[{"data":1,"prerenderedAt":508},["ShallowReactive",2],{"blog-ab-testing-ad-campaigns-zh-CN":3,"blog-ad-banners-zh-CN":480},{"id":4,"title":5,"excerpt":6,"content":7,"coverImage":421,"meta":430,"site":434,"status":455,"slug":456,"author":457,"category":465,"publishDate":474,"featured":45,"updatedAt":475,"createdAt":476,"contentHtml":477,"previewUrl":478,"localeSlugs":479},780,"广告系列 A/B 测试：隔离真正拉动 CPA 的变量（2026）","广告系列 A/B 测试怎么做才有效：一次只改一个变量、公平分流、转化样本下限，再加预写决策规则——放大的是 CPA/ROAS，不是虚荣 CTR。",{"root":8},{"children":9,"direction":19,"format":15,"indent":13,"type":420,"version":18},[10,22,33,50,54,58,124,128,132,175,179,183,209,213,305,313,317,349,353,375,379,383,387,392,396,400,404,408,412,416],{"children":11,"direction":19,"format":15,"indent":13,"type":20,"version":18,"tag":21},[12],{"detail":13,"format":13,"mode":14,"style":15,"text":16,"type":17,"version":18},0,"normal","","广告系列 A/B 测试到底在测什么","text",1,null,"heading","h2",{"children":23,"direction":19,"format":15,"indent":13,"type":32,"version":18,"textFormat":18,"textStyle":15},[24,26,28,30],{"detail":13,"format":18,"mode":14,"style":15,"text":25,"type":17,"version":18},"广告系列 A/B 测试（A/B testing ad campaigns）",{"detail":13,"format":13,"mode":14,"style":15,"text":27,"type":17,"version":18}," 是在可控条件下对比两套（或多套）投放结构，好把 CPA、ROAS 或量级变化归因到",{"detail":13,"format":18,"mode":14,"style":15,"text":29,"type":17,"version":18},"一个刻意改动的变量",{"detail":13,"format":13,"mode":14,"style":15,"text":31,"type":17,"version":18},"，而不是拍卖波动、季节性，或某条素材碰巧火了一周。它不是「开五条广告、留 CTR 最高的那条」。","paragraph",{"children":34,"direction":19,"format":15,"indent":13,"type":32,"version":18,"textFormat":13,"textStyle":15},[35,37,48],{"detail":13,"format":13,"mode":14,"style":15,"text":36,"type":17,"version":18},"把本文当操作手册：测什么、Meta/Google/TikTok 怎么搭才不毁掉学习期、跑多久、以及何时不该用拆分测试（那时该上 ",{"children":38,"direction":19,"format":15,"indent":13,"type":41,"version":42,"fields":43,"id":47},[39],{"detail":13,"format":13,"mode":14,"style":15,"text":40,"type":17,"version":18},"广告增量测试","link",3,{"linkType":44,"newTab":45,"url":46},"custom",false,"https://deepclick.com/zh-CN/resources/blog/incrementality-testing-for-ads/","6aec016a394d0500c85a1653",{"detail":13,"format":13,"mode":14,"style":15,"text":49,"type":17,"version":18},"）。",{"children":51,"direction":19,"format":15,"indent":13,"type":20,"version":18,"tag":21},[52],{"detail":13,"format":13,"mode":14,"style":15,"text":53,"type":17,"version":18},"一次只测一个变量",{"children":55,"direction":19,"format":15,"indent":13,"type":32,"version":18,"textFormat":13,"textStyle":15},[56],{"detail":13,"format":13,"mode":14,"style":15,"text":57,"type":17,"version":18},"有用的广告系列 A/B，主杠杆只有一个：",{"children":59,"direction":19,"format":15,"indent":13,"type":121,"version":18,"listType":122,"start":18,"tag":123},[60,67,74,80,96],{"children":61,"direction":19,"format":15,"indent":13,"type":66,"version":18,"value":18},[62,64],{"detail":13,"format":18,"mode":14,"style":15,"text":63,"type":17,"version":18},"素材",{"detail":13,"format":13,"mode":14,"style":15,"text":65,"type":17,"version":18},"：钩子、版式、时长或卖点表达——定向与出价不变。","listitem",{"children":68,"direction":19,"format":15,"indent":13,"type":66,"version":18,"value":73},[69,71],{"detail":13,"format":18,"mode":14,"style":15,"text":70,"type":17,"version":18},"受众 / 信号",{"detail":13,"format":13,"mode":14,"style":15,"text":72,"type":17,"version":18},"：相似受众种子、兴趣组合或一方人群——素材与预算规则不变。",2,{"children":75,"direction":19,"format":15,"indent":13,"type":66,"version":18,"value":42},[76,78],{"detail":13,"format":18,"mode":14,"style":15,"text":77,"type":17,"version":18},"出价 / 优化目标",{"detail":13,"format":13,"mode":14,"style":15,"text":79,"type":17,"version":18},"：成本上限 vs 尽量多转化，或优化购买 vs 加购。",{"children":81,"direction":19,"format":15,"indent":13,"type":66,"version":18,"value":95},[82,84,86,93],{"detail":13,"format":18,"mode":14,"style":15,"text":83,"type":17,"version":18},"结构",{"detail":13,"format":13,"mode":14,"style":15,"text":85,"type":17,"version":18},"：单系列 vs 多广告组——先读 ",{"children":87,"direction":19,"format":15,"indent":13,"type":41,"version":42,"fields":90,"id":92},[88],{"detail":13,"format":13,"mode":14,"style":15,"text":89,"type":17,"version":18},"Facebook 广告账户结构",{"linkType":44,"newTab":45,"url":91},"https://deepclick.com/zh-CN/resources/blog/facebook-ads-account-structure/","6aec016a394d0500c85a1654",{"detail":13,"format":13,"mode":14,"style":15,"text":94,"type":17,"version":18},"，别把「重组」伪装成测试却同时改了学习池。",4,{"children":97,"direction":19,"format":15,"indent":13,"type":66,"version":18,"value":120},[98,100,102,109,111,118],{"detail":13,"format":18,"mode":14,"style":15,"text":99,"type":17,"version":18},"落地 / 深链路径",{"detail":13,"format":13,"mode":14,"style":15,"text":101,"type":17,"version":18},"：仅当广告与拍卖侧完全一致；否则你在混测媒体和页面。