At five people you cannot give feedback privately, because privacy scales with headcount and you do not have enough of it. Give it the same way every time, in a slot that already sits in the calendar soon after the work, and keep the whole conversation on one piece of work. Ask what the person saw before you say what you saw. Put the correction behind a closed door and the credit in front of everyone, with the same precision on both. Say nothing about one person to another, because in a room that size it gets back to them.
Privacy is not available to you at five people
You can close the door. You cannot make the closing of the door invisible. Everybody saw you take one person into the front room, and a version of it reaches the rest of the team.
Two things follow. Your method has to survive being described, since a technique that works only while nobody knows what you did does not work. And some of the talking between staff about pay and conditions is protected activity, so it is worth knowing how far that goes. The National Labor Relations Board sets out that employees covered by the National Labor Relations Act have the right to join together to improve their wages and working conditions, with or without a union, and its own examples of protected activity include two employees raising pay with their employer and two employees discussing work issues with each other. Most private-sector employees are covered. The Act excludes supervisors, independent contractors, and people employed by federal, state, or local government, and in a firm this size the senior people may sit outside it. This is not legal advice. Treat the running conversation about pay and conditions as part of the working environment, and manage inside it.
So secrecy does not work as a control. Consistency does. If the team knows that corrections go to one person at a time, soon after the work, about the work, there is nothing to repeat, because what you did is public and what you said belongs to the two people in the room.
What the feedback studies found
Kluger and DeNisi pooled 607 effect sizes from 131 studies in Psychological Bulletin in 1996 and found an average effect of d = .41, a moderate gain. In the same analysis, over a third of the measured feedback interventions lowered performance. Their own summary of the split is one sentence: "The results suggest that FI effectiveness decreases as attention moves up the hierarchy closer to the self and away from the task." Read that share as a share of the interventions they studied, many of them laboratory tasks. It says nothing about the rate at which feedback goes badly in your firm.
Exhibit 1
Feedback aimed at the task raises performance and feedback aimed at the person is where the gain fades
Hattie and Timperley opened their 2007 review in Review of Educational Research with the point that feedback is powerful, and that "this impact can be either positive or negative". Across 12 meta-analyses of feedback in classrooms, 196 studies and 6,972 effect sizes, they reported an average effect size of 0.79 against a schooling average of 0.40. The review is of schools. Its structure travels to a small firm, and its numbers are students' achievement scores.
Name the piece of work in the first sentence
Say what happened in the work. The proposal went out with last year's rate table in it. The client's question sat in the drafts folder. Both are facts about a document and a queue, and the person can act on them next week without defending himself.
Then comes the drift off the work. You are careless, you used to be better at this. Douglas Stone and Sheila Heen published Thanks for the Feedback with Penguin in 2014, and their framing is that the difficulty sits on the receiving end, where a single comment can land as an evaluation of the person. In a firm of five that landing has nowhere to go except home, and it is sitting in the same room as you on Monday.
"We need to talk about your attitude" is a sentence about a soul, and it invites a defence about a life. Name the proposal, Tuesday morning, the rate table.
A shape with a fixed order
Hattie and Timperley's three questions are "Where am I going?", "How am I going?" and "Where to next?". In a small firm they turn into a short conversation with a fixed order, and the order matters because the second move is the one that gets skipped.
Exhibit 2
Five moves take a correction from a piece of work to a changed next instance
Name the work. Ask what they saw, then wait for the answer. Say what you saw, at task level. Agree what changes. Fix the next instance out loud, in the room, in words the person will say next.
The second move earns its place. Ask first, because the person may know a cause you cannot see from your desk, which was in the handover, or in the fact that the client's second job landed the same morning. Asking first also tells you whether the conversation ahead of you is useful or redundant. On delegating the first real responsibility the same habit carries the handover, and a correction aimed at a handover that was never complete belongs to you.
Timing is the other half. Within the week, one piece of work at a time. A correction delivered a month late arrives as a character assessment, because by then the only thing left in memory is the person.
Corrections in private, credit in the open
Under pressure this inverts. Praise lands in the corridor, where it costs nothing, and criticism in front of four colleagues.
Exhibit 3
What the whole team hears and what stays behind a closed door
Credit in the open is cheap, and it compounds. The team learns which specific behaviour you value, which is information a compliment in a corridor does not carry. Corrections happen one person at a time, behind a door. What the group hears is the standard and the wins.
Sandwiches are legible in a small room
The praise sandwich is folk practice. No study in our reading supports it, and in a room this size its mechanics become visible fast. The pattern is learned fast, at which point the compliments stop being compliments and become the sound the meeting makes before it gets to the point. You have spent your supply of praise on packaging.
Ratio advice does not rescue it. Jack Zenger and Joseph Folkman wrote in Harvard Business Review in 2013 that the highest-performing leadership teams in one study averaged 5.6 positive comments for every negative one. That study covered 60 business-unit leadership teams at one large company. It describes meetings of leadership teams at a large company, and it sets no quota for anyone managing five people. The widely quoted 5 to 1 figure comes from John Gottman's research on couples, which Zenger and Folkman noted as a comparison. In American Psychologist in 2013, Nicholas Brown, Alan Sokal and Harris Friedman concluded that the claim of a critical minimum positivity ratio of 2.9013 was entirely unfounded. Their objection is to the mathematics behind the ratio. Use a ratio as a prompt to notice your own balance across a week.
