Psy testsUnderstand yourself betterScanMyLoveDecode your relationshipScanMyJobSee clearly at work

Running Performance Reviews for Your Team: Preparation, Wording, and What Remains Afterwards


A performance review lasts an hour and is poorly prepared. Not out of negligence: because it arrives fifteen times over in three weeks, in a busy period, with a form that came from elsewhere and a budget framing that limits what can be announced.

What follows does not try to add work. It tries to move it: two hours of preparation up front are worth more than the four weeks of repair a botched review costs afterwards.

The four biases to know before you start

Recency

Without deliberate effort, a performance review assesses the last eight weeks. It is the most massive bias and the easiest to correct. It produces two symmetrical injustices: someone whose year was solid and whose last quarter was difficult will be assessed badly; someone whose year was average and whose November was brilliant will be assessed well.

The correction: rebuild the year quarter by quarter, from the calendar and the correspondence, before forming a judgement. It takes twenty minutes per person and it changes the appraisal in one case out of three.

Halo

One dominant impression — often tied to the relationship or to a striking episode — colours every criterion. You rate "communication", "rigour" and "autonomy" at roughly the same level, because what you are actually rating is one single thing.

The sign: if your appraisals are flat — every criterion at the same level for the same person — halo is probably at work. A real person is rarely homogeneous. The correction: assess criterion by criterion, for everyone, before looking at the whole. And actively look for the counter-example: what is this person better at than my general impression suggests?

Asymmetric attribution

This is the most insidious one, and it has been documented for a long time: we spontaneously explain some people's results by the circumstances and other people's by their character. The same delay becomes "he was held up by the supplier" for one and "he struggles to meet deadlines" for another.

The sign: read your own wording. If some sentences talk about context and others about traits, the asymmetry is there. The correction: for every gap you observe, look for two situational causes before concluding. Most of the time they exist — and they are not an excuse: they are information about how the work is organised.

Leniency and severity

Two managers in the same organisation do not assess on the same scale, and the difference is considerable. The person being assessed often knows it, and nothing damages the credibility of an appraisal system more.

The correction: it is not an individual one. That is the role of collective calibration, where it exists. Failing that, keeping your own scale stable from one year to the next, and saying so explicitly, is already progress.

Preparation, in three stages

1. Reread the year, person by person

Twenty minutes each. The calendar gives the chronology — meetings, dense periods, turning points. The correspondence gives the rest: the requests that were made, the deadlines imposed, the warnings raised, the thanks, the periods of overload visible in the timestamps.

This is the only real remedy for recency, and it is also what makes feedback credible: quoting an episode from May proves you have looked at the year, and it completely changes how everything that follows is received. Putting working correspondence in series — volumes, channels, moments — is precisely what ScanMyJob produces, without adding any judgement.

2. Write three facts per person

Two positive, one to work on. Each with a date and an observable effect. If you cannot manage to write three dated facts for someone, that is information in itself: you have not had enough visibility on their work this year, and that is something to say rather than to mask with a general appraisal.

Discover your psychological profile

202 validated tests. Answer 60 questions, instant result, full PDF report from €4.99.

Take a test →

Analyze your conversations

Upload a conversation and get an analysis of your relationship dynamics.

Analyze

3. Decide what you can commit to

Before the meeting, not during. What depends on you, what depends on a decision above you, what is not decided. Improvising a promise in the room under the pressure of an awkward moment is the mistake that costs most — because it will be recalled, rightly, the following year.

The feedback formula, in three parts

It comes in three pieces and it solves the central problem: saying something difficult without talking about the person.

The situation — when, where. The behaviour — what was done or said, factually. The effect — what that produced on the work or on other people.
"In the committee meeting on 12 September, the version presented was not the one signed off beforehand. The client noticed the discrepancy in the room, and the file had to be redone the following week."

Compare that with: "You lack rigour." The second statement is shorter, easier to say, and unusable: it cannot be dated, it cannot be corrected, and it invites a defence of the person — after which nothing else said will be heard.

What the formula forbids, and what gives it all its value: attributed intent. "You didn't want to check" is undemonstrable and will be denied. "The version was not checked" is a fact. The same tool works for the positive, and it pays off even better there. "Good work" produces nothing. "In July, when the client threatened to walk, you proposed the recovery plan within three days and it was accepted — that is what saved the contract" gets remembered, and it also indicates what is expected.

