别问：「你身边要是有合适的，帮我介绍一下啊。」——他会答应，而且是真心答应，然后永远想不起来。
要问（画像 + 场景 + 单数）：「你提过你们那批做跨境电商的同行，好几家也在手工对账。他们里面，谁最近抱怨过这件事？不用名单，一个人就行。」
三个零件缺一不可：一、画像具体到能对号入座（「做跨境电商的」，不是「有需要的人」）；二、场景是他亲耳听过的抱怨（那是检索线索不是筛选条件）；三、只要一个——「一个」是可完成的任务，「一些」不是。
他说「我想想」时别停。换个线索再问一次：「上次一起吐槽这事的，都有谁？」名字常常当场就出来了。
拿到名字，立刻把最后一步也做掉：「我写段话发你，你觉得没问题就直接转给他。」
EN What you want isn't "a referral" — it's a name surfacing in their head, and those two things differ by an order of magnitude in difficulty. Agreeing costs nothing; retrieving, with no cue at all, the one person among hundreds who happens to be struggling with exactly this is something human memory is very bad at. So the sincere "sure, I'll keep you in mind" is followed by nothing. Turn the open question into a fill-in-the-blank: give a persona concrete enough to match against, a scene they have personally witnessed, and ask for exactly one person. Not "if you know anyone suitable" but "those cross-border sellers you mentioned — which one of them complained about this recently? One name is enough." Three parts, all required: a persona specific enough to recognise, a scene that works as a retrieval cue rather than a filter, and the singular — one is a completable task, several isn't. If they say "let me think", don't stop there; offer a second cue ("who was in the room when you were all complaining about it?") and the name often arrives on the spot. Then remove the last cost too: write the message for them to forward, and tell them to edit it freely.
机制：记忆是线索驱动的。存在脑子里（可获得）和当场取得出来（可提取）是两回事，中间隔着有没有一个对的线索。你以为在测他的意愿，其实在测他的提取能力——而无线索提取的成功率低到荒谬。
证据强度：两项都是同行评审的经典实验，方向可靠；但都不是在转介绍场景里做的，属于机制迁移。别把「3% → 28%」当成你的转介绍会涨十倍。
第一步，让他自己把成果说出来，别替他总结。问：「跟三个月前比，这块具体差在哪？」——他答出的那句话是之后一切的原材料。你的版本没用，要拿去转述的人是他。
第二步，当场接住：「这句话我能引用吗？」——先要一个最小的许可，再要人。
第三步才是那道填空题：「你刚说那批对账以前要两天，现在半小时。你那圈人里，谁还在过两天那种日子？」
三个触发时刻（都是他刚说完好话时，别另约会）：第一个成果被确认那次通话；他主动来道谢的任何时候——最强的信号，也最常被浪费；他公开说了好话之后的 48 小时内。
别越的线：没拿到成果就别开口。那时你要的不是推荐，是让他替一个没验证的东西作保——他事后会尴尬，而这份尴尬算在你头上。
EN Signing day is when they feel best and hold nothing: all they can say is "I just bought something, looks good", which carries zero information to a friend and arguably negative information — it sounds like vouching for something unverified. What they can pass on is never the product, it's a result they can retell. And that window closes at both ends: too early and they have nothing to say; too late and they no longer remember how bad the old way was, because a solved problem rapidly starts to feel like it never existed. So the moment to ask is right after they have said the result out loud themselves — not the date your calendar reached. Make them phrase it: "compared with three months ago, what's actually different?" Their sentence is the raw material; your version is useless, because the person who has to retell it is them. Then take the smallest permission first — "can I quote that?" — before asking for a person, and only then run the fill-in-the-blank. Three best triggers, all of them moments they have just said something good, none of them a separately booked meeting: the call where the first result gets confirmed; any time they spontaneously thank you, which is the strongest and most wasted signal of all; and within 48 hours of them praising you publicly. One line not to cross: don't ask before they have a result. At that point you aren't asking for a referral, you're asking them to vouch for something unproven, and the embarrassment that follows gets charged to you.
