Almost nobody likes performance reviews. Managers postpone them, employees dread them, and HR spends six weeks chasing forms that get filled in the evening before the deadline. The usual conclusion is that the process is bad, and the usual response is a new form.
The form is rarely the problem. The problem is that a single meeting is being asked to do several jobs that pull against each other, and nobody decided which one it is actually for.
The four jobs a review is asked to do
A performance review is usually expected to deliver honest developmental feedback, decide a pay increase, inform a promotion decision, and create a written record in case of a dispute. Those four are not compatible.
Consider it from the employee's side. If the same conversation determines your increment, you will present the strongest possible version of your year. That is rational, and it is the opposite of what a development conversation requires, which is an honest account of what you found hard. Ask someone to be candid about their weaknesses in the meeting that sets their salary and you are asking them to bid against themselves.
The fix is not complicated and it is unpopular because it means more meetings. Separate the development conversation from the pay conversation by four to six weeks. The rating still informs the increment, but the two discussions are not the same event, and the developmental one stops being a negotiation.
A conversation that decides your money and coaches you on your weaknesses will always resolve into the first thing. People are not being difficult when this happens. They are responding sensibly to what is at stake.
Annual, quarterly, or continuous
The annual cycle gets criticised heavily, and the criticism is mostly right: twelve months is too long to wait for feedback, and it produces a document about the last six weeks with the earlier part of the year quietly forgotten.
What replaces it is often worse in practice. Companies move to quarterly reviews, discover they have quadrupled the administrative load without changing what happens in the conversations, and quietly drift back. Continuous feedback works well and fails when it becomes the excuse for having no formal moment at all, at which point nobody has a record and pay decisions get made on impressions.
A workable middle: keep one formal cycle a year for ratings and pay, add a lighter mid-year check with no rating attached, and make the real work happen in one to ones through the year. The formal cycle should confirm what people already know rather than deliver news.
What has to happen before the review
Most of the fairness of a review is determined months before anyone opens the form.
- Expectations written down at the start of the period: what this person is accountable for, and what good looks like at their level
- Goals that are still relevant, or an honest record of why they changed, because judging someone against goals the business abandoned in month three is a common and avoidable injustice
- Feedback given during the year, so the review confirms rather than reveals
- A standard applied consistently, which is what calibration later checks
The first item is the one most often missing. A rating against expectations nobody wrote down is a rating against what the manager assumed, and two managers assume different things.
Evidence beats recall
Human memory of a year of work is dominated by the last few weeks and by two or three vivid moments. Left alone, a review reflects the recent past and whatever was memorable, which is not the same as what mattered.
The countermeasure is small and boring. Keep a running note per person, a few lines a month, capturing what they did, what went well, what did not, and anything a colleague said about working with them. Two minutes a week per report. When the cycle opens, the manager has a year of evidence instead of an impression, and the difference in the quality of the conversation is obvious to the employee.
Peer input helps if it is asked for specifically. "Any feedback on Priya?" produces politeness. "What was it like working with Priya on the migration, and what would you want her to do differently next time?" produces something usable. Collect it from people who actually worked with the person, not from everyone.
The self-assessment, and its one flaw
Self-assessments are worth having. They surface work the manager did not see, which in any distributed or matrixed team is a substantial amount, and they show you how the person reads their own performance, which is diagnostic on its own.
The flaw is that people do not self-promote equally. Confidence in self-description varies by personality, culture and background, and it correlates with things that have nothing to do with performance. If the self-assessment is weighted heavily, you end up rewarding the people who write about themselves most assertively.
So treat it as input, not as evidence. Ask for specific contributions with outcomes rather than an overall self-rating. If you do ask for a self-rating, read it after you have formed your own view, so it informs rather than anchors.
Rating scales people can actually use
Four or five points, with words rather than numbers alone, and a written description of what each level means for this role family. Whatever scale you pick, two things decide whether it works.
First, the middle rating has to be genuinely acceptable. Where "meets expectations" is treated as a euphemism for disappointment, managers inflate everyone to the level above, the scale collapses into its top half, and the ratings stop distinguishing anything. That has to be fixed with words from leadership, repeatedly, because everyone has learned otherwise somewhere else.
Second, the scale needs anchors. "Exceeds expectations" means nothing on its own. "Delivered beyond the scope of the role and raised how others in the team work" means something you can look for evidence of. Write the anchors once per role family and reuse them.
Calibration, and why it beats forced distribution
Forced distribution, the bell curve, requires a fixed percentage of people to land in each rating band. It remains common in large Indian companies and it causes specific damage.
In a team of eight, a mandated bottom decile means someone gets a poor rating because of arithmetic. Managers respond by hiring a weak performer to protect their good ones, or by trading ratings across teams, and everyone learns that the number reflects distribution rules rather than performance. It also breaks the moment a team is genuinely strong, which is precisely the team you least want to demoralise.
Calibration is the thing people actually want when they reach for a curve. Managers meet before ratings are communicated, present their proposed ratings with the evidence, and challenge each other. The purposes are specific.
- Check that different managers are applying the same standard, since one manager's "exceeds" is another's "meets"
- Require evidence for outlier ratings in both directions, because an unexplained top rating distorts the scale as much as an unfair low one
- Look for patterns: are ratings clustering by gender, tenure, location, or by who sits near the manager in a hybrid team
- Surface people doing significant work that their own manager has under-described
A calibration session is not a negotiation about quotas and should not become one. If the outcome is that a strong team gets strong ratings, that is a correct outcome, and the response is to check the evidence rather than to redistribute.
One rule makes it work: nothing is communicated to any employee until calibration is finished. A manager who has already told someone their rating will defend it in the room regardless of what the evidence shows.
