一个 Yes 的流水线,与触达的公地悲剧 · AI 与销售产业 · 2026-07The assembly line of Yes, and the tragedy of the commons · AI & sales · Jul 2026

成交从来不是一个动作,而是一串小 Yes 的累积。AI 接管了 Yes 之前的一切——找人、打标、起草、跟进、记录——却在最后一个 Yes 面前停下:信任仍然只认人 A deal was never one act — it's a chain of small Yeses. AI took over everything before the Yes — finding, tagging, drafting, following up, logging — then stopped at the final Yes: trust still only recognizes a person

诚实层:好名单上的平庸话术,永远胜过烂名单上的完美话术。杠杆排序＝ICP 与名单质量＞触达量＞话术;购买信号时机＞说服力。销售组织的头号死法,是用虚高的活动量指标(外呼数 / 发信数)掩盖管道质量的腐烂——AI 把活动量做得更大,也把这种死法加速了。The honesty layer: a mediocre pitch to a great list beats a perfect pitch to a bad list. Leverage = ICP & list quality ＞ outreach volume ＞ script; timing of buying signals ＞ persuasion. A sales org's number-one way to die is covering rotten pipeline quality with inflated activity metrics (dials / sends) — AI makes the activity bigger, and the dying faster.

最大危机＝「触达的公地悲剧」:当 AI SDR 把冷邮件的边际成本压到零,所有人同时踩下油门,收件箱这块公地被吃干——冷邮件回复率从 2019 的 8.5% 崩到 2026 的 3.43%。悖论随之而来:真人手写的触达反而成了稀缺的高价值信号,而买卖双方各自架起 AI,「AI 对 AI 的谈判」军备竞赛开幕。The biggest crisis = the «tragedy of the commons of outreach»: when AI SDRs drive cold email's marginal cost to zero, everyone floors it at once and the inbox — the commons — is grazed bare. Reply rates crashed from 8.5% (2019) to 3.43% (2026). Then the paradox: a human, hand-written touch becomes the scarce, high-value signal, while both sides raise their AIs and the «AI-vs-AI negotiation» arms race begins.

主脊是销售漏斗八节点(线索→资格→需求信任→演示→异议→谈判成交→交付→续约),每节点生产一种 Yes,标注传统 vs AI + 替代强度。另含四大张力、AI 已侵入的流水线 vs 攻不下的信任、六块硬骨头、AI SDR 祛魅、中国现场、产品指南(2C 一线销售 / 2B 销售组织)。这是一张批判性行业解剖,不是工具选型。相邻议题见姊妹图:诈骗 / 冷触达黑产(同一条信任流水线)→security、数字人直播带货 / KOL 变现→creator、per-outcome 定价 / SaaS 商业模式→money。 The spine is the eight-node sales funnel (lead → qualify → discovery/trust → demo → objections → negotiate/close → handover → renewal), each producing a type of Yes, tagged traditional vs AI + replacement strength. Plus four tensions, the AI-taken pipeline vs the uncrackable trust, six hard bones, the AI-SDR reckoning, the China scene, and a product guide (2C front-line sellers / 2B sales orgs). A critical dissection, not tool-selection. Adjacent topics on siblings: fraud / cold-outreach underground (the same trust pipeline) → security, digital-human live commerce / KOL monetization → creator, per-outcome pricing / SaaS models → money.

传统节点Traditional

Yes · 名单 · 成交Yes · list · close

AI agent · 已接管流水线AI agent · pipeline taken

触达公地悲剧 · 危机Commons tragedy · crisis

信任最后一公里 · 硬骨头Last mile of trust

8.5%→3.43%

冷邮件平均回复率崩塌(2019→2026,Instantly 数十亿封基准)。AI 把触达边际成本压到零,收件箱这块公地被吃干;底部 <0.5%Average cold-email reply-rate collapse (2019→2026, Instantly, billions of emails). AI drove touch cost to zero and the inbox commons was grazed bare; bottom <0.5%

15–25%

基于购买信号(融资 / 高管变动 / 招聘激增)的个性化触达回复率——通用冷邮件 3.43% 的 5 倍+。同一条杠杆律:名单＞触达量＞话术Reply rate for outreach on a buying signal (funding / exec change / hiring surge) — 5×+ the 3.43% of generic cold email. The same leverage law: list ＞ volume ＞ script

$5.4亿 ARR

Salesforce Agentforce(「数字劳动力」)ARR,同比 +330%(FY2026 Q3);自家客服 38 万+ 交互、84% 无需人工——AI 替掉席位,倒逼定价从 per-seat 转 per-outcomeSalesforce Agentforce ('digital labor') ARR, +330% YoY (FY2026 Q3); its own support handled 380k+ interactions, 84% with no human — AI replaces seats, forcing pricing from per-seat to per-outcome