配合 ",{"children":103,"direction":19,"format":15,"indent":13,"type":41,"version":42,"fields":106,"id":108},[104],{"detail":13,"format":13,"mode":14,"style":15,"text":105,"type":17,"version":18},"移动深链",{"linkType":44,"newTab":45,"url":107},"https://deepclick.com/zh-CN/resources/blog/mobile-deep-linking-guide/","6aec016a394d0500c85a1655",{"detail":13,"format":13,"mode":14,"style":15,"text":110,"type":17,"version":18}," 与 ",{"children":112,"direction":19,"format":15,"indent":13,"type":41,"version":42,"fields":115,"id":117},[113],{"detail":13,"format":13,"mode":14,"style":15,"text":114,"type":17,"version":18},"广告活动链接追踪",{"linkType":44,"newTab":45,"url":116},"https://deepclick.com/zh-CN/resources/blog/link-tracking-for-ad-campaigns-2026/","6aec016a394d0500c85a1656",{"detail":13,"format":13,"mode":14,"style":15,"text":119,"type":17,"version":18},"，保证 UTM 与去向清晰。",5,"list","bullet","ul",{"children":125,"direction":19,"format":15,"indent":13,"type":32,"version":18,"textFormat":13,"textStyle":15},[126],{"detail":13,"format":13,"mode":14,"style":15,"text":127,"type":17,"version":18},"若一次改素材、受众又改出价，结果无法规模化复用。",{"children":129,"direction":19,"format":15,"indent":13,"type":20,"version":18,"tag":21},[130],{"detail":13,"format":13,"mode":14,"style":15,"text":131,"type":17,"version":18},"怎么搭测试（不污染学习）",{"children":133,"direction":19,"format":15,"indent":13,"type":121,"version":18,"listType":173,"start":18,"tag":174},[134,140,146,161,167],{"children":135,"direction":19,"format":15,"indent":13,"type":66,"version":18,"value":18},[136,138],{"detail":13,"format":18,"mode":14,"style":15,"text":137,"type":17,"version":18},"先写决策规则",{"detail":13,"format":13,"mode":14,"style":15,"text":139,"type":17,"version":18},"：例如「胜出方 CPA 优于对照 ≥15%，且达到最低转化数」。没有规则，你最后会选好看的 CTR。",{"children":141,"direction":19,"format":15,"indent":13,"type":66,"version":18,"value":73},[142,144],{"detail":13,"format":18,"mode":14,"style":15,"text":143,"type":17,"version":18},"公平分流",{"detail":13,"format":13,"mode":14,"style":15,"text":145,"type":17,"version":18},"：尽量同地区、同时段、同版位库存。能用平台原生实验（Meta Experiments、Google Experiments）就不要手搓「两个等预算系列」。",{"children":147,"direction":19,"format":15,"indent":13,"type":66,"version":18,"value":42},[148,150,152,159],{"detail":13,"format":18,"mode":14,"style":15,"text":149,"type":17,"version":18},"对齐花费与时间窗",{"detail":13,"format":13,"mode":14,"style":15,"text":151,"type":17,"version":18},"：别拿半天尖峰下结论。归因窗口要和你看的 ",{"children":153,"direction":19,"format":15,"indent":13,"type":41,"version":42,"fields":156,"id":158},[154],{"detail":13,"format":13,"mode":14,"style":15,"text":155,"type":17,"version":18},"效果广告 KPI",{"linkType":44,"newTab":45,"url":157},"https://deepclick.com/zh-CN/resources/blog/performance-marketing-kpis/","6aec016a394d0500c85a1657",{"detail":13,"format":13,"mode":14,"style":15,"text":160,"type":17,"version":18}," 一致。",{"children":162,"direction":19,"format":15,"indent":13,"type":66,"version":18,"value":95},[163,165],{"detail":13,"format":18,"mode":14,"style":15,"text":164,"type":17,"version":18},"追踪完全一致",{"detail":13,"format":13,"mode":14,"style":15,"text":166,"type":17,"version":18},"：同一像素/CAPI 事件、同一排除窗口、同一 UTM 规范。追踪坏了会伪装成「失败变体」。",{"children":168,"direction":19,"format":15,"indent":13,"type":66,"version":18,"value":120},[169,171],{"detail":13,"format":18,"mode":14,"style":15,"text":170,"type":17,"version":18},"中途不改",{"detail":13,"format":13,"mode":14,"style":15,"text":172,"type":17,"version":18},"：改名、换文案、调上限都会毁掉可比性。","number","ol",{"children":176,"direction":19,"format":15,"indent":13,"type":32,"version":18,"textFormat":13,"textStyle":15},[177],{"detail":13,"format":13,"mode":14,"style":15,"text":178,"type":17,"version":18},"页面文案与版式实验请与系列 A/B 