咨询产业 AI 全景图 · 一次判断的出租

一次判断的出租Renting Judgment

AI 先吃掉的不是判断,是判断的「制作过程」 · 咨询产业 AI 全景图 · 2026AI ate not judgment first, but its «manufacture» · the consulting-industry AI atlas · 2026

AI 冲击咨询的顺序出人意料:它先吃掉的不是「判断」,是判断的「制作过程」。桌面研究、访谈纪要、建模、做 deck——养活整个金字塔底座的流水线正在被软件化:头部机构自己承认「还需要一群群分析师做 PPT 吗?不需要」,入门级招聘一年暴跌 54%,应届起薪三年冻结。客户第一次能用 AI 生成「大牌级」的分析框架,199 元/月的 AI 战略咨询代理把原本 1% 大企业才买得起的东西卖给所有人;卖 PDF 信息差的分析师机构最先被资本市场重定价——市值自高点回撤逾 52%。但当报告可以民主化,为报告署名背锅的责任无法外包给模型:政府合同因 AI 编造引用而退款、45 条引注仅 5 条准确的报告被撤回。于是咨询没有消失,而是被拦腰切开——劳动密集的下半身被软件吃掉,信用密集的上半身反而更贵。一次判断力的出租,正在变成一次信用的出租。The order of AI’s impact on consulting is unexpected: it ate not «judgment» but judgment’s «manufacture». Desk research, interview notes, modelling, building the deck — the assembly line that feeds the whole pyramid base is being softwareised: a top firm itself admits «do we still need armies of analysts making PPTs? No», entry-level hiring plunged 54% in a year, and graduate starting pay is frozen for a third year. Clients can, for the first time, generate a «top-tier» analytical framework with AI, and a ¥199-a-month AI strategy agent sells to everyone what only the top 1% of firms could afford; the analyst houses selling «PDF information asymmetry» were repriced first — market cap down over 52% from its peak. But once a report can be democratised, the responsibility of signing for it cannot be outsourced to a model: a government contract was refunded over AI-fabricated citations, a report with only 5 of 45 accurate references was withdrawn. So consulting did not vanish; it was cut in half — the labour-intensive lower body eaten by software, the credit-intensive upper body only dearer. The rental of judgment is becoming the rental of credit.

咨询无执照、无法定责任——它靠三样东西兜底:Consulting has no licence, no statutory liability — it rests on three things: 声誉 · 合同 · 署名 — reputation · contract · a signature。这三样,AI 恰好都没有。AI has none of the three.

✓ 本图相信This map believes判断的「制作过程」被打穿是实测不是预言(758 人实验 + 95% 月度使用 + 入门招聘 -54%);头部机构「左手革命」已写进财务报表(AI 业务占某巨头营收 25%);纯信息中介最先被资本市场重定价(Gartner 市值回撤逾 52%、Forrester 退出战略咨询);「谁署名谁背锅」不可外包给模型——两起幻觉事故是行业判例;买方结果导向化真实发生(86% 买方要 AI 资产、「卖配方→卖蛋糕」);专有数据与专用工具强于通用能力(返工率 29% vs 49%)。Judgment’s «manufacture» being pierced is measured, not predicted (a 758-consultant trial + 95% monthly use + entry hiring −54%); the top firms’ «self-revolution» is already on the income statement (AI is 25% of one giant’s revenue); pure information intermediaries were repriced first (Gartner −52% from peak, Forrester exiting strategy consulting); «who signs, who takes the fall» cannot be outsourced to a model — two hallucination incidents are the industry’s case law; the buyer shift to outcomes is real (86% of buyers want AI assets, «selling the cake, not the recipe»); proprietary data and purpose-built tools beat general capability (rework 29% vs 49%).

✗ 本图不相信This map does not believe不信「咨询已死」——冲击是重构价值链不是消灭,「AI 终结咨询未发生」与 Gartner 腰斩同时为真;不信「AI 马上接管判断」——射程外正确率反降约两成、复杂决策依赖 AI 更糟;不信裁员数字=AI 直接替代——需求疲软与 AI 提效叠加,归因要拆;不信厂商效率倍数——「15 分钟出战略」「一半成本 10 倍洞察」皆自述;不信「技术可自动化 60-70%」等于「今天已节省 60-70%」——实测多数节省不足 25%;不信任何绕开诚实层的叙事——新人断层、起薪冻结、样本污染、幻觉反噬,AI 每一处都在放大。Not «consulting is dead» — the shock reconstructs the value chain, not erases it; «the end of consulting hasn’t happened» and Gartner halving are both true. Not «AI takes over judgment now» — out of range, accuracy falls ~20 points and reliance worsens complex decisions. Not that layoff figures = direct AI replacement — soft demand and AI efficiency overlap; attribution must be split. Not vendor efficiency multiples — «strategy in 15 minutes», «half the cost, 10× insight» are self-reported. Not that «60–70% technically automatable» = «60–70% saved today» — most measured savings are under 25%. Not any narrative that skips the honesty layer — the missing rung, frozen pay, sample pollution, hallucination blowback, all amplified by AI.

