1. 拿最近成的三单做时间账：每单在四段上各花了多少小时。找人那段多半近乎空白——不是不需要，是一直在被躲。
2. 先切最痛的那一刀：第一刀几乎都切在「找人 / 成交」之间，这两段节奏最冲突。切完看一个月：成交那位的会面数涨没涨？没涨说明瓶颈不在找人。
3. 把交接写成一张纸：三个必填加一句「不满足就退回」。允许退回是关键——不能退回的交接，下游只会私下把上游的活重做一遍。
交接时的一句话：「这单交给你，他原话是『每月底对账要花两天』，预算没问过，拍板的是财务总监，这三条你先确认再排会。」
EN A deal is at least four different jobs: sourcing (screening a crowd for people who might need this — volume and discipline), qualifying (deciding whether it's worth your hours — the nerve to say no), closing (diagnosis, value, objections, negotiation — experience and presence) and keeping (onboarding, renewal, expansion — patience). The skills, tempo and temperament barely overlap. Force all four onto one person and the outcome is always the same: they quietly drop the most uncomfortable stretch, usually sourcing. So the first decision isn't how many people to hire, it's who does each of the four today, where the handoffs sit, and what travels across them. Three handoffs, each with a required payload: sourcing to qualifying carries his own words, not your summary; qualifying to closing carries budget, timing and decision-maker, each marked known or unknown; closing to keeping carries why he bought and what you promised. Make handoffs rejectable — a handoff nobody can send back gets silently redone downstream. And don't split at all while you're under five people, or while a deal takes months of customisation, or while you personally haven't run the whole thing end to end.
机制：分工的收益来自两处——重复带来的熟练，以及不必在不同任务间反复切换心智；代价是每个切口新增的交接损耗与信息丢失。所以不是越细越好，净收益取决于单量：量小时交接损耗大于熟练收益。
诚实边界：上面两项实验的任务以毫秒计，把它换算成「一天换三种活损失百分之多少」是外推，不是研究结论——它支持「切换有代价」，支持不了任何具体数字。另外，分工会让每个人的视野变窄：找人的那位久不到成交现场，判断会慢慢失准，所以定期让他旁听一次真实会面，不是福利，是校准。
给材料（提前发）：你们真实产品的一页介绍 + 一个真实客户的公开信息（网站、招聘页、新闻）。
任务一（15 分钟，现场）：「我是这家公司的运营负责人，你有十分钟。开始吧。」——你只演一个普通难缠的人：忙、有点怀疑、不主动给信息。看什么：他问了几个问题？问的是他想卖的点，还是我的处境？我含糊其辞时他是继续讲，还是追一句。
任务二（15 分钟）：「刚才哪里没做好，重来会怎么改？」看什么：能不能具体复盘。会复盘的人能被流程放大，只会归因于「客户不对」的人不能。
任务三（15 分钟）：给他一个真实异议（「我们已经在用别家了」），让他写三行回应。看什么：先认同再回应，还是直接对撞。
打分表（每项 1–5，所有人同一张）：提问质量 / 听完再说 / 复盘能力 / 抗含糊 / 表达清晰。先各自独立打分再讨论——先讨论会让第一个开口的人污染全场。
EN The most expensive misjudgement in building a sales team usually happens the moment you think "we really clicked." A free-flowing interview measures whether he is likeable to you — which is the single skill every salesperson has practised most, so you have chosen the most easily faked signal available on which to bet a hire. Only two things replace it: the same questions for every candidate scored on the same scale, and watching him do a piece of the real work. Run a forty-five minute sample: ten minutes selling to you while you play a busy, mildly sceptical buyer who volunteers nothing — count how many questions he asks, and whether they are about his pitch or your situation, and whether he follows up when you go vague; then fifteen minutes on what he'd do differently, because a person who can review himself concretely is one a process can amplify and a person who blames the customer is not; then a written answer to one real objection. Score independently before anyone speaks, or the first voice contaminates the room. And drop extraversion from your filter — it isn't the thing you think it is.