分开，别把落地页改动塞进媒体实验。",{"children":180,"direction":19,"format":15,"indent":13,"type":20,"version":18,"tag":21},[181],{"detail":13,"format":13,"mode":14,"style":15,"text":182,"type":17,"version":18},"样本量与时长（实务下限）",{"children":184,"direction":19,"format":15,"indent":13,"type":121,"version":18,"listType":122,"start":18,"tag":123},[185,191,197,203],{"children":186,"direction":19,"format":15,"indent":13,"type":66,"version":18,"value":18},[187,189],{"detail":13,"format":18,"mode":14,"style":15,"text":188,"type":17,"version":18},"看转化，不看点击",{"detail":13,"format":13,"mode":14,"style":15,"text":190,"type":17,"version":18},"：样本建在你要放大的 KPI（购买、合格线索）上，而不是 CTR。",{"children":192,"direction":19,"format":15,"indent":13,"type":66,"version":18,"value":73},[193,195],{"detail":13,"format":18,"mode":14,"style":15,"text":194,"type":17,"version":18},"每臂下限",{"detail":13,"format":13,"mode":14,"style":15,"text":196,"type":17,"version":18},"：方向性结论通常要每臂数十转化；押注大或效应小时要上百级。",{"children":198,"direction":19,"format":15,"indent":13,"type":66,"version":18,"value":42},[199,201],{"detail":13,"format":18,"mode":14,"style":15,"text":200,"type":17,"version":18},"时长",{"detail":13,"format":13,"mode":14,"style":15,"text":202,"type":17,"version":18},"：多数 DTC/应用账户至少覆盖一个完整周；学习慢或成交周期长则更久。",{"children":204,"direction":19,"format":15,"indent":13,"type":66,"version":18,"value":95},[205,207],{"detail":13,"format":18,"mode":14,"style":15,"text":206,"type":17,"version":18},"统计功效",{"detail":13,"format":13,"mode":14,"style":15,"text":208,"type":17,"version":18},"：3% 提升比 25% 提升吃数据多得多。养不起样本就换更大胆的假设——或干脆别测。",{"children":210,"direction":19,"format":15,"indent":13,"type":20,"version":18,"tag":21},[211],{"detail":13,"format":13,"mode":14,"style":15,"text":212,"type":17,"version":18},"能过董事会的决策表",{"children":214,"direction":19,"format":15,"indent":13,"type":304,"version":18},[215,231,245,259,273],{"children":216,"direction":19,"format":15,"indent":13,"type":230,"version":18},[217,224],{"children":218,"direction":19,"format":15,"indent":13,"type":223,"version":18,"backgroundColor":19,"colSpan":18,"headerState":18,"rowSpan":18},[219],{"children":220,"direction":19,"format":15,"indent":13,"type":32,"version":18,"textFormat":13,"textStyle":15},[221],{"detail":13,"format":13,"mode":14,"style":15,"text":222,"type":17,"version":18}," 结果形态 ","tablecell",{"children":225,"direction":19,"format":15,"indent":13,"type":223,"version":18,"backgroundColor":19,"colSpan":18,"headerState":18,"rowSpan":18},[226],{"children":227,"direction":19,"format":15,"indent":13,"type":32,"version":18,"textFormat":13,"textStyle":15},[228],{"detail":13,"format":13,"mode":14,"style":15,"text":229,"type":17,"version":18}," 动作 ","tablerow",{"children":232,"direction":19,"format":15,"indent":13,"type":230,"version":18},[233,239],{"children":234,"direction":19,"format":15,"indent":13,"type":223,"version":18,"backgroundColor":19,"colSpan":18,"headerState":13,"rowSpan":18},[235],{"children":236,"direction":19,"format":15,"indent":13,"type":32,"version":18,"textFormat":13,"textStyle":15},[237],{"detail":13,"format":13,"mode":14,"style":15,"text":238,"type":17,"version":18}," 主 KPI 明显胜出 + 质量稳定（退款、留存、线索接受率） ",{"children":240,"direction":19,"format":15,"indent":13,"type":223,"version":18,"backgroundColor":19,"colSpan":18,"headerState":13,"rowSpan":18},[241],{"children":242,"direction":19,"format":15,"indent":13,"type":32,"version":18,"textFormat":13,"textStyle":15},[243],{"detail":13,"format":13,"mode":14,"style":15,"text":244,"type":17,"version":18}," 渐进放量；大预算可留对照 ",{"children":246,"direction":19,"format":15,"indent":13,"type":230,"version":18},[247,253],{"children":248,"direction":19,"format":15,"indent":13,"type":223,"version":18,"backgroundColor":19,"colSpan":18,"headerState":13,"rowSpan":18},[249],{"children":250,"direction":19,"format":15,"indent":13,"type":32,"version":18,"textFormat":13,"textStyle":15},[251],{"detail":13,"format":13,"mode":14,"style":15,"text":252,"type":17,"version":18}," 只赢 CTR/CPC，CPA/ROAS 持平或更差 ",{"children":254,"direction":19,"format":15,"indent":13,"type":223,"version":18,"backgroundColor":19,"colSpan":18,"headerState":13,"rowSpan":18},[255],{"children":256,"direction":19,"format":15,"indent":13,"type":32,"version":18,"textFormat":13,"textStyle":15},[257],{"detail":13,"format":13,"mode":14,"style":15,"text":258,"type":17,"version":18}," 丢掉或重写——拍卖喜欢，生意不喜欢 ",{"children":260,"direction":19,"format":15,"indent":13,"type":230,"version":18},[261,267],{"children":262,"direction":19,"format":15,"indent":13,"type":223,"version":18,"backgroundColor":19,"colSpan":18,"headerState":13,"rowSpan":18},[263],{"children":264,"direction":19,"format":15,"indent":13,"type":32,"version":18,"textFormat":13,"textStyle":15},[265],{"detail":13,"format":13,"mode":14,"style":15,"text":266,"type":17,"version":18}," 落在决策边界内的平局 ",{"children":268,"direction":19,"format":15,"indent":13,"type":223,"version":18,"backgroundColor":19,"colSpan":18,"headerState":13,"rowSpan":18},[269],{"children":270,"direction":19,"format":15,"indent":13,"type":32,"version":18,"textFormat":13,"textStyle":15},[271],{"detail":13,"format":13,"mode":14,"style":15,"text":272,"type":17,"version":18}," 保留对照，下一轮换更强假设 ",{"children":274,"direction":19,"format":15,"indent":13,"type":230,"version":18},[275,281],{"children":276,"direction":19,"format":15,"indent":13,"type":223,"version":18,"backgroundColor":19,"colSpan":18,"headerState":13,"rowSpan":18},[277],{"children":278,"direction":19,"format":15,"indent":13,"type":32,"version":18,"textFormat":13,"textStyle":15},[279],{"detail":13,"format":13,"mode":14,"style":15,"text":280,"type":17,"version":18}," 赢了但合规/拒审风险更高 ",{"children":282,"direction":19,"format":15,"indent":13,"type":223,"version":18,"backgroundColor":19,"colSpan":18,"headerState":13,"rowSpan":18},[283],{"children":284,"direction":19,"format":15,"indent":13,"type":32,"version":18,"textFormat":13,"textStyle":15},[285,287,294,295,302],{"detail":13,"format":13,"mode":14,"style":15,"text":286,"type":17,"version":18}," 不放大——见 ",{"children":288,"direction":19,"format":15,"indent":13,"type":41,"version":42,"fields":291,"id":293},[289],{"detail":13,"format":13,"mode":14,"style":15,"text":290,"type":17,"version":18},"广告合规与流量路由",{"linkType":44,"newTab":45,"url":292},"https://deepclick.com/zh-CN/resources/blog/ad-compliance-traffic-routing-overseas/","6aec016a394d0500c85a1658",{"detail":13,"format":13,"mode":14,"style":15,"text":110,"type":17,"version":18},{"children":296,"direction":19,"format":15,"indent":13,"type":41,"version":42,"fields":299,"id":301},[297],{"detail":13,"format":13,"mode":14,"style":15,"text":298,"type":17,"version":18},"Facebook 拒审修复",{"linkType":44,"newTab":45,"url":300},"https://deepclick.com/zh-CN/resources/blog/facebook-ad-rejected-how-to-fix/","6aec016a394d0500c85a1659",{"detail":13,"format":13,"mode":14,"style":15,"text":303,"type":17,"version":18}," ","table",{"children":306,"direction":19,"format":15,"indent":13,"type":32,"version":18,"textFormat":13,"textStyle":15},[307,309,311],{"detail":13,"format":13,"mode":14,"style":15,"text":308,"type":17,"version":18},"还要看",{"detail":13,"format":18,"mode":14,"style":15,"text":310,"type":17,"version":18},"边际",{"detail":13,"format":13,"mode":14,"style":15,"text":312,"type":17,"version":18},"回报：日耗 $200 时的赢家，在 5 倍预算打进更冷库存时可能翻车。",{"children":314,"direction":19,"format":15,"indent":13,"type":20,"version":18,"tag":21},[315],{"detail":13,"format":13,"mode":14,"style":15,"text":316,"type":17,"version":18},"广告系列 A/B 的常见翻车",{"children":318,"direction":19,"format":15,"indent":13,"type":121,"version":18,"listType":122,"start":18,"tag":123},[319,325,331,337,343],{"children":320,"direction":19,"format":15,"indent":13,"type":66,"version":18,"value":18},[321,323],{"detail":13,"format":18,"mode":14,"style":15,"text":322,"type":17,"version":18},"天天偷看",{"detail":13,"format":13,"mode":14,"style":15,"text":324,"type":17,"version":18},"，一领先就提前停。",{"children":326,"direction":19,"format":15,"indent":13,"type":66,"version":18,"value":73},[327,329],{"detail":13,"format":18,"mode":14,"style":15,"text":328,"type":17,"version":18},"学习不对称",{"detail":13,"format":13,"mode":14,"style":15,"text":330,"type":17,"version":18},"：一臂继承历史事件密度，另一臂冷启动。",{"children":332,"direction":19,"format":15,"indent":13,"type":66,"version":18,"value":42},[333,335],{"detail":13,"format":18,"mode":14,"style":15,"text":334,"type":17,"version":18},"污染",{"detail":13,"format":13,"mode":14,"style":15,"text":336,"type":17,"version":18},"：再营销池、品牌搜索抬升或受众重叠串臂。",{"children":338,"direction":19,"format":15,"indent":13,"type":66,"version":18,"value":95},[339,341],{"detail":13,"format":18,"mode":14,"style":15,"text":340,"type":17,"version":18},"指标购物",{"detail":13,"format":13,"mode":14,"style":15,"text":342,"type":17,"version":18},"：看完结果再改「主 