ATHE LIFECYCLE · 主脊 · 一次判断出租的一生The spine

咨询项目生命周期九节点:中段被吃,两端更贵The nine-node project lifecycle: the middle eaten, the two ends dearer

最小单元是一次「判断力的出租」。每个项目走一遍生命周期:问题定义 → 售前 → 提案 → 调研访谈 → 建模 → 报告 → 汇报 → 实施 → 续约。越靠近「信息搬运 + 文档产出」的中段(④⑤⑥)冲击越猛;越靠近两端的问题定义与汇报/实施/续约,AI 越难独立完成——判断的生产被压缩,判断的采纳未被触及。⑧实施是唯一「AI 反而做大蛋糕」的节点。每节点标承接节奏:人天层 / 混合层 / 信用层;渗透 % 为编者评估值。The smallest unit is a «rental of judgment». Every project walks the lifecycle: problem definition → pre-sales → proposal → research/interviews → modelling → report → presentation → implementation → renewal. The «information-hauling + document-output» middle (④⑤⑥) is hit hardest; the two ends — problem definition and presentation/implementation/renewal — AI can least do alone: the production of judgment is compressed, its adoption untouched. Node ⑧ implementation is the only one AI enlarges. Each node is tagged by tempo: grunt / hybrid / credit; penetration % are editorial estimates.

BTWO PARADOXES · 两个悖论与一道锯齿Two paradoxes and a jagged edge

左手革命、信任反噬,与 AI 的锯齿边界The self-revolution, the trust blowback, and AI’s jagged frontier

中篇是全图最锋利的三张牌:巨头为何一边卖 AI 一边拆自己、报告民主化为何撞上幻觉反噬、AI 的能力边界为何反而抬高了「人来把关」的价格。The middle holds the map’s three sharpest cards: why the giants sell AI while dismantling themselves, why democratised reports hit hallucination blowback, and why AI’s capability frontier raised the price of «a human gatekeeper».

CHARD BONES · 硬骨头Hard bones

AI 攻不进的八块:报告可以免费,签名依然昂贵Eight things AI cannot breach: a report can be free, the signature stays expensive

利润与地位正向这八堵墙内迁移:能签名担责、能进董事会、能拿到一手事实、能推动变革的位置,最后失守。咨询无执照,全靠声誉 + 合同 + 署名的私法结构兜底——而 AI 恰好三样都没有。Profit and status migrate behind these eight walls: the positions that can sign and be liable, enter the boardroom, obtain first-hand facts and drive change fall last. Consulting has no licence; it rests on the private-law structure of reputation + contract + signature — and AI has none of the three.

DSTANCE MAP · 立场地图Stance map

你站在哪:按角色 × 替代节奏三层过滤Where you stand: filter by role × the three tempos of replacement

选「我是谁」高亮相关卡;切三层——人天层(判断制作,已被吃) / 混合层(争夺中) / 信用层(署名与背锅,人的堡垒)。切换动作本身就是「劳动密集层缩水、信用密集层升值」论点的交互证明。Pick «who I am» to highlight cards; toggle the three tiers — grunt (the manufacture, eaten) / hybrid (contested) / credit (the signature and the fall, the human fortress). The toggling is the interactive proof of «labour-intensive shrinks, credit-intensive appreciates».

我是谁WHO I AM 初级顾问 · 想入行Junior · aspirant 合伙人 · 机构经营Partner · firm 企业买方 · 甲方战略部Buyer · client strategy 信息中介 · 平台Broker · platform 个人军师 · 知识付费Indie advisor 智库 · 政策研究Think tank · policy

哪一层WHICH TIER 人天层 · 已被吃Grunt · eaten 混合层 · 争夺中Hybrid · contested 信用层 · 人的堡垒Credit · fortress

DEBATES · 争议对垒Live debates

五场没打完的仗Five unfinished fights

四份底稿对下面五题立场不一——各方并置,再给编者合议,不裁掉任何一方。The four sources disagree on the five questions below — each side shown, then an editorial synthesis; no side is cut.

TIMELINE · 时间线Timeline

2019–2026:军备 → 冲击 → 反噬三幕2019–2026: arms race → shock → blowback

底稿对 2015–2020 极薄(金字塔 + 按人天计费的「前 AI 稳态」为默认背景,标 ※),2023 起爆发(军备元年),2025–2026 最厚(裁员/股价/幻觉事故/模型厂商下场)——薄区如实留稀。Sources are thin on 2015–2020 (the pyramid + per-diem «pre-AI steady state» as default background, ※) and erupt from 2023 (the arms-race year), thickest in 2025–2026 (layoffs / stock / hallucination incidents / model vendors entering) — sparse zones left honestly sparse.

