「他是哪国人」几乎预测不了他怎么谈——国别的均值差，小于同一个国家里两个人的差

EN Country-level culture is real, and much smaller than the popular story. Hofstede's dimensions came from something like a hundred thousand IBM questionnaires and describe national averages; using them to predict the person sitting across from you is the ecological fallacy, and Hofstede himself kept saying so. The size of the problem is easy to underestimate: Oyserman, Coon and Kemmelmeier's 2002 meta-analysis of 83 studies of individualism and collectivism found the differences smaller and far less consistent than assumed — in several comparisons Japanese and Korean respondents were not more collectivistic than Americans, and within-country variance dwarfed between-country variance throughout. Hall's high- and low-context distinction has the same problem in a sharper form: it is enormously influential and thinly evidenced, and Kittler, Rieg and Franke's 2011 review found the country classifications lack solid empirical grounding. None of this means culture is irrelevant. It means culture is a weak prior, and you should overwrite it fast with three things you can actually watch in the first meeting. Who is in the room — one person or four, who speaks first, who speaks last; the headcount and the speaking order tell you more about the approval chain than any national label. How information is given — ask something concrete like how many hours a month this costs them, and notice whether you get a number or a paragraph of context. And who sets the timetable — a buyer who offers dates unprompted can be worked directly. Those three are behaviour, not guesswork, and they are observable inside one call. Write them into your notes and use them next time instead of a list of etiquette rules.

EN The most reliable finding in cross-cultural negotiation research is unglamorous: intercultural pairs reach lower joint gains than intracultural pairs, and it isn't because anyone is being hostile. Adair, Okumura and Brett (2001, Journal of Applied Psychology) compared American and Japanese negotiators and found they route information differently — Americans share priorities directly through questions and answers, Japanese negotiators convey them indirectly through a sequence of offers, and when the two are paired the information never lands, so value stays on the table. Brett and Okumura (1998, Academy of Management Journal) found the same drop in joint gains for US-Japan dyads. Gunia, Brett, Nandkeolyar and Kamdar (2011, JAP) added the mechanism that generalises: lower baseline trust reduces direct information sharing, which reduces joint gains — which is precisely why the offer channel matters, because it moves information without requiring anyone to disclose anything. So stop trying to force directness and change the probe. Put three equivalent offers on the table at once: twelve months at the lowest unit price paid upfront; six months billed quarterly at 8% more; twelve months billed quarterly at the first price in exchange for a named case study. Whichever they pick tells you what actually binds them — cash, commitment horizon, or budget with a reputational appetite — and they never had to state a preference out loud. Then decode the polite deferral. "We'll discuss it internally" is real only if a name and a date follow it, so supply them: who sees it next, when might feedback come, and if there's a concern is it most likely budget, timing, or replacing what they do today. Offering three options is far easier to answer, and far easier on their face, than asking what their concerns are. And route the sensitive questions — real budget, real decision-maker, internal opposition — through whoever introduced you.

EN A transactional buyer runs three lines: value, price, risk. A relationship-oriented buyer runs a fourth — if I bring this in and it goes wrong, who do I have to explain myself to, and whose favour have I spent. Leave that line unsolved and the first three don't matter. There is a surprisingly specific piece of evidence underneath it. Zhu, Zhang, Fan and Han (2007, NeuroImage) had participants judge trait words about themselves and about their mother: Chinese participants recruited the same medial prefrontal region for both, while Western participants recruited it only for self-judgements, with the mother falling on the "other" side. Treat that carefully — early cultural-neuroscience work, small samples, limited replication, and emphatically not a claim that anyone's brain is built differently; what it supports is that the boundary of "self" can include close others and can shift with context. The practical consequence is that "what's in it for you" is sometimes aimed at the wrong entity. Four moves lower that fourth cost. Take the risk off the individual: a small pilot, written exit conditions, one named person on your side who owns it — what they fear isn't a bad product, it's signing and then having nobody to call. Never let them be proven wrong in front of colleagues: disagreement goes to them privately, the room only hears the conclusion you reached together. Give them credit they can put their name on by writing the one-pager they need for their own internal review — problem, approach, expected result, risks and mitigations — which isn't flattery, it's absorbing their cost of explaining upward. And lean on introductions, whose advantage is amplified here because the introducer prepays part of that guarantee cost. The boundary matters: relationship lowers the cost of trust, it does not substitute for value. A warm relationship on top of a solution that doesn't work only means they'll find it harder to say no — so you learn you lost later than you should have.

EN Localisation is not translating the deck. Two things here have real evidence behind them. Lev-Ari and Keysar (2010, Journal of Experimental Social Psychology) had the same trivia statements spoken by native and non-native speakers and found accented statements were rated less truthful — and the mechanism appears to be processing fluency rather than overt prejudice, since telling participants that accent can bias them reduced the effect for mild accents. Second, Keysar, Hayakawa and An (2012, Psychological Science) found that reasoning in a foreign language weakens framing effects and loss aversion, the so-called foreign-language effect; be honest that later replications are mixed, with framing paradigms holding up better than risk-preference ones and effect sizes smaller than first reported. Put the two together and you get something concrete: your urgency and loss framing arrive discounted, while numbers, live demos and checkable reference customers cross intact. So send the agenda and the key figures 24 hours ahead — trading live listening for reading they can pace themselves is worth more than any speaking technique. Cut sentences to one idea each and strip idioms, sports metaphors, puns and internal acronyms; every two seconds spent decoding "let's ballpark it" is two seconds not spent on your proposal. Have them restate the important terms in their own words — never ask "is that clear?", which has only one possible answer in any culture; ask how they plan to describe this to their colleagues. Wait five seconds after a question, because second-language processing runs a beat or two slower and the sentence you cut off is often the most valuable one in the meeting. And when you're the accented one: written first, slower, key numbers on screen, and stop apologising for your English — repeating that you're sorry is actively cueing the listener to discount what you say.