💡 Core idea
Fairness is a system you run all year. Avoidance is also a system — one that always fails, usually at the expense of the team and of the person you were trying not to hurt.
The “kind” delay turns a coaching problem into a late, formal shock. That is not kindness. It is a surprise, which you already know how to name (Chapter 4), delivered at the moment when options have narrowed.
Ben Kuhn includes delaying difficult conversations among first-year failure modes. It is not only a first-year failure.
Performance is a conversation over time
Set expectations in writing: role, outcomes, behaviors, and how you will review them. A one-pager is enough. A quarterly snapshot — what you observed, what good looks like next, what support you will provide — keeps the conversation from collapsing into a single annual verdict. When you assess the work, look at how much of it is still yours. The 37signals Manager Playbook asks: “If you go back and look at your report’s projects, how much red pen is there? … how much of you or another teammate is represented in the final work product?” Heavy rewrite is a signal about development, and about whether your review has slipped into doing the job twice (Chapter 2). That is not only a fairness habit. Locke and Latham (2002) treat feedback as a moderator of goal effects: goals plus feedback are more effective than goals alone, because people cannot adjust effort or strategy if they do not know how they are doing.
If your current view would surprise the person, you are already late (Chapter 9). Stay SaaSy, in “The Big Management Lie,” argues against over-promising — including to retain someone or to postpone a hard truth. Your word is the relationship. Do not buy a quiet month with a promise you cannot keep.
Grow the people who are doing well
Growth plans are not only for people who are struggling. Career conversations need a sponsor, not a vague encouragement: skills, interests, constraints, next bets, and what you will actually do. If you cannot sponsor the next step, say so. Hope is not a plan.
Promotion is not only a thank-you. The 37signals Manager Playbook is clear: “Strong individual performance is necessary for a promotion, but it’s not enough.” The person should already be doing most of the next-level work, reliably, with evidence you can show — and the promotion should also serve a real business need (ownership, capability, how the team actually operates). Do not use a promised title to retain someone or to delay a hard truth. Until a promotion is approved, “refrain from promising anything on any sort of timeline.”
When coaching must become judgment
Jim Grey is the hinge of this chapter as well as the last: there is a point at which continuing to coach is no longer the honest move. You owe the person a clear statement of the gap, a genuine chance with support and a timeline, and an honest possible outcome — not a performance-improvement plan used as theater.
An improvement plan used in good faith is specific: the standard, the evidence so far, the support, the date, and what happens if the standard is not met. Give a real chance, and a short one. The 37signals Manager Playbook draws a sharp line: “A cycle or two to allow someone to show development is great; six months is not.” Tell the person you are checking, and what you think, while the work is still small enough to fix. If you already know you will exit the person, do not use a plan to buy yourself cover. Follow fair process, with HR and policy, and be clean.
Charity Majors asks a useful, uncomfortable question: if your team is happy, are you doing a good job? Happiness is not a complete definition of the job. Standards still matter. Aviv Ben-Yosef argues that over-protecting a team is not the same as caring, and that the company’s purpose is not “keep everyone comfortable.” Use that carefully. It is not a license to be unkind. It is a warning against confusing comfort with responsibility.
Molly Graham names the “management pretzel”: care that curdles into “over-responsibility for someone’s performance or happiness,” until you are redesigning the org or lowering the bar around one person. Her test: “Would I do this if this person weren’t here?” and, taking the names off the boxes, “what’s the right thing for the business?” “Your job is to create the conditions for success; it’s their job to meet those conditions.” That is not a reason to skip support. It is a reason not to contort the system to avoid a judgment you already owe. The NUMMI case in Chapter 1 is the other half of the test: change the conditions, then hold the standard. Do not pretend a system problem is only a person problem, and do not use the system as a reason never to judge.
Exits with dignity
When someone leaves — especially when you have decided they must — be clear, early enough, and coordinated with policy. No ambush. Care for the person and for the people who remain, who will read the event as information about how this team treats them.
Peers should not have been carrying hidden load for months while you “didn’t want to deal with it.” That cost is also a fairness failure.
Prepare talking points for a high-stakes conversation, not a script. The 37signals Manager Playbook is useful here: write down the gap, the examples, the change you need, and the timeline, then “Stop talking. Say what you want to say, then stop.” In a termination conversation, “It’s not about you, so avoid phrases like ‘I don’t know how to say this’ or ‘I feel terrible’.” An approved AI assistant can stress-test your notes — flag what sounds personal rather than professional, or unsupported — but “Do not use AI to create a script.” You will sound rehearsed, and you will fail to have a two-sided conversation. Follow your employer’s policy and involve HR at the right time.
Tools
- Expectation one-pager. Role, outcomes, behaviors, review rhythm.
- Quarterly performance snapshot. Observations, next standard, support.
- Career conversation guide. Skills, interests, constraints, next bets, what you will sponsor.
- Improvement plan used in good faith. Clear gaps, support, timeline, honest possible outcome.
🤖 AI-era tip: Audit consistency; do not let a model write the rating
An approved tool can help you check that written feedback is specific, behavioral, and consistent in rigor across people — a fairness pass. It must not invent examples you did not observe, and it must not determine a rating or a promotion. You remain accountable for every sentence in a formal record. Keep raw personnel data inside the systems your organization has designated.
Takeaways
- Set expectations in writing and talk all year; surprise at review time is a process failure.
- Do not over-promise to retain someone or to delay the truth.
- Grow people who are doing well; do not only remediate.
- When coaching is no longer honest, judge — early, clearly, and inside fair process.
- Avoidance is paid for by peers and, eventually, by the person you were trying to spare.
Evidence
- Locke, E. A., and Latham, G. P. (2002). Building a practically useful theory of goal setting and task motivation. American Psychologist, 57(9), 705–717. Summary feedback is a moderator: the combination of goals plus feedback is more effective than goals alone.
- Kluger, A. N., and DeNisi, A. (1996). The effects of feedback interventions on performance. Psychological Bulletin, 119(2), 254–284. Feedback is not uniformly helpful; more than one-third of interventions in the meta-analysis decreased performance, especially when attention moved toward the self. A late, formal, character-tinged review is the design this chapter is trying to avoid. (Discussed more fully in Chapter 9.)
- 37signals. The 37signals Manager Playbook: Performance Reviews; Promoting Your Report; Upholding Standards, Underperformance & Terminations; Navigating Difficult Conversations. Cited for the “red pen” test, promotion as performance plus business need (and no promised timeline), a short recovery window, and talking points rather than a script for hard conversations.
- Graham, M. (2026). The Management Pretzel. Practitioner essay: the test is whether you would still make the org change if that person were not in the seat.
- Shook, J. (2010). How to Change a Culture: Lessons From NUMMI. MIT Sloan Management Review, 51(2). See Chapter 1.
- Fisher, M. (2026). Same People. See Chapter 1.