The conversation itself
The single most important rule is that there is nothing new in it. Every point in a review should have been raised at the time it happened. A weakness first mentioned in an annual review is a failure of the manager, not of the employee, and it is the fastest way to lose someone's trust in the whole process.
Structure that works in about forty-five minutes:
- Give the rating in the first few minutes. Holding it back means the person spends the meeting waiting for it and hears none of the rest
- Give the evidence behind it, specifically, including the parts that were strong
- Ask for their reaction and let there be a silence. This is the part managers rush and where the useful conversation usually is
- Move to what happens next, which should take the majority of the time
- Agree two or three things, and a date to look at them again
Delivering a below-expectations rating is the part people avoid, and avoidance is why so many of these arrive as a shock. Be direct about the rating, be specific about the gap, be clear about what changing it requires, and be honest about the timeline. Softening it into ambiguity is kinder in the room and considerably crueller three months later.
When someone disagrees, the useful response is not to defend the rating harder. Ask what they think the evidence shows and what you have missed. Sometimes you have missed something, in which case change it and say why. Where you have not, say plainly that you have heard them, that the rating stands, and record their disagreement in the document. A review with a recorded disagreement is a more honest artefact than one where the employee was worn down into agreeing.
Linking reviews to pay and promotion
People will work out the relationship between rating and increment whether or not you publish it, and the version they invent is usually less generous than the truth. Explaining how ratings translate into pay decisions, and what else affects them such as budget and market position, removes a great deal of suspicion.
Promotion is a different decision from performance and should be treated as one. Performing strongly at your current level is evidence you do that job well, not that you will do the next one well. Where promotion criteria exist, they should describe the next level rather than an outstanding version of the current one, and the two decisions should not be silently merged into a single rating.
Where a large share of the workforce is on an April cycle, as in much of India, the timing pressure is real. That is an argument for starting calibration earlier, not for compressing the conversations.
After the review
The review ends and nothing happens. That is the most common failure and it is the reason people describe the process as theatre.
- Two or three development items, not ten. A plan with ten items is a document
- Each with something concrete attached: a project, a person to learn from, a piece of scope to take on
- A date in the diary to look at them, which is what separates a plan from an intention
- The manager's own commitments recorded too, since half of most development plans depend on the manager doing something
- The written record filed, because it matters for pay decisions, promotion cases and, occasionally, for a dispute
When a performance improvement plan is legitimate
A PIP is legitimate when the intention is genuinely to help someone succeed: a clear description of the gap, specific and achievable targets, a realistic period, the support the company will provide, and a fair assessment at the end that can go either way.
It is not legitimate when the decision has already been made and the plan exists to build a file. People can tell the difference almost immediately, colleagues can tell too, and the damage extends well past the individual. If the decision is that someone should leave, the honest route is that conversation, handled properly, rather than a process with a predetermined outcome.
On the formal side, documentation matters and the requirements are not uniform. Notice, procedure and what counts as fair process depend on the appointment letter, the state Act your establishment falls under, and whether standing orders apply. Take advice before starting anything that could end in an exit, rather than after.
Where reviews go wrong
- Recency bias, where the last six weeks become the year
- The halo effect, where one strong quality lifts every rating on the form
- Central tendency, where a manager rates everyone in the middle to avoid difficult conversations
- Ratings decided by which manager argues hardest in calibration rather than by evidence
- Surprises, which destroy trust in the process faster than any unfair rating
- Judging people against goals the business changed and never rewrote
- The review being the only structured feedback anyone gets all year
- Forms designed for the HR system rather than for the conversation, which is how a process ends up with eleven fields and no discussion
Frequently asked questions
How do you run a performance review?
Set expectations at the start of the period, gather evidence through it, ask for a self-assessment, calibrate ratings across managers before communicating anything, then run a conversation containing no new information: rating first, evidence, their reaction, and most of the time spent on what happens next.
Should performance reviews be linked to salary?
The rating can inform the increment, but hold the two conversations separately, four to six weeks apart. When one meeting decides both, people present their strongest case rather than an honest account, and the development discussion disappears.
What is calibration in performance management?
A session where managers present proposed ratings with evidence before anything is communicated, so the same standard is applied across teams. It also surfaces rating patterns by gender, tenure or location, and unexplained outliers in both directions.
Is the bell curve good for performance appraisal?
Forced distribution creates poor ratings by arithmetic rather than by performance, particularly in small teams, and it penalises genuinely strong teams. Calibration achieves the consistency that companies actually want from a curve, without the quota.
How often should performance reviews happen?
One formal cycle a year for ratings and pay, a lighter mid-year check with no rating attached, and the real feedback happening continuously in one to ones. Moving to quarterly ratings usually multiplies administration without improving the conversations.
What should a manager say in a difficult performance review?
Give the rating early, be specific about the gap with evidence, be clear about what changing it requires and by when, and let there be silence for the person to respond. Nothing in the review should be information the employee is hearing for the first time.
What if an employee disagrees with their rating?
Ask what they think the evidence shows and what you may have missed. Change it if they are right. If not, say plainly that the rating stands and record their disagreement in the document rather than pressing for agreement.
When should a performance improvement plan be used?
When the intention is genuinely to help the person succeed, with a clear gap, achievable targets, a realistic period, defined support and an outcome that can go either way. A plan written to build a file for an exit is recognised as such by everyone involved.
If you change one thing this cycle, separate the pay conversation from the development conversation. It costs one extra meeting per person and it changes what people are willing to say in the first one. If you change two, add calibration before anything is communicated. Between them, those two account for most of the difference between a review process people take seriously and one they endure.
