1. 先定义「第一个成果」：一句话、可观察、由客户说了算。「老李这周不用再手工导那张表」是成果；「完成培训」不是。签约会上就问一句：三十天后什么事发生了，你会觉得这钱花对了？把他的原话抄下来，那就是验收标准。
2. 只配置那一件事需要的东西。全功能培训是上线的头号杀手——信息量把人劝退，而他真正要的可能只有两个按钮。剩下的两个月后再说。
3. 约时间，不约意愿。别问「下周开始用吧？」，问「周二上午九点你处理那批订单的时候，我们一起过一遍，行吗？」——把动作绑到具体的时间、地点、触发场景上。这一句改动有很硬的证据，见下。
4. 头两周的联系密度，比后面十个月加起来都重要。宁可显得烦，也别显得「已经交付完了」。
EN Going live is not opening an account or finishing the training — it is getting them one visible win, in their own words, inside the first two weeks. Not "I know how to use it" but "that annoying thing genuinely stopped happening this week." Two weeks is not arbitrary: motivation, attention and internal backing all peak on signing day and decay from there, and once they get pulled back into their old routine your product stays permanently at "haven't really started yet" — which shows up in month nine as a cancellation. So the number the delivery side should watch is time-to-first-value, not onboarding completion. Four rules. Define the first win as one observable sentence dictated by the customer, and write down their exact words at the kickoff. Configure only what that one win needs; full-feature training is the single biggest killer of adoption. Book a time, never an intention: not "shall we start next week" but "Tuesday at nine, when you run that batch, let's walk through it together" — tie the action to a time, a place and a trigger. And make the first fortnight the densest contact of the whole relationship; better to feel slightly pushy than to feel finished.
机制：签约当天动机最高，此后单调衰减。能不能在动机耗尽之前把那个动作重复成日常，决定了这个客户之后是靠意志用你、还是靠习惯用你。靠意志的最终都会流失，因为意志一定会被别的事抢走。
边界：不是所有生意都有「上线」。一次性交易（卖二手手机、单次服务）没有这个窗口，对应动作是交付后 48 小时内主动确认一次。另外这条只在这单本来就该卖时成立（资格判定那期讲的）——否则再漂亮的上线，也只是把发现问题的时间往后拖半年。
1. 你的冠军换岗或离职。最早也最致命：你的价值住在他脑子里，人一走，你在这家公司就变回一张没人认识的账单。所以从第一天起就要认识第二个人。
2. 使用面变窄：从五个人在用变成一个人在用。总用量可能没跌，但它已经变成一个人的私人工具——那个人一休假，你就消失了。
3. 你的会议开始被改期、被降级（原来是负责人来，现在是助理来）。这是优先级的公开读数，比任何问卷都准。
4. 突然不再提需求、不再抱怨。别读成「终于稳定了」。
5. 话里出现新词：合规、评估、比价、统一采购。出现时基本已经晚了，只剩体面收场。
沉默规矩：任何账户连续 30 天没有一次实质互动（自动邮件不算），就标红，由人主动约一次。别等系统提醒你——系统提醒的时候，信号早就变成结果了。
EN A customer who complains is still talking to you, which means they still intend to stay. The dangerous one says "all good" every time and simply doesn't renew. Satisfaction is a costless score about the past; renewal is a spending decision about the future, and between them sit the alternatives they now know about and the cost of switching — that gap is where every "but they said they were happy" comes from. So stop chasing sentiment and watch five behaviours, all of which show up months before a cancellation notice. Your champion changes role or leaves — the earliest and deadliest, because your value lives in that person's head and once they go you are an invoice nobody recognises. Usage narrows from five people to one, so total volume holds but the product has become one person's private tool. Your meetings start getting rescheduled and the attendee gets more junior. They abruptly stop asking for things and stop complaining. New vocabulary appears: procurement, evaluation, benchmarking. And one standing rule: any account with thirty days without a real human interaction gets flagged and called by a person — don't wait for the system, because by the time a dashboard notices, the signal has already become the outcome.
