your framework.
  • How it is rated. A small, named scale such as Developing, Meets expectations, Exceeds expectations. Some teams prefer written feedback only in their first cycles.
  • Deadlines. Separate dates for self reviews, peer feedback and manager reviews, so each step can use the previous one.

Freeze the framework version a cycle uses. If you change expectations mid-cycle, people are measured against a moving target.

Ask for evidence, not adjectives

"Strong communicator" is an opinion. "Wrote the design document for the billing migration and ran the review with three teams" is evidence.

Ask every reviewer, including the engineer in their self review, to attach examples to their assessment:

  • shipped projects and the decisions behind them,
  • design documents, incident reviews and postmortems,
  • mentoring, onboarding and code review contributions,
  • goals and how they were met.

Evidence keeps reviews grounded and makes it far easier to explain a rating.

Use peer feedback for what peers see

Peers see things managers miss: how someone collaborates, reviews code, unblocks others and handles disagreement. Keep peer feedback focused:

  • Ask for written feedback on a few areas and an overall comment, rather than ratings.
  • Let the engineer suggest peers, and have their manager approve the list.
  • Keep the number of peers small (three to five is usually enough) so each one can write thoughtfully.

Calibrate before the conversation

Calibration is where managers compare their assessments against the same expectations before talking to their reports. It catches the most common sources of unfairness:

  • Leniency and severity: one manager rates everyone highly, another rarely gives Exceeds.
  • Recency: the last month weighs more than the previous eleven.
  • Visibility: work that was loud gets more credit than work that was important.

Keep calibration focused on evidence against expectations. "Does this match what we wrote for L3 Delivery?" is a better question than "Is she better than him?".

The conversation itself

The review conversation should hold no surprises. If the manager has given feedback during the year and goals have been tracked, it becomes a summary and a plan:

  1. Start with the engineer's self review and what they are proudest of.
  2. Walk through each area: the expectation, the evidence, the assessment.
  3. Agree on one or two areas to develop, tied to specific expectations of the current or next level.
  4. Turn them into goals for the next period.

After the cycle

A review cycle is also feedback on your framework. After each cycle, ask managers which expectations were hard to assess, and which ones nobody could tell apart. Revise them in the next version of the framework, before the next cycle starts.

Reviews will never be perfectly objective. They can be consistent: the same expectations, the same structure and the same standard of evidence, for every engineer.

  • · 5 min read

    How to build a software engineering career ladder

    A practical, step-by-step guide to defining tracks, levels, areas and expectations for engineers, without writing a 40-page document nobody reads.

    • Career ladders
    • Engineering management