「名单再长一点」是找客户里唯一确定无效的努力

EN "Mid-sized, growing fast, cares about efficiency, has budget" — the problem with a profile like that isn't that it's wrong, it's that it excludes nobody. A trait is only worth writing down if the share of your won deals that have it differs sharply from the share of your lost deals that have it; a trait that runs equally high on both sides carries zero information and is pure self-consolation. So a profile can't be reasoned outward from your product. It has to be subtracted from a comparison: put last year's wins in one column and last year's dead deals in the other, count both proportions for each candidate trait, and delete every trait where the gap is under about twenty points. What survives becomes a crude scorecard, one point per trait — and you work only the top band. Use the scorecard rather than your instinct, and give every profile the same acceptance test: adding this criterion, does my list of three thousand shrink? If it's still three thousand, the criterion is decoration.

EN What you're short of at a cold start isn't names — it's a single concrete sentence you can say about these people. Sending three hundred emails at that point just mass-prints a hypothesis you haven't found the language for yet. Do thirty real conversations first, twenty minutes each, with exactly one job: copy down the words they use when they describe the problem. Ask for them honestly — "I'm working on something around X, I'm not selling anything, I'd like to hear how you handle it today" — and then genuinely don't sell, because selling closes that door permanently. Ask about the past rather than the future, write nouns and verbs down verbatim, and resist the urge to translate anything into your own vocabulary; the moment you translate, the conversation's entire value evaporates. Any phrase that shows up three times or more goes into a pool, and the most frequent one becomes your opening line, unedited. The test is simple: can your first sentence be found word for word in your notes? If not, you're still talking to yourself.

EN Most people pick a channel by how many people it reaches, but that number is large everywhere, so it can't function as a choice. The unit that decides it is your minutes divided by the real conversations you get back: a channel producing three twenty-minute conversations beats one reaching thirty thousand people. Structurally this is an explore-exploit problem — each channel is a slot machine whose true payout you don't know — and the optimal shape is always the same: spread small equal bets first, then concentrate fast. So run four candidates at an identical small budget for two weeks, score each on one number only (conversations, never opens or likes), cut the worst two in week three, and double down on the survivors. Retest a cut channel with a small bet six months later, because channel economics drift. And track one more thing: whether you can be recognized there. Channels where you have work, shared people, or an identity get cheaper over time; a purely cold channel costs the same on the thousandth attempt as on the first.

EN "Outreach is a numbers game" is true, and almost everyone reads it backwards. It doesn't mean send more; it means that at low reply rates, differences inside a small sample are noise. Fifty emails yielding one reply against fifty yielding three looks like a threefold win, but the two confidence intervals sit almost entirely on top of each other, and telling two percent from four percent reliably takes upwards of a thousand sends per arm. The practical consequence is that polishing wording is unfalsifiable at your sample size, so bet on step changes instead: a different audience, a different concrete fact about the person in your opening line, a different channel — wording ranks last. Fix your batches at thirty or more, hold the audience and the sentence constant until the batch is done, put only absolute numbers on the wall (sent this week, real conversations this week), and commit to a volume — say a hundred attempts — before you're allowed to draw any conclusion at all. Otherwise what you kill is the unlucky batch, not the bad one.