机制：情绪指标记录的是他愿意怎么说，行为指标记录的是他实际把注意力放在哪。注意力比嘴诚实得多，而且它变化在先——人是先不用，才不满，最后才取消。
边界：信号是用来触发一次对话的，不是用来自动降级客户的。看到信号要打电话，别看到信号就群发一封挽留邮件——那反而暴露了你只在数据变差时才想起他。
第一行：你当初说的问题是什么——用他签约时的原话，不要转述。
第二行：当时的基线数 vs 现在的数。保守、精确，宁可写「每周省 3.5 小时」也别写「大幅提升效率」——保守精确的数比大数可信，这在价值销售那期展开过。
第三行：这半年最关键的一次（那次事故、那次赶工期）。这是「峰」，必须点名，因为客户记住的就是它。
第四行：下半年要一起做成的一件事。让这份东西朝前看，而不是一次讨账。
开场一句话：「这半年我们一起把 X 从 A 做到了 B；下半年我想推的是 C，你看对不对？」
EN When your product actually works, the pain stops happening — and once it stops happening, nobody remembers how much it hurt. Only the invoice keeps showing up on time. That is not ingratitude, it is how people work; you don't thank your own front door lock every morning. There is exactly one defence and it has to happen on day one: at signing, write down how much it currently hurts as a number — hours per week, incidents last month, orders mis-shipped in this batch. Without that number you can prove nothing six months later and are reduced to "we've been supporting you throughout", which scores zero in a renewal meeting. Then, sixty days before renewal (not seven), send one page: their original words about the problem, the baseline number against today's number, the single most critical moment of the period, and one thing to achieve together next. Keep the numbers conservative and precise — "3.5 hours a week" beats "a substantial efficiency gain" — and always name the critical moment out loud, because that peak is what they actually remember of you.
机制：记忆不是录像，是几个片段的加权和。人评价一段长期关系时不做积分，只取最强烈的那一刻和最后那一刻。你稳定的日常交付几乎不进入这个计算——这就是「我们服务一直很稳」在续约会上等于没说的原因。
边界：价值复盘的前提是真有价值。数字拿不出手时，正确做法是先承认、再谈怎么补，而不是把几个模糊的正面词包装成一页 PPT——客户手里有他自己的账，粉饰会一次性烧掉你全部可信度。另外别把「制造峰值」理解成故意先让事情变糟再救火：那是操纵，而且客户复盘时看得出来。
（就在他刚确认成果的那次通话里说，别另约会——另约就变成了一次销售会议）
1. 承接他刚说的话：「这次能这么快，主要是因为那批数据是提前理好的，对吧？」
2. 指出同一台机器的下一个瓶颈：「那按同样的逻辑，下个季度卡住的会是审批那一段——现在要三天。」
3. 给一个小的、可退的下一步：「有个客户就是这么干的，我先给你看十分钟，你觉得不值就当我没提。」
不许越的线：客户还没从已经买的东西里拿到价值时，绝不卖第二件。那不是扩张，是在一个尚未验证的假设上加注；他之后一定会发现，而且会连第一件一起退掉。判断标准很简单——上一件东西，他能不能自己说出它带来了什么。说不出来，就先回去做第三节那件事。
EN The same customer paid a risk premium the first time, for the question of whether you could deliver at all. Once one result has been confirmed, that premium is gone: they hold the evidence and the cost of evaluating you has fallen to nearly zero. So the difficulty of expansion was never the customer — it is the timing, and most people put it in the worst possible slot: the renewal negotiation, where you are structurally opposed (they want the price down, you want it held) and anything new you raise reads as a way to charge more. The two weeks right after a confirmed win is the window where you are still on the same side. Say it on that same call, don't book a separate meeting: pick up their own sentence about why it went well, name the next bottleneck in the same machine, then offer one small reversible step — ten minutes, and if it isn't worth it, forget I mentioned it. One line you never cross: never sell the second thing before they have got value out of the first. That isn't expansion, it's raising the stake on an unverified assumption, and when they notice they will return both. The test is simple — can they say, unprompted, what the first purchase gave them?
机制：第一次购买面对的是未知风险，第二次面对的是已知风险，而人对这两者的定价完全不同。叠加现状偏好——续着的东西不需要理由，换掉才需要理由——你天然占着位置。但这个便宜只在他还认为你有用的时候才成立，一旦无用，同一股惯性会变成年底的一次性清算。
边界：不是每个客户都该扩张。有些客户已经用到上限，再加只会让他的投入产出变差、下一次续约更难谈。对这种客户正确的动作是要转介绍，不是要加单——把他的满意换成新客户，而不是换成他的预算。
九十分钟，把「维护关系」换成四件能查的事。