KPI」。",{"children":344,"direction":19,"format":15,"indent":13,"type":66,"version":18,"value":120},[345,347],{"detail":13,"format":18,"mode":14,"style":15,"text":346,"type":17,"version":18},"把 A/B 当成增量",{"detail":13,"format":13,"mode":14,"style":15,"text":348,"type":17,"version":18},"：平台 ROAS 赢了仍可能无净增——大放大前要上 holdout。",{"children":350,"direction":19,"format":15,"indent":13,"type":20,"version":18,"tag":21},[351],{"detail":13,"format":13,"mode":14,"style":15,"text":352,"type":17,"version":18},"一周操作闭环",{"children":354,"direction":19,"format":15,"indent":13,"type":121,"version":18,"listType":173,"start":18,"tag":174},[355,359,363,367,371],{"children":356,"direction":19,"format":15,"indent":13,"type":66,"version":18,"value":18},[357],{"detail":13,"format":13,"mode":14,"style":15,"text":358,"type":17,"version":18},"周一：选一个绑金钱 KPI 的假设。",{"children":360,"direction":19,"format":15,"indent":13,"type":66,"version":18,"value":73},[361],{"detail":13,"format":13,"mode":14,"style":15,"text":362,"type":17,"version":18},"搭两臂，只在该假设上不同；其余冻结。",{"children":364,"direction":19,"format":15,"indent":13,"type":66,"version":18,"value":42},[365],{"detail":13,"format":13,"mode":14,"style":15,"text":366,"type":17,"version":18},"跑到既定样本/时长——周中不「微调」素材。",{"children":368,"direction":19,"format":15,"indent":13,"type":66,"version":18,"value":95},[369],{"detail":13,"format":13,"mode":14,"style":15,"text":370,"type":17,"version":18},"周五：按预写规则打分；记录赢家、输家与下一假设。",{"children":372,"direction":19,"format":15,"indent":13,"type":66,"version":18,"value":120},[373],{"detail":13,"format":13,"mode":14,"style":15,"text":374,"type":17,"version":18},"归档预算、ID、UTM，保证下一轮可对比。",{"children":376,"direction":19,"format":15,"indent":13,"type":20,"version":18,"tag":21},[377],{"detail":13,"format":13,"mode":14,"style":15,"text":378,"type":17,"version":18},"结论",{"children":380,"direction":19,"format":15,"indent":13,"type":32,"version":18,"textFormat":13,"textStyle":15},[381],{"detail":13,"format":13,"mode":14,"style":15,"text":382,"type":17,"version":18},"广告系列 A/B 测试要成立，关键是：隔离单一杠杆、预先承诺决策规则、用生意指标而不是虚荣互动来判。媒体杠杆优先用平台拆分实验，落地页实验分开做；当问题从「哪条广告赢」变成「这钱有没有净增结果」时，升级到增量设计。",{"children":384,"direction":19,"format":15,"indent":13,"type":20,"version":18,"tag":21},[385],{"detail":13,"format":13,"mode":14,"style":15,"text":386,"type":17,"version":18},"FAQ",{"children":388,"direction":19,"format":15,"indent":13,"type":20,"version":18,"tag":391},[389],{"detail":13,"format":13,"mode":14,"style":15,"text":390,"type":17,"version":18},"广告系列 A/B 一次能改几个变量？","h3",{"children":393,"direction":19,"format":15,"indent":13,"type":32,"version":18,"textFormat":13,"textStyle":15},[394],{"detail":13,"format":13,"mode":14,"style":15,"text":395,"type":17,"version":18},"只改一个主变量。预算四舍五入、投放时刻等次要差异尽量压到最小，结果才可执行。",{"children":397,"direction":19,"format":15,"indent":13,"type":20,"version":18,"tag":391},[398],{"detail":13,"format":13,"mode":14,"style":15,"text":399,"type":17,"version":18},"CTR 能当广告系列 A/B 的主指标吗？",{"children":401,"direction":19,"format":15,"indent":13,"type":32,"version":18,"textFormat":13,"textStyle":15},[402],{"detail":13,"format":13,"mode":14,"style":15,"text":403,"type":17,"version":18},"很少能。CTR 是诊断项。主指标应是 CPA、ROAS、扣广告后贡献利润或合格商机——财务认的那个。",{"children":405,"direction":19,"format":15,"indent":13,"type":20,"version":18,"tag":391},[406],{"detail":13,"format":13,"mode":14,"style":15,"text":407,"type":17,"version":18},"广告 A/B 要跑多久？",{"children":409,"direction":19,"format":15,"indent":13,"type":32,"version":18,"textFormat":13,"textStyle":15},[410],{"detail":13,"format":13,"mode":14,"style":15,"text":411,"type":17,"version":18},"跑到预设转化下限，并且至少覆盖一个周需求周期；除非平台实验工具在你接受的功效设置下更早宣布胜者。",{"children":413,"direction":19,"format":15,"indent":13,"type":20,"version":18,"tag":391},[414],{"detail":13,"format":13,"mode":14,"style":15,"text":415,"type":17,"version":18},"什么时候该用增量测试而不是 A/B？",{"children":417,"direction":19,"format":15,"indent":13,"type":32,"version":18,"textFormat":13,"textStyle":15},[418],{"detail":13,"format":13,"mode":14,"style":15,"text":419,"type":17,"version":18},"当你需要相对 holdout（地理、用户或 PSA 