Your mood is the loudest feedback in the building
Whatever you say in the slot, the team spends the rest of the week reading your face and your tone in the group chat. Marcus Buckingham and Ashley Goodall argued in Harvard Business Review in 2019 that telling people what we think of their performance does not help them thrive; the piece is an argument about management practice from the March to April 2019 issue, and it reports no study. Kim Scott's Radical Candor, published by St. Martin's Press in 2017, puts two behaviours together, care personally and challenge directly. Her company's site names the two ways that pair fails: caring without speaking up is ruinous empathy, and speaking up without caring is obnoxious aggression.
The first failure is the common one here, because most of the staff are people you like and some are people you have known for years. Ruinous empathy in a small company has a specific cost. The client finds out what you knew, later than you did, and the rest of the team notices who let it through.
One-to-ones carry most of the weight
Andrew Grove wrote in High Output Management in 1983 that the one-on-one is the principal way a supervisor's working relationship with a person is maintained, that the meeting belongs to the subordinate who sets its agenda and tone, and that an hour is the minimum, since a short meeting pushes the conversation toward whatever can be settled quickly. Gallup's guidance runs the other way on length. A weekly conversation of 15 to 30 minutes does more than an occasional long one, as long as it happens every week. Gallup reports that 80% of employees who say they received meaningful feedback in the past week are fully engaged, and that among nearly 15,000 employees only 16% said the last conversation with their manager was extremely meaningful. Those are Gallup's own survey numbers from pages updated in 2026, and the survey year is not printed on the page.
Take both. A weekly slot that cannot be cancelled, long enough for the agenda to go where the subordinate wants it, beats a monthly performance meeting you prepare for. Grove is right about something the clock does not capture. If the only time a person can raise something is a slot built around your agenda, they will not raise it.
Safety as the byproduct
Amy Edmondson defined team psychological safety in Administrative Science Quarterly in 1999 as a shared belief that the team is safe for interpersonal risk taking. She tested it in a study of 51 work teams at a manufacturing company and found psychological safety associated with learning behavior, with learning behavior mediating between safety and team performance. Google's re:Work guide reports that Project Aristotle studied 180 Google teams, 115 engineering project teams and 65 sales pods, and ranked five team dynamics in order of importance, with psychological safety first; the teams ranged from three to fifty people, with a median of nine. Google describes it as people feeling able to admit a mistake or offer a half formed idea without being punished for it by teammates. Those were Google teams measured on Google's own measures, so treat the ordering as a lens for your firm.
A correction that names the work and ends with a next instance is a low-risk act to be on the receiving end of. A correction that names the person gives people a reason to hide the next mistake, and at five people a hidden mistake reaches the client.
The person who is also a friend
Use the same shape, with more of the first question. The friendship is a reason to be exact about the work, because vagueness is what a friendship absorbs badly. "This one had the old rate table in it" is a sentence a friend can hear. A general assessment of their standards is the friendship wearing a work coat.
Say the thing in the slot. The barbecue is where the friendship lives. If a conversation has to be prepared for, an AI pass on your notes puts the facts in order and writes the first sentence, and that is most of what preparation is for.
Write down the serious ones
After a correction that matters, send one dated paragraph covering what happened, what it cost, what was agreed, and what happens at the next instance. The EEOC tells small businesses to put factual detail into evaluations, on the grounds that details help an employee see why a decision was made and help managers apply the same standard to everyone; the same guidance starts earlier, with telling people your performance standards when they join. SHRM reports an employment attorney's advice to record what the employee said as well as what the manager said, because a two way note is a fairer record. That item is one attorney's conference remarks from 2022.
The note protects both sides, and mostly it protects the conversation. Later nobody has to reconstruct what was agreed, and neither of you has to argue about whether it happened. Keep the standard in writing from the first week and keep the notes factual and two way. A checklist for the work itself removes a class of corrections altogether, which is cheaper than getting better at giving them.
The quarterly version of the same conversation
Once a quarter, widen it to role, rate, the next piece of responsibility, and what is in the way. This is the same shape at longer range, and it goes badly if it arrives as a surprise after a quarter of silence, which is why the weekly slot comes first. What goal-setting research says about quarterly goals covers what makes a target stick, and the weekly review is where you notice the correction you owe before it becomes a serious one.
When you can stop doing this yourself
The sign is a Tuesday when somebody flags their own mistake in front of the team, and someone else says the useful thing about it before you get there. That is the whole standard. An annual calibration cycle and a performance framework with descriptors are beyond a firm this size; you get a habit and one dated paragraph in a file.
Where this becomes the recurring problem, with standards nobody was ever given and corrections that land as moods, our management consulting work starts on those two things, the standard written down and the weekly slot that holds.