Most of the work in knowing how to run performance reviews happens outside the meeting room: a record kept through the year, the employee’s own account collected first, a conversation about specific pieces of work, and next steps with owners and dates. The case for working from a record rests on an uncomfortable finding. In a 2000 study of how managers were rated by their bosses, peers, direct reports and themselves, Steven Scullen, Michael Mount and Maynard Goff found that each rater’s own tendencies accounted for more of the variation in scores than the managers’ performance did.¹ A review drawn from one person’s memory of the year leans on the shakiest part of the process.

Here is the sequence, with what each step looks like on an ordinary week:

1. Prepare evidence from the whole period (a few dated notes per person, every week)
2. Ask for the employee’s view before you write yours (three short questions, sent a week ahead)
3. Keep the conversation on the work (one dated example at a time, each followed by a question)
4. Agree next steps and book the first check-in (two or three actions, each with an owner and a date)

Why reviews disappoint when they become paperwork

Performance reviews disappoint when they turn into a form completed once a year instead of the summary of a year of conversations. Elaine Pulakos and Ryan O’Leary made that case in a 2011 paper: formal systems reduced performance management to prescribed steps, cut off from the everyday work that makes it effective, such as setting clear expectations and short-term objectives and giving guidance as the work happens.²

Theirs is a reasoned case, not a tested result, and the logic is easy to follow. A once-a-year form asks memory to do the job of a record, and it asks one meeting to carry pay, status, criticism and plans at once.

Myth

A better review form will fix performance reviews.

Fact

Pulakos and O'Leary argue that new forms and systems alone do not make performance management work; the everyday conversations between reviews matter more.

Picture a designer who hears in December that a handoff in March went badly. The comment may be fair, but it arrives nine months too late to fix anything, and it lands as a verdict rather than as help. If anything in a review would surprise the employee, the gap opened months earlier, in the ordinary weeks between reviews.

This guide opens our coverage of performance management among the guides to hiring, onboarding and developing people. It stays with the review meeting itself; the wider system and how often to hold reviews are separate questions.

How to run performance reviews in four steps

The four steps follow the order of the work: gather, listen, discuss, agree. Each rests on a different body of research, and none of that research tested a whole review process from start to finish. Reviewing 100 years of appraisal studies in 2017, Angelo DeNisi and Kevin Murphy named performance management as an area still in need of research.³

1. 1Prepare evidencea log kept through the period
2. 2Ask for their viewtheir account, a week ahead
3. 3Discuss the workdated events, not character
4. 4Agree next stepsowners, dates, a booked check-in

Then repeat from “Prepare evidence”

The loop matters more than any single meeting. What you agree in step 4 becomes what you log in step 1 next time, so the following review starts from a record rather than from a blank page.

1. Prepare evidence from the whole period

Before you decide what you think, collect what happened: dated examples, results against what was agreed, and what colleagues or customers saw. The reason is the rater problem that opened this guide, and it is larger than most managers assume.

The study

Moderate evidence

Seven raters per manager: what the scores tracked

Each manager was rated on three performance dimensions by two bosses, two peers, two direct reports and themselves. Raters’ idiosyncratic tendencies accounted for 62% and 53% of the variation in ratings in the two samples. The managers’ performance, general and dimension-specific together, accounted for 21% and 25%, less than half as much.¹

Put plainly, the scores said more about each rater than about the manager being rated. Think of a tough grader and a generous one scoring the same year of work: the gap between them is about them, not the work. Two caveats apply. These were ratings collected for development rather than pay, and a 2010 study of two large samples of multisource ratings by Brian Hoffman and colleagues found that a rater’s level, such as boss or peer, explains much more of the variation than earlier work suggested.⁴ Either way, one rater’s impression is a thin basis for a judgment.

The practical answer is a record kept as you go. In field experimentsfield experiment: An experiment run in a real setting, such as a workplace, shop or website, with people going about their ordinary lives, usually with random assignment. It shows how an effect plays out in practice, though often for one setting and one group at a time.Full entry in the glossary in two organizations, reported in 1996 by Angelo DeNisi and Lawrence Peters, raters who kept a structured diary of performance, or recalled it in a structured way, remembered more at review time, reacted better to the appraisal process, and gave ratings that were less inflated and better at telling people apart.⁵

1. Memory alone: an illustration, not data; without a record, a review can draw only on the moments that come to mind
2. A running log: in field experiments in two organizations, raters who kept a structured diary recalled more and gave ratings that told people apart better

A log need not be elaborate. Five minutes every Friday is enough to note, for each person, one thing that went well, one thing that did not, and the date. By review time you have examples from every month, not only the vivid ones. Rater bias and calibration meetings, the wider fixes organizations use, will get guides of their own.

2. Ask for the employee’s view before you write yours

A week or so before the meeting, ask the person for their own account of the period: what went well, what was hard, what they need and what they want next. Having a say is linked to how people receive the review that follows.