Running a difficult review

Announce the structure at the start. "We'll do the review of the year, then the points to work on, then your requests and the objectives." Someone who knows in what order things are coming listens better to the first part, instead of waiting for the axe to fall. Do not drown the difficult point. The technique of wrapping a criticism between two compliments is very widely recognised by those on the receiving end, and it produces two effects: it discredits the compliments, and it makes the difficult point vague. Better to say the positive for its own sake, then announce the shift: "there is one point I want us to work on this year." Let them answer, genuinely. Feedback with no effective right of reply does not produce correction: it produces surface compliance. And it lastingly damages the relationship — research distinguishes here between procedural justice, which is about the process, and interactional justice, which is about how a person is treated. It is the second that produces the strongest and most durable reactions, and it costs nothing: it comes down to letting people speak, not interrupting, and not concluding beforehand. One single thing to work on, not five. Beyond one point, nothing is retained, and the whole is received as a global judgement. Choose the one with the most effect. Never characterise the person. That is the absolute rule, and it holds both humanly and practically: a behaviour can be dated, documented and corrected; a character trait can be argued over indefinitely, cannot be proved, and puts you in difficulty if the record is reread by others.

The case of the decision you did not take

This is the most frequent situation and the one first-line managers find hardest: announcing no pay rise, an objective judged unrealistic, a reorganisation.

One form of words holds without lying and without disowning: say what has been decided, say what you argued for, do not say what you think.

"This is the budget; there is no increase this year in this scope. I put your case forward with the elements we have just gone through; the decision came back this way. What I can do is A and B."

That sentence is verifiable, it claims nothing, and it does not make you look like the author of a decision you did not take. The two forms of words that fail are "that's just how it is" — which attributes the decision to you — and "personally I agree with you, but they…" — which discredits you for everything you announce afterwards.

That position has a name and a cost: it is a loyalty conflict, and it wears people down more than workload does.

The written record, and what remains

The record will be reread in six months by people who were not in the room. Three rules.

Dated facts, not appraisals. "Average turnaround down from 9 to 4 days between April and September" travels; "good level of engagement" tells nobody anything. Not a word about the person. Neither in the positive nor in the negative. A record that characterises commits the organisation and commits you. Commitments with a deadline. What you do, what the person does, and the date of the follow-up point. A review with no review date will have no sequel: the only way to reopen a subject then becomes a fresh incident. And keep what was promised. This is the point that decides all the rest. Work on the psychological contract shows that it is unkept commitments — far more than severe appraisals — that produce lasting disengagement. A forgotten promise costs more than a frank conversation.

What the person opposite you is going through

One last point, often useful to bear in mind. A negative appraisal produces a physiological state comparable to that of a threat: accurate memory and judgement are temporarily less available. What you say after the difficult point will not be heard, and what the person remembers the next day will be harsher than what you actually said.

Two practical consequences: put the difficult point in the middle and not at the end, and write down what matters — because the written record, unlike the conversation, will be reread when things have cooled, and it will correct the reconstructed version. What the person goes through in the forty-eight hours that follow explains a good part of what happens next.

The tests on the platform help you take stock of your own relationship to feedback — giving difficult feedback draws on exactly the same resources as receiving it. The assistant can help word a delicate point along the three parts.

Three things this article does not do

It says nothing about the applicable frame. Whether these meetings are compulsory, how often they must happen, what formalities apply and what an unsigned document implies belongs to your HR department, an employee representative or the relevant labour authority. It characterises no situation. Neither legally nor clinically. If the state of someone in your team worries you, the person to turn to is the occupational health doctor, accessible at that person's own request and bound by medical confidentiality — not the performance review. It describes nobody. The whole method rests on the fact that a review is about dated behaviours and their effects, never about what a person supposedly is.
In brief: Running fifteen performance reviews in three weeks, on top of your own work, using an imposed form you did not choose: it is the most common and least supported managerial exercise there is. This article takes it from the side of the person running it. It starts with the four biases that distort most appraisals and that are thoroughly documented — recency, which makes the whole year rest on the last eight weeks; halo, where one dominant impression colours every criterion; asymmetric attribution, which explains one person's results by their situation and another's by their character; and leniency or severity error, which makes two assessors incomparable. It then proposes a three-stage preparation built on rereading the actual year rather than on memory, followed by the three-part feedback formula — the situation, the behaviour, the effect — which makes it possible to say difficult things without ever characterising a person. It details how to run a difficult review, how to announce a decision you did not take, and what the written record should contain. It closes on what is at stake beyond the hour: a review run without stated criteria produces lasting disengagement, however good the relationship.
Gildas Garrec, Psychopraticien TCC

About the author

Gildas Garrec · CBT Psychopractitioner

Certified psychopractitioner in cognitive-behavioral therapy (CBT), author of 16 books on applied psychology and relationships. Over 1000 clinical articles published across Psychologie et Serenite.

Was this article helpful?

Join our exchange community on WhatsApp

Join the Community (waiting list)

Need personalized support?

Video sessions (€90 / 75 min) or in-office in Nantes.

Book a video session
WhatsApp
Messenger
Instagram