FIELD GUIDE · 产品指南Field guide

四种人的「现在能用 / 别碰 / 戒律」«Use now / avoid / discipline» for four roles

厂商效率数字一律 ⚑D 级;所有清单是行业结构信息,不构成任何建议。Vendor efficiency figures are grade D ⚑; these lists are industry-structure information, not any form of advice.

FIVE LAWS · 五条规律Five laws

把整张图压缩成五句判断The whole map, compressed into five judgments

先吃的是判断的制作,不是判断本身What’s eaten first is the manufacture, not the judgment

AI 先软件化的是养活金字塔底座的流水线——桌面研究、纪要、建模、做 deck;而判断的采纳(背书、政治、合法性、担责)长在人与制度上,AI 卡在接口。评估一个咨询岗位危不危险,不看它会不会做 deck,看它签不签得了名、进不进得了董事会。AI first softwareises the line that feeds the pyramid base — desk research, notes, modelling, decks; the adoption of judgment (endorsement, politics, legitimacy, ownership) lives in people and institutions, where AI jams. To judge a consulting role’s risk, don’t ask whether it can build a deck; ask whether it can sign its name and enter the boardroom.

谁署名,谁背锅:AI 没有可被摧毁的抵押品Who signs takes the fall: AI has no destroyable collateral

客户买咨询,买的从来不只是分析,还有「万一失败时的背锅权」。AI 能生成建议却无法承担风险,因此无法积累信任——「没有所有权的建议只是表演」。咨询无执照,全靠声誉 + 合同 + 署名的私法结构兜底;两起幻觉事故(政府合同退款、45 条引注仅 5 条准确)证明:报告可以民主化,署名不能外包。Clients buy consulting not only for analysis but for «the right to take the fall if it fails». AI can generate advice but cannot bear risk, so it cannot accumulate trust — «advice without ownership is only a performance». Consulting has no licence; it rests on reputation + contract + signature; two hallucination incidents (a refunded government contract, only 5 of 45 accurate citations) prove: a report can be democratised, a signature cannot be outsourced.

纯信息中介最先被资本市场重定价Pure information intermediaries are repriced first

第一批被打穿的不是咨询公司,是「卖 PDF 信息差」的分析师机构:Gartner 市值自高点回撤逾 52%、遭证券集体诉讼,报道归因「客户用模型厂商工具自建、绕过付费洞察」;Forrester 连亏裁员退出战略咨询。悖论:Gartner 自己还是 AI 支出最大的鼓吹者。信息不对称是最脆的护城河,专有数据才是硬的。The first pierced were not consultancies but the analyst houses selling «PDF information asymmetry»: Gartner down over 52% from peak with a securities class action, reports attributing it to «clients building their own with model-vendor tools, bypassing paid insights»; Forrester loss-making, cutting staff, exiting strategy consulting. The paradox: Gartner is itself the loudest booster of AI spending. Information asymmetry is the most brittle moat; proprietary data is the hard one.

同一场军备,有人造护城河,有人拆自己的桥One arms race: some build a moat, some dismantle their own bridge

巨头把 AI 化能力卖给客户,同时拆自己的金字塔(某巨头自称 2.5 万 AI agent、AI 占营收 25%);模型厂商反向下场,联合金融资本 15 亿美元自建企业 AI 服务公司。分野在「AI 能不能吃到别人没有的数据」:有专有知识库/专家转录/行业基准者,AI 化是军备;只靠信息不对称与通用方法论者,AI 化是自我拆解加速器。The giants sell AI-ification to clients while dismantling their own pyramid (one claims 25,000 AI agents, AI at 25% of revenue); model vendors enter in reverse, building a $1.5B enterprise-AI services firm with financial capital. The divide is «whether AI can feed on data others lack»: those with proprietary knowledge bases / expert transcripts / industry benchmarks make it an arms race; those relying on information asymmetry and general methods make it a self-dismantling accelerant.

塌的是底座,断的是梯子The base collapses; the ladder breaks

入门招聘 −54%、起薪三连冻、砍 45% MBA 招聘——底座缩水是硬数据;但顶层杠杆是重构而非消失(金字塔变沙漏/方尖碑)。真正的次生灾害是新人断层:如果新人不再做 100 个市场 sizing、50 个访谈纪要、30 个 deck,判断力的学徒制断了,未来的合伙人从哪长出来?四源都没有答案。塌的是底座,断的是培养未来判断者的梯子。Entry hiring −54%, a third year of frozen pay, 45% fewer MBA hires — the shrinking base is hard data; but the top’s leverage is reconstructed, not gone (the pyramid becomes an hourglass/obelisk). The real second-order harm is the missing rung: if juniors no longer do 100 market-sizings, 50 interview notes, 30 decks, the apprenticeship of judgment breaks — where do future partners grow from? None of the four sources has an answer. The base collapses; the ladder that grows future judges breaks.