A 1998 meta-analysismeta-analysis: A study that combines the results of earlier studies on the same question into one overall estimate. Pooling makes the estimate more precise, but it cannot repair the studies it pools: a meta-analysis of surveys is still survey evidence.Full entry in the glossary of 27 field studies by Brian Cawley, Lisa Keeping and Paul Levy found that participation in the appraisal was linked to employees’ reactions, most of all their satisfaction. Being heardemployee voice: Employees speaking up with ideas, questions or concerns meant to improve how things work, especially to people above them. How often it happens depends heavily on how open managers are seen to be.Full entry in the glossary for its own sake mattered more for most reactions than having a say in order to change the result.⁶ The studies were correlational: asking for a view goes with better reactions, which does not prove the asking produces them.

Expect the two accounts to differ. A 2009 meta-analysis by Heike Heidemeier and Klaus Moser found only modest agreement between self-ratings and supervisors’ ratings, and in Western samples people rated themselves more generously than their supervisors did.⁷

Keep the request short and open. Three questions are enough: what are you proudest of from this period, what got in the way, and what would you like more of or less of next time? Ask for a few lines, not an essay, and read them before you draft your own notes, so your first impression is not the only one on the table.

Read the gap as information. If someone rates a project highly that you saw as late, one of you is weighing evidence the other has not seen, and the conversation now has a real question to answer. A planned guide on self-assessments will look at whether the written form earns its place.

3. Keep the conversation on the work

Open with their account, then set yours beside it with specific, dated examples, and keep every comment on the work rather than on the person. Feedback can backfire. In a 1996 meta-analysis, Avraham Kluger with Angelo DeNisi found that it helped on average but left performance worse in over a third of the results pooled, and comments that threatened self-esteem were linked to weaker results.⁸

The site’s guide to keeping feedback on the work in one-on-one meetings covers that evidence and how it sounds in practice, so it is not repeated here. The review is simply the meeting where the risk is highest, because the person knows something is being decided about them.

The relationship around the meeting counts for a great deal. Shaun Pichler’s 2012 meta-analysis found that how employees reacted to an appraisal tracked the quality of their relationship with the supervisor, such as trust and support, more closely than it tracked participation or the rating itself.⁹ A good review script cannot repair a year of neglect; a year of regular conversations makes a hard review easier to hear.

Go through the period in order, and let the employee respond to each example before moving on. A review that lists five criticisms and then asks “any thoughts?” gives the person nothing to answer except the whole pile at once.

In practice, compare “You’re not reliable with deadlines” with “The supplier report was due on the 3rd and came on the 10th; what got in the way?” The first invites a defense. The second invites an explanation, and often a fix.

If pay or a formal rating is attached to the review, say so plainly and early rather than letting it hang over the discussion. How to connect the two without poisoning the feedback belongs to a separate guide we plan to publish.

4. Agree next steps and book the first check-in

End with two or three concrete actions, each with an owner and a date, and put the first check-in in both calendars before you leave the room. In Kluger and DeNisi’s analysis, feedback that supplied the correct solution was linked to larger gains, while support for goals adding to the effect was only marginal.⁸

This is the everyday behavior Pulakos and O’Leary placed at the center of good performance management: short-term objectives with deadlines, followed up in ordinary weeks.² A useful next step reads like “draft the supplier report template by the 15th; we review it together on the 18th,” not “improve reliability.” The first can be checked in two weeks. The second can only be argued about at the next review. A goal written well is specific enough to check; the site’s guide to writing SMART goals shows how. Where the review has surfaced a real performance problem, the guide on how to manage an underperforming employee takes it from there.

WiserHours

Before and after the review

• Weekly log: one dated example that went well, one that did not
• Employee asked for their own account a week ahead
• Your examples sorted by date, with results against what was agreed
• Opening question ready: how do you see the period?
• Two or three next steps, each with an owner and a date
• First check-in booked in both calendars

Find this checklist online:
https://wiserhours.com/talent-development/performance-reviews/#checklist-performance-reviews-prep

What Deloitte and Adobe changed, in their own words

Two well-known redesigns point the same way, but both are company accounts rather than independent evaluations. In a 2015 Harvard Business Review article, Deloitte reported that forms, meetings and ratings had consumed close to 2 million hours a year, and cited Scullen’s research on rater effects as a reason to change.¹⁰

Its new design asked team leaders about their own future actions toward each person, such as whether they would always want that person on their team, rather than asking them to rate skills, and called for a weekly check-in between each team leader and each team member.¹⁰ Adobe says that in 2012 it replaced annual performance reviews and ratings with Check-in, an ongoing two-way conversation between manager and employee about performance, growth and feedback.¹¹

Neither company published a controlled comparison, and both accounts were written by the organizations themselves. What they share is the lesson of this guide: the formal meeting carries less weight than the frequent conversations around it. Whether to keep ratings at all, and what companies learned after moving to continuous feedback, are questions for planned guides of their own.

Which steps have the firmest research behind them

The steps rest on evidence of uneven strength, mostly correlational and mostly from the 1990s and 2000s. The table sums it up.

The pattern is consistent even where each link is modest: reviews go better when they draw on a record, include the employee’s voice and sit inside a working relationship.

The bottom line

A performance review is only as good as the evidence and the relationship behind it. Keep a short log all year, ask for the employee’s account first, talk about dated events rather than character, and leave with next steps and a booked check-in. Then the review becomes a summary of conversations you have already had.