847→11→1

AI SDR 祛魅:独立测评中,明星公司 11x 的 Alice 发 847 封邮件仅换来 11 个回复、1 场会议;11x 另被曝虚增 ARR、试用流失 70–90%The AI-SDR reckoning: in an independent test, star startup 11x's Alice sent 847 emails for 11 replies and 1 meeting; 11x was also exposed for inflating ARR and 70–90% trial churn

口径警告:本页是批判性行业分析,非工具选型 / 投资建议,厂商自述与独立验证并置。冷邮件回复率主用 3.43%(Instantly,数十亿封),⚠️另有 3.1%(Cleanlist)等基准。AI SDR 效果:厂商自述(Ava 覆盖 80% 流程、HubSpot +65% 线索 / 2× 回复)与独立测评(847→11→1、平均 3.43%)严重背离,图上并置对照。Agentforce ARR / 单量、硅基智能份额 / 营收、中国智能客服 / 数字人市场规模均有口径冲突,已逐条标注;市场规模预测(百亿 / 480.6 亿 / 767.93 亿美元艾媒预测)为预测值非已实现。厂商自述 / 预测打 D 级;财报 / 官方 / 权威调研为 A。连给出业务成果的那一家,分母也是自家客户:某数据情报厂商 2025 年的客户影响报告基于逾 11,000 名收入专业人士的调查,称使用其平台的公司总可服务市场平均扩大 40%、管道规模增加 32%,平均交易额自 7 万美元起提升,会议数、连接率、赢单率与销售周期均有改善——⚠️ 这是自家客户的自报调查,愿意回答与不回答的客户之间存在选择偏差,且「使用平台的公司改善了」不等于「平台造成了改善」。可核实的那一侧是形态事实:另一家 2025 年把 AI 助手、预测性成交评分、引导式动作与会议助手嵌进销售 / 客户成功 / 服务台三个工作空间——形态是事实,效果是承诺。每张卡片右上角 A/B/C/D＝证据强度;产品指南非推荐背书。 Basis warning: a critical industry analysis, not tool-selection / investment advice, placing vendor claims beside independent checks. Cold-email reply rate uses 3.43% (Instantly, billions of emails); ⚠️other benchmarks include 3.1% (Cleanlist). AI-SDR results: vendor claims (Ava covers 80% of outreach, HubSpot +65% leads / 2× replies) diverge sharply from independent tests (847→11→1, 3.43% average) — shown side by side. Agentforce ARR/units, SiliconIntelligence share/revenue, and China smart-CS / digital-human market sizes carry basis conflicts, flagged individually; market-size forecasts (¥10B / ¥48B / $76.8B, iiMedia) are forecasts, not realized. Vendor claims/forecasts are grade D; filings/official/authoritative surveys are A. Even the one vendor that publishes business outcomes computes them on its own customers: a data-intelligence firm 2025 customer-impact report, based on a survey of over 11,000 revenue professionals, reports customer companies seeing total addressable market up about 40% and pipeline up 32%, average deal size rising from a $70k base, plus gains in meetings, connect rates, win rates and cycle time — ⚠️ a self-reported survey of its own customers, where respondents differ systematically from non-respondents and «companies using the platform improved» is not «the platform caused the improvement». The checkable side is form: another vendor embedded an AI assistant, predictive deal scoring, guided actions and a meeting assistant into its sales, customer-success and help-desk workspaces in 2025 — form is a fact, effect is a promise. Each card's top-right A/B/C/D = evidence strength; the product guide is not an endorsement.

◆ 诚实层 · 四大张力The honesty layer · four tensions

触达的公地悲剧:无限供给,杀死了触达The commons tragedy: infinite supply killed the touch

先看清一件事:当任何人都能用大模型十分钟发出成千上万封「个性化」邮件,「个性化」这个词就失去了含义,买方的收件箱被彻底淹没。无限供给直接制造了「触达的公地悲剧」——这是全图最强的一条张力,也是一切 AI 销售乐观叙事必须先面对的诚实反面。See one thing clearly first: when anyone can fire off thousands of «personalized» emails in ten minutes with an LLM, the word «personalized» loses its meaning and the buyer's inbox drowns. Infinite supply manufactured the «tragedy of the commons of outreach» — the map's strongest tension, and the honest flip side every AI-sales optimism must face first.

▚ 冷邮件平均回复率(Instantly 数十亿封基准)· 崩塌 vs 幸存Average cold-email reply rate (Instantly, billions) · crash vs survivor

8.5%

2019 冷邮件2019 cold email

20252025

3.43%

2026 底部 <0.5%2026 bottom <0.5%

15–25%

购买信号触达(幸存者·5 倍+)signal-based touch (survivor · 5×+)

买方反制(军备竞赛的另一侧):Google/Yahoo 强制 DMARC + 垃圾投诉率 0.3% 红线 / 超 0.5% 全面封杀;买方部署 AI 过滤与 AI 采购代理,AI 对 AI 的谈判开幕(Forrester 预测 2026 年 20% B2B 卖家将进行 agent 主导的报价谈判)。军备竞赛的净结果耐人寻味:卖方用 AI 压低触达成本,买方用 AI 抬高筛选门槛,最后胜出的反而是更像真人、时机更对、上下文更深的那一次触达——机器越多,人味越贵。The buyer's counter (the other side of the arms race): Google/Yahoo enforce DMARC + a spam-complaint 0.3% red line / over 0.5% = full block; buyers deploy AI filters and procurement bots, opening the AI-vs-AI negotiation (Forrester forecasts 20% of B2B sellers will run agent-led price negotiations in 2026). The net result is telling: sellers cut touch cost with AI, buyers raise the filter with AI, and what wins is the touch that feels most human, lands at the right moment, carries the deepest context — the more machines, the pricier the human note.