式）的因果增量，尤其是平台归因噪声大、准备大幅放量之前。","root",{"id":422,"alt":423,"updatedAt":424,"createdAt":424,"url":425,"thumbnailURL":19,"filename":426,"mimeType":427,"filesize":428,"width":429,"height":429},1293,"muvloom inspiration r26b-46","2026-10-20T10:06:33.957Z","https://cms-r2.deepclick.com/r26b-46-25cb62d727af.jpg","r26b-46-25cb62d727af.jpg","image/jpeg",460301,1254,{"title":431,"description":432,"image":433},"广告系列 A/B 测试：拉动 CPA 的变量（2026）","用单变量隔离、平台原生实验、转化样本下限与预写决策规则做广告系列 A/B 测试，避免被 CTR 带跑、放大错误赢家。",{"id":422,"alt":423,"updatedAt":424,"createdAt":424,"url":425,"thumbnailURL":19,"filename":426,"mimeType":427,"filesize":428,"width":429,"height":429},{"id":18,"key":435,"name":436,"prodHost":437,"testHost":438,"blogPath":439,"docPath":440,"zhPrefix":441,"deployHookTest":442,"deployHookProd":443,"brandAuthor":444,"brandKit":448,"enabled":452,"updatedAt":453,"createdAt":454},"deepclick","DeepClick","https://deepclick.com","https://www-test-deepclick.qiliangjia.one","/resources/blog/{slug}","/docs/{slug}","/zh-CN","https://api.cloudflare.com/client/v4/pages/webhooks/deploy_hooks/60a9adef-153b-4c89-8d07-7118e91e9522","https://api.cloudflare.com/client/v4/pages/webhooks/deploy_hooks/05323321-f694-4ce8-a5af-173c507b8bae",{"id":73,"name":436,"avatar":445,"updatedAt":446,"createdAt":447},25,"2026-04-22T08:09:35.299Z","2026-04-22T06:42:49.116Z",{"logoLight":19,"logoDark":19,"primaryColor":449,"accentColor":450,"tagline":451,"wechatName":19,"xhsHandle":19,"wechatQr":19},"#1a73e8","#ff6600","一次点击，多重价值",true,"2026-08-25T06:40:57.663Z","2026-07-14T06:26:38.962Z","published","ab-testing-ad-campaigns",{"id":73,"name":436,"avatar":458,"updatedAt":446,"createdAt":447},{"id":445,"alt":436,"updatedAt":459,"createdAt":459,"url":460,"thumbnailURL":19,"filename":461,"mimeType":462,"filesize":463,"width":464,"height":464},"2026-04-22T08:09:22.606Z","https://cms-r2.deepclick.com/头像-白.png","头像-白.png","image/png",26626,1024,{"id":466,"site":467,"titleZh":469,"titleEn":470,"slug":471,"order":120,"updatedAt":472,"createdAt":473},7,{"id":18,"key":435,"name":436,"prodHost":437,"testHost":438,"blogPath":439,"docPath":440,"zhPrefix":441,"deployHookTest":442,"deployHookProd":443,"brandAuthor":73,"brandKit":468,"enabled":452,"updatedAt":453,"createdAt":454},{"logoLight":19,"logoDark":19,"primaryColor":449,"accentColor":450,"tagline":451,"wechatName":19,"xhsHandle":19,"wechatQr":19},"技术导航","Tech Guides","tech-guides","2026-04-27T08:37:10.576Z","2026-04-23T02:59:13.436Z","2026-11-05T04:09:22.082Z","2026-11-05T04:09:23.492Z","2026-11-05T04:09:04.904Z","\u003Cdiv class=\"payload-richtext\">\u003Ch2>广告系列 A/B 测试到底在测什么\u003C/h2>\u003Cp>\u003Cstrong>广告系列 A/B 测试（A/B testing ad campaigns）\u003C/strong> 是在可控条件下对比两套（或多套）投放结构，好把 CPA、ROAS 或量级变化归因到\u003Cstrong>一个刻意改动的变量\u003C/strong>，而不是拍卖波动、季节性，或某条素材碰巧火了一周。它不是「开五条广告、留 CTR 最高的那条」。\u003C/p>\u003Cp>把本文当操作手册：测什么、Meta/Google/TikTok 怎么搭才不毁掉学习期、跑多久、以及何时不该用拆分测试（那时该上 \u003Ca href=\"https://deepclick.com/zh-CN/resources/blog/incrementality-testing-for-ads/\">广告增量测试\u003C/a>）。\u003C/p>\u003Ch2>一次只测一个变量\u003C/h2>\u003Cp>有用的广告系列 A/B，主杠杆只有一个：\u003C/p>\u003Cul class=\"list-bullet\">\u003Cli\n          class=\"\"\n          style=\"\"\n          value=\"1\"\n        >\u003Cstrong>素材\u003C/strong>：钩子、版式、时长或卖点表达——定向与出价不变。\u003C/li>\u003Cli\n          class=\"\"\n          style=\"\"\n          value=\"2\"\n        >\u003Cstrong>受众 / 信号\u003C/strong>：相似受众种子、兴趣组合或一方人群——素材与预算规则不变。\u003C/li>\u003Cli\n          class=\"\"\n          style=\"\"\n          value=\"3\"\n        >\u003Cstrong>出价 / 优化目标\u003C/strong>：成本上限 vs 尽量多转化，或优化购买 vs 加购。\u003C/li>\u003Cli\n          class=\"\"\n          style=\"\"\n          value=\"4\"\n        >\u003Cstrong>结构\u003C/strong>：单系列 vs 多广告组——先读 \u003Ca href=\"https://deepclick.com/zh-CN/resources/blog/facebook-ads-account-structure/\">Facebook 广告账户结构\u003C/a>，别把「重组」伪装成测试却同时改了学习池。\u003C/li>\u003Cli\n          class=\"\"\n          style=\"\"\n          value=\"5\"\n        >\u003Cstrong>落地 / 深链路径\u003C/strong>：仅当广告与拍卖侧完全一致；否则你在混测媒体和页面。配合 \u003Ca href=\"https://deepclick.com/zh-CN/resources/blog/mobile-deep-linking-guide/\">移动深链\u003C/a> 与 \u003Ca href=\"https://deepclick.com/zh-CN/resources/blog/link-tracking-for-ad-campaigns-2026/\">广告活动链接追踪\u003C/a>，保证 UTM 与去向清晰。\u003C/li>\u003C/ul>\u003Cp>若一次改素材、受众又改出价，结果无法规模化复用。