机制：选拔的准确度取决于信号里有多少和工作真相关的东西。自由聊天把标准交给临场好感与相似性，等于往判断里灌噪音；固定问题固定尺子，是把人放在同一条基线上比；工作样本更直接——用行为预测行为，中间少一层翻译。
边界与红线：结构化不等于死板，追问和临场跟进都可以，固定的是问题清单和评分维度，不是对话。工作样本的前提是任务真像那份工作，别拿脑筋急转弯充数。还有一条硬红线：试岗不能变成免费劳动——超过一两小时、或你打算真拿去用，就付钱。
1. 这个指标怎么作弊最省力？按成交额 → 送折扣；按新签 → 不管续费；按活动量 → 刷记录。想清楚后，要么换指标，要么把作弊路径直接堵上（最常用的一招：折扣从他的提成基数里扣，让让利变成他自己的钱）。
2. 数不清但重要的事，谁在管？交接质量、说真话、帮同事——这些永远算不进提成。可行的只有三种：把单一指标的强度调低（别搞纯提成）、把一部分挂到团队或续费上、把兑现延后。再加一条提成来补，通常只会催生新的作弊面。
3. 阶梯要不要？阶梯能拉动冲刺，代价就是上图那个形状。真要用就把周期拉长、门槛设在他跳一跳够得着的地方；差太远的人会放弃本季度、把单藏到下季度。
4. 比例怎么定？没有普适数字，只有推法：算一单毛利，定销售成本能占毛利多少（由你的单位经济学定，不是同行定），乘以预期单量，就是一个人的总薪包上限，再切成底薪和提成。业内常见做法是两者大致对半（可查的惯例，不是研究结论）；单越简单、越靠个人，提成比重可以越高。
EN Commission is never about how much, it's about what you pay on. People move their hours toward whatever is both measurable and convertible into money, and the unmeasurable dimensions don't merely shrink — they get systematically crowded out. Pay only on new logos and nobody handles handoffs or renewals. Pay only on booked revenue and discounts get handed out freely, because a discount barely touches the base your rep is paid on. Pay on activity and you'll receive a pile of call logs and zero customers. So before you design anything, ask the one useful question: what is the laziest way to game this number? Whatever that is will happen — not because people are bad, but because you are paying for it. Three fixes actually work for the things you can't count: lower the intensity of the single metric, tie part of the payout to the team or to renewals, and delay vesting. Adding one more commission line to cover the gap mostly just creates a new surface to game. And charge discounts against the rep's own commission base, so giving margin away costs him his money, not only yours.
机制：当一个人要同时干好几件事、而其中只有一部分能被准确计量时，对可计量部分给强激励，会把努力从不可计量的部分抽走。这不是态度问题，是理性反应——他在按你给的价格表分配时间。
边界：提成解决「愿不愿意」，解决不了「会不会」——转化率低通常是能力和流程问题，这时加提成只会让好的人先失望离开。对个人也一样：只奖励能数的指标（发了几条、见了几个人），那些数不清却真正复利的事就会被你自己挤掉。
1. 字段只留四个，其余全部选填。每多一个必填，数据质量掉一截；报表字段你自己从这四个算，别摊派给一线。
2. 「下一步」必须写成动作 + 日期，不能是状态。「保持跟进」「等客户回复」都不算；「11 月 6 日前把对账那段做成 20 分钟演示发给小张」才算。
3. 永远别用 CRM 里的活动量做考核。一旦打几个电话变成考核项，你就把上一节的挤出效应搬进了系统，从此数据漂亮又无用。CRM 该是销售自己的工作台：替他记住该做什么，而不是替你盯着他做了什么。
周会上的一句话：「你这三单的下一步日期都过了，我们只聊这三单——是卡住了，还是该关掉？」
EN Most CRMs become a filing punishment for exactly one reason: the fields were designed for management's report, not for the seller's next action. The honest test is whether someone can take a colleague's deals from the CRM alone while he's away for two weeks. If not, what you're storing is a record, not a process. So make only four fields mandatory: what the next action is, who does it, the date it's done by, and which specific obstacle the deal is stuck on. The first three turn a vague "keep following up" into a plan with a trigger, and the fourth forces people to name the blocker instead of parking the deal in the comfort of "in progress" forever. Everything else stays optional — each extra required field costs you data quality, and reports should be computed from those four rather than levied on the front line. One rule matters more than the software: never make CRM activity counts a performance metric. The moment calls logged becomes something people are paid or judged on, you have imported the crowding-out problem straight into your system, and from then on the data is beautiful and useless.
机制：意图变成行动靠的不是决心，是把动作挂到具体的时间和情境上——到了那刻不必重新决定。同时，把只长在个人脑中的隐性流程写成可交接的东西，团队才可能不依赖单个人；但外化本身没有魔力，起作用的是人真的照它改变行为。
边界：两三个人时 CRM 的收益主要是防遗忘，一个共享表格够用；真正该上系统的信号只有两个——交接开始出现、预测开始重要。上太早，你会花几个月配置一个没数据的系统。另外，清单管的是「别漏」，管不了「该不该做这单」——那是资格判定那件事，别指望字段替你判断。
九十分钟，拿你现在的团队（哪怕只有你）走一遍。