▸ 张力① 话术神话 vs 名单现实Tension 1 · pitch myth vs list reality

话术是放大器,不是发动机The pitch is an amplifier, not the engine

数据把培训产业的神话拆穿了:Belkins 分析 1650 万封邮件,50 封以内小而精名单回复率 5.8%,500+ 大名单仅 2.1%;6sense 的更狠——任一时刻 ICP 中只有 5–10% 账户真正准备购买。也就是说,你的大多数触达注定发给了此刻不买的人,再完美的话术也改变不了这一点。好名单上的平庸话术,永远胜过烂名单上的完美话术。The data dismantles the training industry's myth: Belkins analyzed 16.5M emails — a tight list under 50 replies at 5.8%; a 500+ list at just 2.1%; 6sense goes further — at any moment only 5–10% of ICP accounts are truly in-market. Which means most of your touches were always destined for people not buying right now, and no perfect script changes that. A mediocre pitch to a great list beats a perfect pitch to a bad list.

▸ 张力② CRM 虚构文学 vs 对话智能Tension 2 · CRM fiction vs the gaze

CRM 是「企业里最大的虚构文学」The CRM is 'the biggest work of fiction'

数字触目惊心:68% 销售说 CRM 录入最耗时、仅约 2% 真正信任数据、约 79% 商机数据从未进 CRM;Validity 调研(1241 人)更直白——75% 承认会捏造数据、82% 被要求「找数据支撑某个说法」。Gong 那句判词值得抄下来:「关于 deal 的真相活在对话里,不在 CRM 字段里」。AI 在这里的第一功不是修复销售管理,而是把旧底座的失真晒到了光下。The numbers are stark: 68% call CRM entry their most time-consuming task, only ~2% truly trust the data, ~79% of deal data never enters the CRM; a Validity survey (1,241) is blunter — 75% admit fabricating data, 82% are asked to 'find data to back a claim.' Gong's verdict is worth copying out: 'the truth about a deal lives in the conversation, not in the CRM field.' AI's first achievement here wasn't fixing sales management — it was dragging the old foundation's distortion into the light.

▸ 张力④ 定价范式 · per-seat → per-outcomeTension 4 · pricing · per-seat → per-outcome

最有意思的反噬发生在卖工具的人身上:AI agent 替代人力席位,吃掉的正是 per-seat 收入,定价被迫从「按人头」转向「按结果」。Salesforce 18 个月换了三套($2/对话 → Flex Credits $0.10/action → $125/user/月)再推 pay-per-resolution(按解决计费);HubSpot Breeze Customer Agent 从 $1.00/对话降为 $0.50/已解决对话。反噬自证:Salesforce 自家客服用 Agentforce 处理 38 万+ 交互、84% 无需人工——AI 不只是功能插件,它在改写软件行业自己的计费逻辑。商业模式深挖见 money。The most interesting cannibalization hits the toolmakers themselves: AI agents replace seats, eating exactly the per-seat revenue, forcing pricing from 'per head' to 'per outcome.' Salesforce ran three schemes in 18 months ($2/conversation → Flex Credits $0.10/action → $125/user/mo) then pay-per-resolution; HubSpot's Breeze Customer Agent went from $1.00/conversation to $0.50/resolved conversation. Self-proof: Salesforce's own support handled 380k+ interactions, 84% with no human — AI isn't a feature plug-in; it's rewriting how the software industry bills itself. The business-model deep-dive → money.

◆ 判断层杠杆排序 · 结论卡The leverage stack · the conclusion

ICP 与名单质量ICP & list quality ＞触达量volume ＞话术script

购买信号时机timing of buying signals ＞说服力persuasion

头号死法:用活动量指标(外呼数 / 发信数)掩盖管道质量腐烂。AI SDR 若建在脏 CRM 与模糊 ICP 上,只会「以机器速度制造垃圾」——80–90% 的成功来自修管道(路由 / 打分 / 数据卫生),不来自买工具。先修管道,再谈 AI。Number-one way to die: covering rotten pipeline quality with activity metrics (dials / sends). An AI SDR on a dirty CRM and fuzzy ICP just 'manufactures garbage at machine speed' — 80–90% of success comes from fixing the pipeline (routing / scoring / data hygiene), not from buying tools. Fix the pipeline first; then talk AI.

从这张图看到的五条规律Five patterns this map makes visible