\u003C/p>\u003Ch2>怎么搭测试（不污染学习）\u003C/h2>\u003Col class=\"list-number\">\u003Cli\n          class=\"\"\n          style=\"\"\n          value=\"1\"\n        >\u003Cstrong>先写决策规则\u003C/strong>：例如「胜出方 CPA 优于对照 ≥15%，且达到最低转化数」。没有规则，你最后会选好看的 CTR。\u003C/li>\u003Cli\n          class=\"\"\n          style=\"\"\n          value=\"2\"\n        >\u003Cstrong>公平分流\u003C/strong>：尽量同地区、同时段、同版位库存。能用平台原生实验（Meta Experiments、Google Experiments）就不要手搓「两个等预算系列」。\u003C/li>\u003Cli\n          class=\"\"\n          style=\"\"\n          value=\"3\"\n        >\u003Cstrong>对齐花费与时间窗\u003C/strong>：别拿半天尖峰下结论。归因窗口要和你看的 \u003Ca href=\"https://deepclick.com/zh-CN/resources/blog/performance-marketing-kpis/\">效果广告 KPI\u003C/a> 一致。\u003C/li>\u003Cli\n          class=\"\"\n          style=\"\"\n          value=\"4\"\n        >\u003Cstrong>追踪完全一致\u003C/strong>：同一像素/CAPI 事件、同一排除窗口、同一 UTM 规范。追踪坏了会伪装成「失败变体」。\u003C/li>\u003Cli\n          class=\"\"\n          style=\"\"\n          value=\"5\"\n        >\u003Cstrong>中途不改\u003C/strong>：改名、换文案、调上限都会毁掉可比性。\u003C/li>\u003C/ol>\u003Cp>页面文案与版式实验请与系列 A/B 分开，别把落地页改动塞进媒体实验。\u003C/p>\u003Ch2>样本量与时长（实务下限）\u003C/h2>\u003Cul class=\"list-bullet\">\u003Cli\n          class=\"\"\n          style=\"\"\n          value=\"1\"\n        >\u003Cstrong>看转化，不看点击\u003C/strong>：样本建在你要放大的 KPI（购买、合格线索）上，而不是 CTR。\u003C/li>\u003Cli\n          class=\"\"\n          style=\"\"\n          value=\"2\"\n        >\u003Cstrong>每臂下限\u003C/strong>：方向性结论通常要每臂数十转化；押注大或效应小时要上百级。\u003C/li>\u003Cli\n          class=\"\"\n          style=\"\"\n          value=\"3\"\n        >\u003Cstrong>时长\u003C/strong>：多数 DTC/应用账户至少覆盖一个完整周；学习慢或成交周期长则更久。\u003C/li>\u003Cli\n          class=\"\"\n          style=\"\"\n          value=\"4\"\n        >\u003Cstrong>统计功效\u003C/strong>：3% 提升比 25% 提升吃数据多得多。养不起样本就换更大胆的假设——或干脆别测。\u003C/li>\u003C/ul>\u003Ch2>能过董事会的决策表\u003C/h2>\u003Cdiv class=\"lexical-table-container\">\n        \u003Ctable class=\"lexical-table\" style=\"border-collapse: collapse;\">\n          \u003Ctbody>\u003Ctr class=\"lexical-table-row\">\n        \u003Cth\n        class=\"lexical-table-cell lexical-table-cell-header-1\"\n        \n        \n        style=\"border: 1px solid #ccc; padding: 8px;\"\n      >\n        \u003Cp> 结果形态 \u003C/p>\n      \u003C/th>\n    \u003Cth\n        class=\"lexical-table-cell lexical-table-cell-header-1\"\n        \n        \n        style=\"border: 1px solid #ccc; padding: 8px;\"\n      >\n        \u003Cp> 动作 \u003C/p>\n      \u003C/th>\n    \n      \u003C/tr>\u003Ctr class=\"lexical-table-row\">\n        \u003Ctd\n        class=\"lexical-table-cell lexical-table-cell-header-0\"\n        \n        \n        style=\"border: 1px solid #ccc; padding: 8px;\"\n      >\n        \u003Cp> 主 KPI 明显胜出 + 质量稳定（退款、留存、线索接受率） \u003C/p>\n      \u003C/td>\n    \u003Ctd\n        class=\"lexical-table-cell lexical-table-cell-header-0\"\n        \n        \n        style=\"border: 1px solid #ccc; padding: 8px;\"\n      >\n        \u003Cp> 渐进放量；大预算可留对照 \u003C/p>\n      \u003C/td>\n    \n      \u003C/tr>\u003Ctr class=\"lexical-table-row\">\n        \u003Ctd\n        class=\"lexical-table-cell lexical-table-cell-header-0\"\n        \n        \n        style=\"border: 1px solid #ccc; padding: 8px;\"\n      >\n        \u003Cp> 只赢 CTR/CPC，CPA/ROAS 持平或更差 \u003C/p>\n      \u003C/td>\n    \u003Ctd\n        class=\"lexical-table-cell lexical-table-cell-header-0\"\n        \n        \n        style=\"border: 1px solid #ccc; padding: 8px;\"\n      >\n        \u003Cp> 丢掉或重写——拍卖喜欢，生意不喜欢 \u003C/p>\n      \u003C/td>\n    \n      \u003C/tr>\u003Ctr class=\"lexical-table-row\">\n        \u003Ctd\n        class=\"lexical-table-cell lexical-table-cell-header-0\"\n        \n        \n        style=\"border: 1px solid #ccc; padding: 8px;\"\n      >\n        \u003Cp> 落在决策边界内的平局 \u003C/p>\n      \u003C/td>\n    \u003Ctd\n        class=\"lexical-table-cell lexical-table-cell-header-0\"\n        \n        \n        style=\"border: 1px solid #ccc; padding: 8px;\"\n      >\n        \u003Cp> 保留对照，下一轮换更强假设 \u003C/p>\n      \u003C/td>\n    \n      \u003C/tr>\u003Ctr class=\"lexical-table-row\">\n        \u003Ctd\n        class=\"lexical-table-cell lexical-table-cell-header-0\"\n        \n        \n        style=\"border: 1px solid #ccc; padding: 8px;\"\n      >\n        \u003Cp> 赢了但合规/拒审风险更高 \u003C/p>\n      \u003C/td>\n    \u003Ctd\n        class=\"lexical-table-cell lexical-table-cell-header-0\"\n        \n        \n        style=\"border: 1px solid #ccc; padding: 8px;\"\n      >\n        \u003Cp> 不放大——见 \u003Ca href=\"https://deepclick.com/zh-CN/resources/blog/ad-compliance-traffic-routing-overseas/\">广告合规与流量路由\u003C/a> 与 \u003Ca href=\"https://deepclick.com/zh-CN/resources/blog/facebook-ad-rejected-how-to-fix/\">Facebook 拒审修复\u003C/a> \u003C/p>\n      \u003C/td>\n    \n      \u003C/tr>\u003C/tbody>\n        \u003C/table>\n      \u003C/div>\u003Cp>还要看\u003Cstrong>边际\u003C/strong>回报：日耗 $200 时的赢家，在 5 倍预算打进更冷库存时可能翻车。