# Performance reviews that are fair and useful Source: https://sarang.pro/performance-reviews-fair-useful/ Author: Sarang | Published: 25 October 2026 Ask a team what they think of performance reviews and you will often hear the same complaints. The review came as a surprise. It covered only the last few weeks. The rating seemed to depend on which manager you had. None of these problems comes from the idea of a review itself. They come from how reviews are run. This guide sets out a fair, useful process for a small or growing organisation: clear objectives, feedback through the year, evidence instead of impressions, awareness of rating bias, and a calibration step to keep standards consistent. It draws on UK guidance from Acas and the CIPD. ## What a review is for [Acas describes reviews](https://www.acas.org.uk/performance-management) (sometimes called appraisals) as a chance to discuss what an employee does well, where they could improve, any support or training they need, and their career and development goals. It advises reviewing employees regularly, ideally at least once a year, and keeping a written record of what was discussed that is shared with the employee afterwards. Acas also says formal reviews should sit alongside informal conversations: ongoing feedback, coaching and regular one-to-ones. The annual review is where you summarise and plan. Day-to-day feedback is where most of the performance management happens. Be clear about what else the review is used for. If ratings feed pay or promotion decisions, people will treat the conversation as a negotiation. Some organisations hold the development conversation and the pay conversation at different times for that reason. ## Set objectives people can be judged against A review is only as fair as the objectives behind it. Acas recommends objectives that are fair, reflect the person's usual tasks and workload, and can stretch them. It suggests the SMART test: specific, measurable, achievable, relevant and time-bound. The [CIPD's performance management factsheet](https://www.cipd.org/uk/knowledge/factsheets/performance-factsheet/) adds some useful detail from the research: - Specific, stretching goals tend to work well, but for complex or knowledge-based work, learning goals or open-ended goals may work better than rigid targets. - Team goals suit shared work, but individual and team objectives should not pull against each other. - Letting employees set their own goals is not generally better, which the CIPD notes runs against popular opinion. Agree objectives together instead. - Narrow targets can be counterproductive. Write down how each objective will be judged before the period starts. "Improve customer service" is a hope. "Reduce average first response time on customer emails from two working days to one by March" is something both of you can check. ## Make feedback continuous The CIPD describes a shift away from annual appraisals towards ongoing, flexible performance conversations, with more coaching and less emphasis on judging past performance for administrative purposes. Its factsheet also notes a move away from forced ranking and guided distributions of ratings. In practice this means a simple rule: nothing in the annual review should be news. If a problem is serious enough to appear in the review, it was serious enough to raise when it happened. Use one-to-ones to give feedback as you go, and treat the review as a summary of conversations you have already had. ## Collect evidence through the year Memory is a poor record. Keep a short file for each person and add to it monthly. Useful evidence includes: - progress against each objective, with dates and numbers where they exist - examples of work, good and bad, with the context - feedback from customers, peers or other teams - notes from one-to-ones, including support you agreed to give - changes in role, workload or priorities that affected the objectives Ask the employee to bring their own evidence too. A short self-assessment before the meeting often surfaces work the manager did not see. ## Watch for common rating biases Even careful managers make predictable errors. A guide to [common rating errors](https://portal.cca.edu/working/office-human-resources/performance-management/manager-guide-annual-performance-evaluations/avoiding-common-rating-errors/) from California College of the Arts lists several, including these: BiasWhat happensHow to reduce it Recency effectRecent events outweigh the whole review periodKeep notes throughout the year and review them before rating Halo or horns effectStrength or weakness in one area colours ratings in unrelated areasRate each objective or competency separately, with its own evidence Similar-to-me effectPeople like the rater get higher ratings than others performing at the same levelCheck ratings against written standards, not personal rapport Central tendencyEveryone is rated in the middle of the scaleUse the full scale where the evidence supports it Contrast effectPeople are compared with colleagues instead of the job's standardsRate against the role's standards first First impressionAn early judgement shapes later ratingsWeigh evidence from the whole period In the UK there is a legal dimension too. [GOV.UK](https://www.gov.uk/discrimination-your-rights) lists the protected characteristics under the Equality Act 2010, including age, disability, race, religion or belief, sex, sexual orientation, pregnancy and maternity, and gender reassignment, and confirms that discrimination at work is unlawful. Ratings that consistently disadvantage one group are a fairness problem and may become a legal one. If you have enough people, compare rating patterns across groups each cycle. ## Calibrate ratings across managers Calibration is a meeting where managers compare proposed ratings before they are shared, so the same standard applies across teams. The University of Colorado Boulder [describes its process](https://www.colorado.edu/hr/performance-calibration) in two steps: a reviewer, usually the supervisor's manager, reads each evaluation and sends back any that lack clear reasoning. Supervisors then meet to examine rating distributions, check that standards are applied consistently and that no supervisor gives everyone the same score, and raise ratings they are unsure about. Its guidance stresses ground rules that let people challenge each other's ratings, confidentiality outside the room, and accepting that the process is imperfect. A small business can run a light version: two or three managers spend an hour going through proposed ratings, each one backed by evidence, before any review meeting takes place. ## A worked example An example (illustration only). A 20-person marketing agency runs reviews once a year. Last year, one manager rated all five of her team "exceeds expectations" and another rated all six of his "meets". Staff noticed. This year the agency changes four things. Objectives are agreed in January, with a measure for each. Managers keep a monthly note per person. Before reviews, the three managers meet for 90 minutes to calibrate: each proposed rating needs two pieces of evidence, and the group questions any rating that rests on a single recent project. Two ratings move up and one moves down. Every employee sees a written summary after the meeting, as Acas recommends. The ratings are not perfect. They are explainable, and people can see how they were reached. ## Common mistakes, and when to adapt - Surprises in the review. Raise issues when they happen. - Rating the person instead of the work. Tie each rating to objectives and evidence. - Overloading the meeting. Pay, promotion, development and problems in one hour rarely go well. - No written record. Acas advises keeping one and sharing it. - Skipping the forward look. Spend at least half the meeting on next period's objectives and support. Very small teams may not need ratings at all; a written summary and agreed objectives can be enough. Where performance falls well short, a review is not the right tool on its own. Acas has separate guidance on handling problems with an employee's performance, and a formal capability process should follow it.