机制：「我会介绍的」是一个意图，而意图和行为之间的落差是系统性的，不是他不诚实。缩小它只有两条路——把行动成本降到接近零（上一节），或趁触发条件正好齐全的那一刻（这一节）。另外，让他亲口说出成果本身就会改变他：说过的版本会反过来加固他自己的判断。
边界：这套逻辑要求真的交付了一个可观察的成果。价值要一年后才显现的生意（保险、咨询），窗口得人为造——定期把基线和现在摆在一起给他看。客户成功那期讲过：有效的东西会隐形。
1. 先把最坏结果承包掉：「聊了要是不合适，我自己会说不合适然后走开，不缠着他。」——这句解决的正是他真正担心的事。
2. 让引荐可撤回：「我写段话你转给他，他没兴趣就当没这回事，你不用替我追。」把一次担保降级成一次转发。
3. 只让他背书体验，别让他背书能力：「你就说你们用着还行、让他自己判断」，比「你跟他说这个特别好」轻得多——后者是替你打包票，前者只是分享经历。
4. 主动闭环：见完面一定告诉他结果，成没成都说。这条最常被跳过，却决定有没有第二次——人不怕帮忙，怕帮完石沉大海，连自己是帮了还是添了乱都不知道。
EN Making one introduction stakes their standing as a good judge in front of a friend. If you deliver, they gain a little; if you blow it, they lose a real relationship — the ledger is inherently asymmetric, so the default answer is to do nothing. Which tells you where to push: cut their cost and risk rather than raise their reward. Most people instinctively add to the reward side with bounties and commissions, which neither offsets the risk nor stays neutral: money quietly converts "doing a friend a favour" into "a piece of business", and people follow completely different rules in those two modes. Four sentences that cut risk. Underwrite the worst case: "if it isn't a fit I'll say so myself and go away, I won't chase him." Make the introduction reversible: send them text to forward, so a guarantee downgrades into a message. Let them vouch for their experience, never for your capability — "just say it's been working for you and let him judge" is far lighter than "tell him it's great." And close the loop, unprompted: tell them how the meeting went, deal or no deal. That last one gets skipped most and decides whether there's ever a second referral — people don't mind helping, they mind helping into silence, never learning whether they helped or created a mess.
机制：引荐发生在社会规范里（人情、互惠、长期关系），一旦标上价格，它就被切换进市场规范（一次性交易、按价钱算值不值）。切换是单向的、而且很难切回来——因为标价同时给了一个可以拒绝的理由：不值这个价。
证据强度：两项都是同行评审的经典研究，机制方向可靠；但「动机挤出」的适用范围学界一直有争论（不同元分析结论不一致）。别当成「给钱一定坏事」的定律，当成一个真实存在、值得防的风险即可。
1. 分数只用来触发，不用来汇报。给 9–10 分的人，48 小时内必须被一个有名字的人问一次——问上面那道填空题，不是「谢谢您的评分」。过了那阵，劲就散了。
2. 真正的信息在后面那个开放题。「最主要的原因是什么？」——他写的那句话就是你的下一段文案、下一个案例标题，也是他以后转述你时会用的话。分数是个位数，那句话是原材料。
3. 别拿它考核团队。一旦分数决定奖惩，你收到的就是被打点过的分数：挑人发问卷、发之前先打招呼。考核什么，就拿到它的表演版。
4. 样本小就别算。三十个回复的净推荐值，抖动几乎全是噪声，还被回应偏差放大——愿意填问卷的人本来就偏两头。客户少时用「这个月发生了几次真实引荐」代替。
EN Net Promoter asks how likely you are to recommend, counts 9–10 as promoters and 0–6 as detractors, and subtracts. Two problems: it measures an intention rather than an act — the 28% gap from the previous section lands exactly here — and it is not better than conventional satisfaction measures at predicting growth, which isn't an opinion but a published replication. So stop running meetings on the number. Demote it to a switch: whoever just scored 9 or 10 is the person to ask, specifically, today. The score has no value; that moment does. Five rules. Use the score only as a trigger — a named person contacts every 9–10 within 48 hours, with the fill-in-the-blank question, not a thank-you. Treat the open follow-up as the real payload: the sentence they write is your next piece of copy, your next case study headline, and the words they'll use to describe you later. Never appraise a team on it, or you'll receive a groomed number. Don't compute it on small samples, where the movement is noise amplified by response bias; count actual introductions instead, which is a behaviour you can verify. And retire "an unhappy customer tells ten people" — it has no reliable source, and the research below points the other way.
机制：单题的意图量表，量的是他在被问那一刻愿意怎么表态。它和收入之间隔着三样东西——意图到行为的缺口、他有没有对具体的人开口、那个人会不会买。任何一环没人负责，分数再高也不变成钱。
诚实补一句：批评净推荐值不等于满意度不重要。它的真问题是被当成了预测器和考核指标，而它只是一个被压缩过头的表态。要预测收入就看行为——续约、扩张、实际发生的引荐次数。
六十分钟，把「多让客户帮我们说说」换成四件能查的事。