\u003C/p>\u003Ch2>广告系列 A/B 的常见翻车\u003C/h2>\u003Cul class=\"list-bullet\">\u003Cli\n          class=\"\"\n          style=\"\"\n          value=\"1\"\n        >\u003Cstrong>天天偷看\u003C/strong>，一领先就提前停。\u003C/li>\u003Cli\n          class=\"\"\n          style=\"\"\n          value=\"2\"\n        >\u003Cstrong>学习不对称\u003C/strong>：一臂继承历史事件密度，另一臂冷启动。\u003C/li>\u003Cli\n          class=\"\"\n          style=\"\"\n          value=\"3\"\n        >\u003Cstrong>污染\u003C/strong>：再营销池、品牌搜索抬升或受众重叠串臂。\u003C/li>\u003Cli\n          class=\"\"\n          style=\"\"\n          value=\"4\"\n        >\u003Cstrong>指标购物\u003C/strong>：看完结果再改「主 KPI」。\u003C/li>\u003Cli\n          class=\"\"\n          style=\"\"\n          value=\"5\"\n        >\u003Cstrong>把 A/B 当成增量\u003C/strong>：平台 ROAS 赢了仍可能无净增——大放大前要上 holdout。\u003C/li>\u003C/ul>\u003Ch2>一周操作闭环\u003C/h2>\u003Col class=\"list-number\">\u003Cli\n          class=\"\"\n          style=\"\"\n          value=\"1\"\n        >周一：选一个绑金钱 KPI 的假设。\u003C/li>\u003Cli\n          class=\"\"\n          style=\"\"\n          value=\"2\"\n        >搭两臂，只在该假设上不同；其余冻结。\u003C/li>\u003Cli\n          class=\"\"\n          style=\"\"\n          value=\"3\"\n        >跑到既定样本/时长——周中不「微调」素材。\u003C/li>\u003Cli\n          class=\"\"\n          style=\"\"\n          value=\"4\"\n        >周五：按预写规则打分；记录赢家、输家与下一假设。\u003C/li>\u003Cli\n          class=\"\"\n          style=\"\"\n          value=\"5\"\n        >归档预算、ID、UTM，保证下一轮可对比。\u003C/li>\u003C/ol>\u003Ch2>结论\u003C/h2>\u003Cp>广告系列 A/B 测试要成立，关键是：隔离单一杠杆、预先承诺决策规则、用生意指标而不是虚荣互动来判。媒体杠杆优先用平台拆分实验，落地页实验分开做；当问题从「哪条广告赢」变成「这钱有没有净增结果」时，升级到增量设计。\u003C/p>\u003Ch2>FAQ\u003C/h2>\u003Ch3>广告系列 A/B 一次能改几个变量？\u003C/h3>\u003Cp>只改一个主变量。预算四舍五入、投放时刻等次要差异尽量压到最小，结果才可执行。\u003C/p>\u003Ch3>CTR 能当广告系列 A/B 的主指标吗？\u003C/h3>\u003Cp>很少能。CTR 是诊断项。主指标应是 CPA、ROAS、扣广告后贡献利润或合格商机——财务认的那个。\u003C/p>\u003Ch3>广告 A/B 要跑多久？\u003C/h3>\u003Cp>跑到预设转化下限，并且至少覆盖一个周需求周期；除非平台实验工具在你接受的功效设置下更早宣布胜者。\u003C/p>\u003Ch3>什么时候该用增量测试而不是 A/B？\u003C/h3>\u003Cp>当你需要相对 holdout（地理、用户或 PSA 式）的因果增量，尤其是平台归因噪声大、准备大幅放量之前。\u003C/p>\u003C/div>","https://deepclick.com/zh-CN/resources/blog/ab-testing-ad-campaigns",{"zh-CN":456,"en":456},[481],{"id":42,"title":482,"site":483,"image":485,"mobileImage":494,"targetUrl":503,"enabled":452,"postDateFrom":504,"postDateTo":505,"updatedAt":506,"createdAt":507},"2026.10.20 Jakarta summit (deepclick)",{"id":18,"key":435,"name":436,"prodHost":437,"testHost":438,"blogPath":439,"docPath":440,"zhPrefix":441,"deployHookTest":442,"deployHookProd":443,"brandAuthor":73,"brandKit":484,"enabled":452,"updatedAt":453,"createdAt":454},{"logoLight":19,"logoDark":19,"primaryColor":449,"accentColor":450,"tagline":451,"wechatName":19,"xhsHandle":19,"wechatQr":19},{"id":486,"alt":487,"updatedAt":488,"createdAt":488,"url":489,"thumbnailURL":19,"filename":490,"mimeType":462,"filesize":491,"width":492,"height":493},1069,"TrafficTalking Jakarta summit 2026.10.20 (zh)","2026-09-04T10:44:19.548Z","https://cms-r2.deepclick.com/traffictalking-jakarta-banner-zh-aa032503f69b.png","traffictalking-jakarta-banner-zh-aa032503f69b.png",1128551,2000,377,{"id":495,"alt":496,"updatedAt":497,"createdAt":497,"url":498,"thumbnailURL":19,"filename":499,"mimeType":462,"filesize":500,"width":501,"height":502},1071,"TrafficTalking Jakarta summit 2026.10.20 (mobile-zh)","2026-09-04T10:44:29.391Z","https://cms-r2.deepclick.com/traffictalking-jakarta-banner-mobile-zh-372fe26c1a2d.png","traffictalking-jakarta-banner-mobile-zh-372fe26c1a2d.png",649961,1200,400,"https://luma.com/g4tnhetc","2026-09-03T00:00:00.000Z","2026-10-04T00:00:00.000Z","2026-09-22T01:42:15.402Z","2026-09-04T10:45:03.031Z",1793933267632]